HVACR performance degradation monitoring and relationship builder

Abstract

The invention relates to a method for HVACamp; a monitoring system for an R system, the monitoring system comprising: a data acquisition processor operable to acquire information about HVACamp; observation of the R system, including fluid temperature measurements of the condenser and the evaporator, and also including compressor input power parameter measurements corresponding to the fluid temperature measurements; a relationship builder operable to learn a compressor input power parameter (CIPP) relationship between fluid temperature measurements of evaporator inlet temperature and condenser inlet temperature and compressor input power parameter measurements, and is further operable to learn an evaporator temperature drop (ETD) relationship between a fluid temperature measurement of an evaporator inlet temperature and a condenser inlet temperature and an evaporator temperature drop; and a temperature map comprising a plurality of cells, each cell corresponding to a temperature tuple consisting of a condenser entry temperature and an evaporator entry temperature, the temperature map is configured to receive and store, for each unit from the relationship builder, a summary statistics of measured compressor input power parameters or a measured derived evaporator temperature drop, or both, corresponding to a temperature tuple of the unit; wherein the processor executable instructions further cause the at least one processor to calculate a predicted value of a compressor input power parameter and / or a predicted value of an evaporator temperature drop using the CIPP relationship or the ETD relationship, respectively, and declare HVACamp based on the predicted values; the R system has performance degradation.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This international application claims the benefit of priority to U.S. non-provisional application No. 18 / 105,776, filed on February 3, 2023, entitled “HVAC&R Performance Degradation Monitor and Relation Builder,” which is a continuation-in-part of U.S. non-provisional application No. 17 / 463,476, filed on August 31, 2021, entitled “Continuous Learning Compressor Input Power Predictor,” which are incorporated herein by reference. This international application also claims the benefit of priority to commonly assigned U.S. non-provisional application No. 18 / 105,767, filed on February 3, 2023, entitled “Monitoring HVAC&R Performance Degradation Using Relative COP,” and commonly assigned U.S. non-provisional application No. 18 / 105,773, filed on February 3, 2023, entitled “Monitoring HVAC&R Performance Degradation Using Relative COP from Joint Power and Temperature Relations,” and is incorporated herein by reference. Technical Field

[0003] The disclosed embodiments relate generally to heating, ventilation, and air conditioning and refrigeration (HVAC&R) systems and, more particularly, to systems and methods for early detection of potential problems in such HVAC&R systems using relative coefficient of performance (COP) relationships. Background Art

[0004] HVAC&R systems, which can include residential and commercial heat pumps, air conditioning, and refrigeration systems, use a vapor compression cycle (VCC) to transfer heat between a low-temperature fluid and a high-temperature fluid. In many VCC-based systems, known as direct exchange systems, the "fluid" is the air in the conditioned space or the external ambient environment. In other VCC-based systems, including indirect exchange systems such as chillers and geothermal heat pumps, the fluid with which heat is exchanged can be a liquid, such as water or antifreeze.

[0005] VCC-based systems are generally known in the art and employ a refrigerant as a medium to facilitate heat transfer. The system is mechanically "closed" in that the refrigerant is contained within the system's mechanical boundaries, with a mechanical buffer present where heat is exchanged between the refrigerant and an external fluid. In these systems, the refrigerant circulates within the system, passing through the compressor, condenser, and evaporator. At the evaporator, the refrigerant absorbs heat from the space being cooled (in the case of an air conditioner or refrigerator), or from the external environment or other heat sources (in the case of a heat pump). At the condenser, the heat is rejected to the external environment (in the case of an air conditioner or refrigerator), or to the space being conditioned (in the case of a heat pump).

[0006] However, existing VCC-based systems lack sufficient capabilities for early detection and detection of potential problems and performance degradation. This lack of early problem detection is due in part to the inability of existing VCC-based systems to do so quickly and reliably. Typically, detecting performance degradation in VCC-based systems requires acquiring and processing large amounts of data over extended periods of time in order to provide a sufficient level of reliability. Over the years, the large amounts of data and processing required have proven prohibitively complex, making implementation impractical for most VCC-based systems.

[0007] Therefore, there is a need for a method to monitor and detect potential problems and performance degradation early in a VCC-based system in an efficient manner while also providing a sufficient level of reliability and accuracy. Summary of the Invention

[0008] Embodiments disclosed herein relate to improved systems and methods for monitoring HVAC&R systems that employ a vapor compression cycle. One embodiment described herein provides a monitoring application or agent that uses observations of evaporator and condenser inlet temperatures, evaporator discharge temperature, and compressor input power parameters to learn the operating characteristics of the HVAC&R system under new maintenance conditions. Thereafter, the agent continuously or periodically calculates a relative coefficient of performance, or relative COP, for the system under subsequently observed ambient conditions and correlates the current instantaneous efficiency of the HVAC&R system under the observed ambient conditions with the instantaneous efficiency of the system under the new maintenance conditions. The relative COP can be used to detect system degradation and quantify the energy usage and cost attributable to the degradation. Such a monitoring application or agent can also be extended to other types of systems besides HVAC&R systems.

[0009] In general, in one aspect, embodiments disclosed herein relate to a monitoring system for an HVAC&R system. The monitoring system includes, among other things, at least one processor and a memory device coupled to the at least one processor, the memory device having processor-executable instructions stored thereon, including instructions that, when executed by the at least one processor, cause the at least one processor to instantiate a data acquisition processor. The data acquisition processor is operable to acquire observations about the HVAC&R system, the observations comprising condenser fluid temperature measurements and evaporator fluid temperature measurements, and the observations also comprising compressor input power parameter measurements corresponding to the fluid temperature measurements. The processor-executable instructions further cause the at least one processor to instantiate a relationship builder operable to learn a compressor input power parameter (CIPP) relationship between the fluid temperature measurements of the evaporator inlet temperature and the condenser inlet temperature and the compressor input power parameter measurements. The relationship builder is further operable to learn an evaporator temperature drop (ETD) relationship between the fluid temperature measurements of the evaporator inlet temperature and the condenser inlet temperature and the evaporator temperature drop. The processor-executable instructions further cause the at least one processor to instantiate a temperature map comprising a plurality of cells, each cell corresponding to a temperature tuple consisting of a condenser inlet temperature and an evaporator inlet temperature, the temperature map being configured to receive and store, for each cell, from the relationship builder, summary statistics of a measured compressor input power parameter or a measured-derived evaporator temperature drop corresponding to the temperature tuple of the cell, or both. The processor-executable instructions further cause the at least one processor to calculate a predicted value of the compressor input power parameter and / or a predicted value of the evaporator temperature drop using a CIPP relationship or an ETD relationship, respectively, and to declare performance degradation of the HVAC&R system using the predicted value of the compressor input power parameter or the predicted value of the evaporator temperature drop or both.

[0010] In general, in another aspect, embodiments disclosed herein relate to a method for monitoring an HVAC&R system. The method includes, inter alia, acquiring, at a data acquisition processor, observations regarding the HVAC&R system, the observations comprising a condenser fluid temperature measurement and an evaporator fluid temperature measurement, the observations also comprising compressor input power parameter measurements corresponding to the fluid temperature measurements. The method also includes, at a relationship builder, learning a compressor input power parameter (CIPP) relationship between the fluid temperature measurements of the evaporator inlet temperature and the condenser inlet temperature and the compressor input power parameter measurements, and is further operable to learn an evaporator temperature drop (ETD) relationship between the fluid temperature measurements of the evaporator inlet temperature and the condenser inlet temperature and the evaporator temperature drop. The method also includes, at a temperature map comprising a plurality of cells, each cell corresponding to a temperature tuple consisting of a condenser inlet temperature and an evaporator inlet temperature, receiving and storing, from the relationship builder, summary statistics of the measured compressor input power parameter or the measured evaporator temperature drop, or both, corresponding to the temperature tuple of the cell. The method also includes the monitoring system calculating a predicted value of the compressor input power parameter and / or a predicted value of the evaporator temperature drop using the CIPP relationship or the ETD relationship, respectively, and the monitoring system declaring the presence of performance degradation in the HVAC&R system using the predicted value of the compressor input power parameter or the predicted value of the evaporator temperature drop, or both.

[0011] According to any one or more of the preceding embodiments, power to the HVAC&R system is shut off in response to declaring that performance degradation exists in the HVAC&R system.

[0012] According to any one or more of the aforementioned embodiments, the relationship builder learns the CIPP relationship and the ETD relationship using a machine learning based learning process, respectively.

[0013] According to any one or more of the aforementioned embodiments, the neighborhood extractor is operable to define, for a given temperature tuple and its elements, a temperature tuple range surrounding the given temperature tuple, the temperature tuple range being usable by the monitoring system to calculate the compressor input power parameter and the evaporator temperature drop.

[0014] According to any one or more of the foregoing embodiments, the neighborhood extractor defines a temperature tuple range for a given temperature tuple by: constructing a set of observed temperature tuples from the temperature tuples in the temperature map; determining whether the set of observed temperature tuples satisfies a predefined minimum number of temperature tuples; and determining whether the given temperature tuple is within the convex hull of a subset of the set of observed temperature tuples.

[0015] According to any one or more of the foregoing embodiments, the parameterized predictor operates to calculate predicted values ​​of the compressor input power parameter and the evaporator temperature drop using the set of observed temperature tuples.

[0016] According to any one or more of the foregoing embodiments, a parameterized predictor calculates predicted values ​​of the compressor input power parameter and the evaporator temperature drop using a summary value table constructed from a set of observed temperature tuples and a temperature map and using parameter coefficients derived from the summary value table.

[0017] According to any one or more of the aforementioned embodiments, the relationship builder includes a CIPP relationship builder configured to learn CIPP relationships and an ETD relationship builder configured to learn ETD relationships.

[0018] According to any one or more of the aforementioned embodiments, the temperature map includes a CIPP temperature map configured to receive and store summary statistics of measured compressor input power parameters from a CIPP relationship builder, and an ETD temperature map configured to receive and store summary statistics of evaporator temperature drops from an ETD relationship builder.

[0019] The embodiment according to any one or more of the preceding embodiments, wherein the neighborhood extractor is a joint neighborhood extractor operable to define a temperature tuple range for a given temperature tuple using both a CIPP temperature map and an ETD temperature map.

[0020] The method of any one or more of the preceding embodiments, wherein the parameterized predictor comprises a CIPP parameterized predictor operable to calculate a prediction of a compressor input power parameter and an ETD parameterized predictor operable to calculate a prediction of an evaporator temperature drop.

[0021] In general, in another aspect, embodiments disclosed herein relate to a monitoring and detection system. The monitoring and detection system includes, among other things, at least one processor and a storage device coupled to the at least one processor, the storage device having processor-executable instructions stored thereon, the instructions including instructions that, when executed by the at least one processor, cause the at least one processor to instantiate a data acquisition processor. The data acquisition processor is operable to acquire observations about a system, the observations comprising a specified system temperature measurement and an input power parameter measurement corresponding to the specified temperature measurement. The processor-executable instructions further cause the at least one processor to instantiate a relationship builder operable to learn a relationship between the specified system temperature measurement and the input power parameter measurement, and to learn a relationship between the specified system temperature measurement and a specified system temperature drop. The processor-executable instructions further cause the at least one processor to instantiate a temperature map comprising a plurality of cells, each cell corresponding to a temperature tuple consisting of the specified system temperature measurement, the temperature map being configured to receive and store, for each cell, from the relationship builder a summary statistic of the measured input power parameter corresponding to the cell's temperature tuple, or a measured-derived specified system temperature drop, or both. The processor-executable instructions further cause the at least one processor to calculate a predicted value of an input power parameter and a predicted value of a specified system temperature drop using the relationship, respectively, and are further configured to declare that system performance degradation exists using the predicted value of the input power parameter and the predicted value of the specified system temperature drop.

[0022] In general, in yet another aspect, the disclosed embodiments relate to a non-transitory computer-readable medium containing program logic that, when executed through the operation of one or more computer processors, causes the one or more processors to perform a method according to any of the embodiments described herein. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The foregoing and other advantages of the disclosed embodiments will become apparent upon reading the following detailed description and referring to the accompanying drawings, in which:

[0024] FIG1 shows a known HVAC&R system using a vapor compression cycle (VCC);

[0025] Figure 2 shows a simplified view of an exemplary HVAC&R system as a "black box" in accordance with aspects of the disclosed embodiments;

[0026] Figure 3 An exemplary HVAC&R system equipped with a monitoring and early problem detection system according to aspects of the disclosed embodiments is shown;

[0027] Figure 4A and Figure 4B shows a graph illustrating steady-state operation of an HVAC&R system according to aspects of the disclosed embodiments;

[0028] Figure 5A and 5B shows a block diagram illustrating how learned relationships may be used in monitoring and early problem detection systems in accordance with aspects of the disclosed embodiments;

[0029] Figure 6 An exemplary implementation of a monitoring agent according to aspects of the disclosed embodiments is shown;

[0030] Figure 6A An exemplary prediction processor according to aspects of the disclosed embodiments is shown;

[0031] Figure 6B A flow chart illustrating an exemplary VCC state generator according to aspects of the disclosed embodiments is shown;

[0032] Figure 6C A flow chart illustrating exemplary debounce logic that may be used with a VCC state generator in accordance with aspects of the disclosed embodiments is shown;

[0033] Figure 6D A flow chart illustrating exemplary stability logic that may be used with a VCC state generator in accordance with aspects of the disclosed embodiments is shown;

[0034] Figure 6E An exemplary relational learner according to aspects of the disclosed embodiments is shown;

[0035] Figure 6F A flowchart illustrating exemplary neighborhood extractor logic that may be used with a relational learner in accordance with aspects of the disclosed embodiments is shown;

[0036] Figure 6G A flowchart illustrating exemplary parameter prediction logic that may be used with a relational learner in accordance with aspects of the disclosed embodiments is shown;

[0037] Figure 6H An exemplary degradation detection processor according to aspects of the disclosed embodiments is shown;

[0038] Figure 6I An exemplary limit detector according to aspects of the disclosed embodiments is shown;

[0039] Figure 7 shows a timing diagram for constructing a temperature map according to aspects of the disclosed embodiments;

[0040] Figure 8 shows a flow chart for determining whether to compensate for an observation according to aspects of the disclosed embodiments;

[0041] Figure 9shows a functional block diagram for updating a residual sequence estimator according to aspects of the disclosed embodiment;

[0042] Figure 10 An HVAC&R system having multiple compressors equipped with monitoring agents according to aspects of the disclosed embodiments is shown;

[0043] Figures 11A to 11C shows an exemplary convex hull for determining whether to present a CIPP or ETD prediction in accordance with aspects of the disclosed embodiments;

[0044] Figure 12A and Figure 12B shows an alternative way of using learned relationships in accordance with aspects of the disclosed embodiments;

[0045] Figure 13 shows an alternative exemplary implementation of a monitoring agent in accordance with aspects of the disclosed embodiments;

[0046] Figure 14 An alternative exemplary joint relation learner according to aspects of the disclosed embodiments is shown;

[0047] Figure 15A A flowchart illustrating exemplary federated neighborhood extractor logic that may be used with an alternative exemplary federated relation learner in accordance with aspects of the disclosed embodiments;

[0048] Figure 15B A flowchart illustrating exemplary parameter prediction logic that may be used with an alternative exemplary joint relational learner in accordance with aspects of the disclosed embodiments; and

[0049] Figure 16 An exemplary implementation of a system parameter monitoring agent in accordance with aspects of the disclosed embodiments is shown. DETAILED DESCRIPTION

[0050] First, it should be understood that the development of actual, real-world commercial applications encompassing various aspects of the disclosed embodiments will require many implementation-specific decisions to achieve the developer's ultimate goal for the commercial embodiment. Such implementation-specific decisions may include, and may not be limited to, compliance with system-related, business-related, government-related, and other constraints, which may vary from time to time depending on the specific implementation, location, and environment. While the developer's efforts may be complex and time-consuming in absolute terms, such efforts will be routine tasks for those skilled in the art having the benefit of this disclosure.

[0051] It should also be understood that the embodiments disclosed and taught herein are susceptible to many and various modifications and alternative forms. Therefore, the use of singular terms, such as but not limited to "one," is not intended to limit the number of items. Similarly, any relative terms used in the written description, such as but not limited to "top," "bottom," "left," "right," "upper," "lower," "downward," "upward," "side," etc., are for clarity and specific reference to the drawings and are not intended to limit the scope of the present invention.

[0052] Various embodiments disclosed herein relate to systems and methods, as well as computer or processor-executable instructions, for early monitoring and detection of potential problems in a VCC-based HVAC&R system. As described above, the HVAC&R monitoring system and method employ a monitoring application or agent that uses continuous machine learning and one or more temperature maps to learn relationships between a measured compressor input power parameter (sometimes referred to as a "power parameter") (i.e., a fixed, measurable, positive-definite function of the power consumed by the compressor) and measured condenser and evaporator inlet fluid temperatures, the measured compressor input power parameter resulting from the application of one or more system compressors, and relationships between temperature parameters (e.g., measured evaporator inlet or discharge temperatures or corresponding measured (or measured-derived) evaporator temperature drops) and measured condenser and evaporator inlet fluid temperatures. These relationships are learned based on observations (i.e., measurements) of the inlet fluid temperature, evaporator discharge temperature, and compressor input power parameter for each operating compressor and evaporator temperature drop when the HVAC&R system is new or in a "fresh maintenance" condition. The monitoring agent can then use the learned relationships to predict the expected compressor input power parameters and evaporator temperature drop values ​​for the HVAC&R system, based on subsequent observations of the HVAC&R system, indicating a "new maintenance" condition. The agent can then compare the predicted compressor input power parameters and evaporator temperature drop values ​​with the observed compressor input power parameters and evaporator temperature drop values ​​to detect performance degradation early, infer the possible causes of the degradation, and issue appropriate alarm signals.

[0053] Thereafter, the agent may determine a relative coefficient of performance or relative COP (rCOP) for each compressor within the HVAC&R system using the predicted compressor input power parameters and evaporator temperature drop values ​​and the actual measured compressor input power parameters and temperature drop values ​​for each operating compressor.

[0054] In practice, the relative COP is the ratio of the coefficient of performance (COP) calculated from the measurements to the expected or reference COP, such that if the HVAC&R system is functioning properly, the relative COP should be 1 or 100%. In some embodiments, for a given set of measured condenser and evaporator inlet fluid temperatures, this relative COP can be calculated by calculating the ratio of the predicted compressor input power parameter value to the measured compressor input power parameter value multiplied by the ratio of the measured evaporator discharge temperature drop (ETD) to the predicted evaporator temperature drop (which is defined as the numerical difference between the evaporator inlet temperature and the evaporator discharge temperature).

[0055] Another aspect of the embodiments herein is a novel relational learner that uses a novel temperature map, relation builder, neighborhood extractor, and parameterized predictor to learn the relations required to calculate the normalized residuals and relative COP described above. The relational learner has the ability to learn to predict the attribute values ​​required to calculate the normalized residuals and relative COP of the present invention, but can also learn in the presence of degradation and can also determine when the predictions are likely to be accurate and when they are inaccurate.

[0056] In general, embodiments of the present disclosure may detect system degradation based on one or more of: power parameter prediction based on a 2-dimensional temperature map (via normalized residuals), evaporator temperature drop prediction based on a 2-dimensional temperature map (via normalized residuals), relative COP based on a ratio involving either the above-mentioned 2-dimensional power parameter or evaporator temperature drop prediction, power parameter prediction based on a 3-dimensional temperature map (via normalized residuals), evaporator temperature drop prediction based on a 3-dimensional temperature map (via normalized residuals), relative COP based on a ratio involving the above-mentioned 3-dimensional power parameter prediction, relative COP based on a ratio involving the above-mentioned 3-dimensional evaporator temperature drop prediction, or relative COP based on a ratio involving both the above-mentioned 3-dimensional power parameter and evaporator temperature drop prediction.

[0057] Now, the following discusses exemplary embodiments of predicted compressor input power parameter values ​​that use a learned relationship from observed time observations of measured compressor input power parameter values ​​and certain measured temperatures, as well as exemplary embodiments of predicted evaporator temperature drop values ​​that also use a learned relationship from observed time observations of measured or calculated evaporator temperature drop values ​​and certain measured temperatures. The specific temperatures measured can be the same for both learned relationships. Both relationships are learned using the novel relationship learner described above. These predicted values ​​can be combined with the corresponding observed values ​​to produce a sequence of metrics that can be used for early detection of performance degradation in HVAC&R systems. This is followed by a discussion of determining and using relative COP for early detection of performance degradation in HVAC&R systems, as well as metrics that quantify the costs associated with performance degradation. The use of these metrics, both individually and in combination, to infer possible causes of degradation and to issue appropriate alarm signals when observed will also be discussed.

[0058] As described above, the ability of the disclosed systems and methods to detect problems early stems from certain intuitions of the present inventors based on the observation that, given a set of measurable external conditions of temperature, evaporator fan speed (and, in some cases, condenser fan speed), and a known combination of compressor states (i.e., which compressors are on and off at the time in a multi-compressor system), the power consumed by the employed refrigerant compressor and the corresponding temperature drop in the vapor compression cycle are both time-invariant and repeatable in a steady-state environment, as long as the physical conditions of the system do not change. More specifically, assuming other aspects of the system remain constant, once the HVAC&R system has been operating long enough for the internal refrigerant conditions to stabilize, a known relationship exists and should exist between compressor input power parameters (e.g., real power, current, volt-amperes, etc.) and certain observed temperatures. This time-invariant relationship between the compressor input power parameter and the condenser and evaporator inlet temperatures represents the behavior of the HVAC&R system when placed under new maintenance conditions and can be learned by the relationship learner of embodiments herein. The resulting learned relationship is referred to herein as the compressor input power parameter (CIPP) relationship, or simply the "CIPP relationship."

[0059] Additionally, once the HVAC&R system has been running long enough for the aforementioned internal refrigerant conditions and the temperatures of the physical heat transfer mechanisms (e.g., fan coils, etc.) to stabilize, a known relationship exists and should exist between (a) the measured fluid temperature drop across the evaporator, or equivalently, the temperature drop across the evaporator (calculated as the difference between the measured fluid entry temperature and the measured fluid exit temperature of the evaporator), and (b) some observed temperature (assuming other aspects of the system remain constant). This time-invariant relationship between the temperature drop across the evaporator and condenser and the evaporator entry temperature, representing the behavior of the HVAC&R system when it is under new maintenance conditions, can be learned by the relationship learner of embodiments herein, and the resulting learned relationship is referred to herein as the evaporator temperature drop relationship, or simply "ETD relationship."

[0060] These learned CIPP and ETD relationships can be used individually and in combination to detect system degradation in many different applications (e.g., air conditioners, heat pumps, refrigerators, and other related systems), and can also be used to infer the likely conditions leading to degradation and issue appropriate alarm signals. Note that in a heat pump system designed to transfer heat from an external ambient heat source to a fluid such as air or water, the evaporator temperature drop as defined above will be a negative quantity.

[0061] Referring now to FIG. 1 , a flow diagram of a basic HVAC&R system 100 employing a vapor compression cycle is shown. The aforementioned CIPP relationship can be illustrated by examining the VCC-based system 100 in FIG. This system 100 represents the majority of HVAC&R systems deployed today, and therefore the discussion herein focuses primarily on early monitoring and detection of problems within such systems. Those skilled in the art will appreciate that the principles and teachings herein are equally applicable to other types of HVAC&R systems and equipment available to commercial and industrial users. Indeed, the principles and teachings discussed herein are generally applicable to any deterministic system or equipment where a parameter result or value is reliably generated for a given parameter of interest and, therefore, can be rapidly learned and predicted given another parameter or set of parameters (and their values) using the techniques described herein. Such deterministic systems and equipment are numerous and varied and involve many types of parameters, such as flow control parameters (e.g., flow rate, viscosity, etc.), power control parameters (e.g., voltage, current, etc.), motion control parameters (e.g., speed, altitude, etc.), and the like.

[0062] The operation of the HVAC&R system 100 is well known in the art and will only be described generally here. Starting at point "A" in the diagram, refrigerant in the form of a low-pressure vapor is extracted via suction from the evaporator 102, which is essentially a heat exchanger that absorbs heat from the fluid (i.e., air) at the evaporator environment 103 and transfers the heat to the refrigerant flowing within the evaporator to the compressor 104. The compressor 104 receives the low-pressure vapor, compresses it into a high-pressure vapor, and sends it to the condenser 106, in the process raising the temperature of the refrigerant to a temperature higher than that of the fluid (i.e., air in the case of a direct exchange system) at the condenser environment 107.

[0063] At the condenser 106, the condenser coil (not explicitly shown) allows heat from the higher temperature vapor refrigerant to be transferred to the lower temperature condenser ambient fluid as indicated by arrow H. c As shown in FIG. 1 . This heat transfer causes the high-pressure vapor refrigerant in the condenser coil to condense into a liquid. From the condenser 106, the liquid refrigerant (still at high pressure) enters the expansion valve 110, which atomizes the refrigerant and releases (i.e., sprays) it as an aerosol into the evaporator 102. The temperature of the liquid refrigerant drops significantly as it moves from the inlet side of the expansion valve 110, which is at high pressure, to the outlet side of the expansion valve 110, which is at a relatively low pressure.

[0064] At the evaporator 102, the reduced temperature refrigerant cools the evaporator coil (not explicitly shown) to a temperature much lower than the evaporator ambient fluid in a normal operating system, thereby absorbing heat and causing the refrigerant to evaporate into vapor in the process. Heat from the evaporator ambient fluid flow is then absorbed by the evaporator coil (not explicitly shown) in the process, as shown by arrow H. e The low pressure vapor in the evaporator is then pulled into the compressor 104 via suction at A, and the cycle repeats.

[0065] In FIG1 , compressor 104 is driven by compressor motor 104a, which is powered by an AC power source (e.g., AC power line 112). AC power line 112 provides power from AC mains 118, which is typically fed through branch feeder circuits 114. Branch feeder circuits 114 are used to isolate HVAC&R system 100 and provide short-circuit and overcurrent protection for HVAC&R system 100. Many branch feeder circuits have current or power measurement capabilities built into their circuit breakers or otherwise embedded, which can provide a signal indicating the input power used by the load. Examples include the NQ and NF series distribution panels with integrated energy meters from Schneider Electric USA. In some installations, HVAC&R system 100 may also include auxiliary equipment (shown in dashed lines), generally indicated at 116, such as fans and other auxiliary electrical loads, electrical disconnect boxes, etc., which also receive power from feeder circuits 114. The auxiliary equipment 116 is typically located within the physical housing that also houses the compressor of the system 100 and may be connected in series or in parallel with the motor 104a.

[0066] As will be explained in the following description, one way to detect system degradation is by monitoring the input power consumed by the compressor motor 104a on the feeder circuit 114 and the AC power line 112, and comparing this compressor input power with the compressor input power predicted by the CIPP relationship described above. Generally, if the comparison indicates that the observed compressor input power differs (i.e., is greater than or less than) the compressor input power predicted by the CIPP relationship by more than a predefined threshold amount (e.g., 5%, 10%, 15%, etc.), this may be an indication of performance degradation.

[0067] As used herein, the terms "evaporator environment" and "condenser environment" refer to the ambient environment surrounding the evaporator and condenser functions, respectively. When system 100 is operating in air conditioning mode or as a refrigerator, the evaporator environment is the space to be cooled or "conditioned," and is typically a building or room, but can also be the interior space of a refrigerator or freezer or a food storage area. In this mode, the condenser environment is typically the outdoor environment in the case of air conditioning and some refrigeration systems, and can be the indoor environment outside the equipment in the case of refrigeration. In other words, a direct exchange air conditioner or refrigerator absorbs heat from the air in the conditioned space and rejects the heat to the outdoors or external environment. When system 100 is operating in heating mode as a heat pump, the roles of physical condenser 106 and physical evaporator 104 are reversed, such that physical condenser 106 is used to absorb heat from the nominally cooler outdoor environment, and physical evaporator 102 is used to deliver heat to the heated building or room.

[0068] The HVAC&R system 100 of FIG1 is a "direct exchange" system, in which heat is transferred directly to and from the ambient air surrounding the evaporator and condenser via the evaporator 102 and condenser 106. However, the embodiments disclosed herein are also applicable to non-direct exchange systems, including "indirect exchange" systems, such as a chiller operating as an air conditioner, or a geothermal heat pump. In a chiller, the evaporator cools a fluid, such as chilled water, which is then distributed throughout the building to independently cool the spaces therein via heat exchangers located remotely from the chiller. In some systems, heat is rejected from the condenser into a liquid fluid (such as water or an antifreeze solution), which is then transferred to a cooler environment via, for example, a cooling tower. Thus, the disclosed embodiments can be used with systems that transfer heat directly to and from the air in the desired space, as in conventional direct exchange systems, or with indirect exchange systems that transfer heat to and from a liquid fluid (such as water), which is then used to cool or heat the desired space.

[0069] In the following description, the term "fluid temperature," when used to describe the inlet or outlet temperature of an evaporator or condenser (or its function), will be understood to refer to air in the case of a direct exchange system and to refer to a liquid or fluid in the case of an indirect exchange system such as a chiller. Mixed-mode systems, such as geothermal heat pumps that use water or antifreeze to exchange heat with the ground and air to exchange heat within a building, are also within the scope of the disclosed embodiments.

[0070] Figure 2 A simplified view of an HVAC&R system 100 is shown in the form of a so-called "black box" 200 having certain inputs and outputs. Treating the HVAC&R system 100 in this manner allows the system to be analyzed in terms of its external inputs and outputs (i.e., its transfer characteristics) without interfering with the internals of the HVAC&R system, which makes the invention disclosed herein ideal for retrofit applications to existing systems. When viewed as a black box 200, the inputs to the system 100 include: a heat exchanger having a specific heat capacity C 200; ... pc , with mass flow rate And at a certain temperature T ci The condenser enters the fluid under operation; with a specific heat C pe , with mass flow rate And at a certain temperature T ei The evaporator enters the fluid under operation; and the compressor input power W with a measurable power parameter P c The output from the black box 200 includes: pc , with mass flow rate And at a certain temperature T cdA condenser discharge fluid operating under the condition of pe , with mass flow rate And at a certain temperature t ed The evaporator discharges fluid when operating under

[0071] As an additional simplification, it may be assumed that the specific heats C of the fluids moving through the condenser and evaporator, respectively, are pc and C pe does not vary with time. This is generally true to a first order approximation. Furthermore, for the system 100 operating in steady state, the mass flow rates through the condenser and evaporator are and This is the case in the simplest system, where one or more single-speed fans are employed in normal operation to move fluid through the condenser and evaporator assemblies (the single-speed fans run continuously and are not cycled on and off with temperature or pressure to maintain head pressure).

[0072] The condenser inlet and outlet fluids have the same specific heat and mass flow rate, which comes from the fact that 1) they are the same fluid, and 2) the physical system viewed in this way has no fluid storage capacity, so the net mass flow must be zero. This is also the case for the evaporator fluid.

[0073] The above assumptions are the design basis for most HVAC&R systems operating in a steady state where the temperature is regulated by cycling the compressor on and off as needed to maintain the temperature within a selected range. This represents the majority of HVAC&R systems in use today, including most residential split and unit systems and simple refrigerators. For such HVAC&R systems, it has been found that the condenser entering fluid temperature, T ci , evaporator inlet fluid temperature T ei and the compressor input power parameter P are sufficient to establish a first time-invariant relationship that can be used to detect system degradation when the vapor compression cycle is operating in steady state. Similarly, it has been found that the condenser inlet temperature T ci , evaporator inlet temperature T ei and evaporator discharge temperature T ed It is sufficient to establish a second time-invariant relationship that can be used to detect system degradation when the vapor compression cycle is operating in steady state.

[0074] Likewise, increased refrigerant temperature in the condenser or evaporator function generally results in increased refrigerant pressure within the refrigerant circuit, and more compressor power is required to maintain the pressure and move the refrigerant through the system. The power required to move the refrigerant through the system also depends on the amount of refrigerant in the circuit, as does a drop in evaporator temperature.

[0075] refer to Figure 2 The HVAC&R system 100 discussed in the present invention is a simplified view of a black box 200, considering that the system experiences a specific pair of condenser and evaporator entering fluid temperatures (T ei ,T ci ) is considered to be the condition of the fluid under "new maintenance" conditions. It is also considered that the system is in a "new maintenance" condition, and the mass flow rates through the condenser and evaporator coils are also fixed and nominal. The term "new maintenance" condition as used herein refers to the condition of the HVAC&R system immediately after it has been properly serviced, wherein the intention of the service is to put the system in the best possible condition (i.e., as close to factory specifications as is practical for the life of the system). As described above, for the system 100 operating in this state, both the compressor power consumed and the evaporator temperature drop should be repeatable, meaning that any time the system 100 is subjected to this same set of conditions, once the refrigerant conditions have stabilized, the power consumed by the compressor and the evaporator temperature drop should be the same. At the same temperature tuple (T ei ,T ci ), any condition that reduces the rate at which heat is extracted from the condenser coil will increase the temperature of the refrigerant in the condenser, causing the pressure in the condenser to increase, and causing the compressor to consume more power than it would otherwise. These conditions include things that will reduce the mass flow rate, such as a failed condenser fan, obstructions in the condenser (including extreme condenser fouling), and surface effects (such as condenser fouling), even if the ultimate mass flow rate is not reduced. Therefore, if for a given set of inlet temperatures (T ei ,T ci )'s compressor power is higher than expected, then: 1) there is a problem with the system and its efficiency may have degraded, and 2) the likely cause of the problem is a problem in the condenser subsystem.

[0076] In a similar way, for the incoming fluid temperature tuple (T ei ,T ci ), any condition that results in a reduced rate of heat absorption in the evaporator will result in a reduced average internal temperature of the fluid in the evaporator, thereby resulting in a reduced pressure, and resulting in a reduced compressor power. This includes phenomena such as fouling the evaporator via accumulation of dirt or frost, which reduces the rate of heat transfer from the evaporator coil to the evaporator fluid, or anything that results in a reduction in the mass flow of the evaporator fluid, which can include the phenomena mentioned above, but also dirty filters, broken evaporator fan belts, and other phenomena. Therefore, in addition, if for a given set of inlet temperatures (T ei ,T ci )'s compressor power is lower than expected, then: 1) there is a problem with the system and its efficiency may have degraded, and 2) the likely cause of the problem is some problem in the evaporator subsystem.

[0077] For a pair of fixed condenser and evaporator inlet mass flow rates and temperatures that are equal, the power required to move refrigerant through the system is a positive definite function of the total amount of refrigerant moved through the system. Importantly, refrigerant leakage (which is very common in HVAC&R systems and affects both system efficiency and the environment via ozone depletion) manifests as a general reduction in compressor power, regardless of inlet temperature.

[0078] Thus, for the basic HVAC&R system 100 described above, information about the overall health of the system can be obtained from a simple black box model where the system is in a “fresh maintenance” condition based on the incoming fluid temperature (T ei ,T ci ) and a compressor input power parameter P to learn a CIPP relationship. Once the learned CIPP relationship is established, it can be used to predict potential performance degradation and problems based on subsequent observations (i.e., measurements) of certain compressor input power parameters. The observed compressor input power parameters may include, for example, real power, current (e.g., one of two phases of current), volt-amperes, etc.

[0079] Additional or alternative information about the overall health of the system can also be obtained from simple black box models where the inlet fluid temperature (T ei ,T ci ) Learn the ETD relationship, where the evaporator temperature drop is calculated as the evaporator entering temperature T observed when the system is in the "new maintenance" condition ei Compared with the observed evaporator discharge temperature T ed Once this learned ETD relationship is established, it can be used to predict potential performance degradation and problems based on calculating the evaporator temperature drop from subsequent observations of the evaporator entry and discharge temperatures.

[0080] Next, refer to Figure 3 In accordance with an embodiment of the present disclosure, an HVAC&R monitoring and early problem detection system 300 has now been installed on the HVAC&R system 100. The monitoring and early problem detection system 300 is designed to learn and use the CIPP relationship and the ETD relationship discussed above to monitor performance degradation in the HVAC&R system 100. To this end, the system 100 is equipped with a plurality of temperature sensors, such as sensors 302, 304, 306, and 308, installed at selected points on the system. These temperature sensors 302, 304, 306, and 308 obtain selected temperature measurements that can be used by the monitoring and early problem detection system 300: (i) condenser inlet fluid temperature, T ci (ii) Condenser discharge fluid temperature T cd (iii) Evaporator inlet fluid temperature T ei, commonly referred to as the “return” temperature in commercial and residential direct exchange air conditioners; and (iv) the evaporator discharge fluid temperature, T ed , often referred to as the "supply" temperature in commercial and residential direct exchange air conditioning systems.

[0081] Although four temperature measurements are mentioned, the CIPP aspect of the monitoring and early problem detection system 300 can operate using only two of the four temperature measurements: the evaporator's inlet or outlet fluid temperature (T ei or T ed ), and the condenser inlet or outlet fluid temperature (T ci or T cd ), depending on the implementation. For example, in one embodiment, the monitoring and early problem detection system 300 may use the fluid temperature T at the inlet of the evaporator 102 ei and the fluid temperature T at the inlet of the condenser 106 ci , and these temperature measurements are preferred when readily available because they are not directly affected by the operation of the HVAC&R system. Therefore, in one embodiment, a temperature sensor 302 is mounted at or near the inlet of the evaporator 102 to measure the evaporator inlet fluid temperature T ei , and the second temperature sensor 304 is installed at or near the inlet of the condenser 106 to measure the condenser inlet fluid temperature T ci Alternatively, in some embodiments, the condenser discharge fluid temperature T cd Can replace T ci , or the evaporator discharge fluid temperature T ed Can replace T ei In such an embodiment, a third temperature sensor 306 may also be optionally installed at the discharge outlet of the evaporator 102 to measure the evaporator discharge fluid temperature T ed Alternatively, a fourth temperature sensor 308 may be optionally installed at the outlet of the condenser 106 to measure the condenser outlet fluid temperature T cd These temperature sensors 302, 304, 306, and 308 may be any suitable temperature sensors known to those skilled in the art, including voltage-based temperature sensors employing thermocouples or thermistor devices.

[0082] In addition to the incoming fluid temperature measurement, measurements of the compressor input power parameter are also obtained for use in the monitoring and early problem detection system 300. Examples of compressor input power parameter measurements that may be obtained include measurements of current, real power, reactive power, and apparent power, as well as voltage in some embodiments. As discussed further below, the compressor input power parameter that is typically measured is current because current measuring equipment is relatively low cost compared to, for example, a power meter. Furthermore, most power meters and other power measurement equipment already require current measurements for actual power measurements. Therefore, compressor input current is almost always one of the compressor input power parameters that is measured.

[0083] To implement the ETD aspect of the monitoring and early problem detection system 300, the evaporator inlet temperature T ei and evaporator discharge temperature T ed Both are used to establish the evaporator temperature drop. As in the case of CIPP, the condenser temperature (T ci or T cd , where T ci is preferred because it is not directly affected by the operation of the HVAC&R system) to fully establish the ETD relationship. Therefore, three temperature measurements are used to implement the ETD aspect of the monitoring and early problem detection system 300.

[0084] In a typical residential installation, the compressor 104 (and motor 104a) are fed via AC power line 112 from a branch feeder circuit 114, which is fed by an AC mains 118. In some systems, the AC power line 112 can be a 3-wire single-phase power line with a midpoint neutral. Other configurations are also possible, including two-wire AC systems and 3-phase AC configurations. Thereafter, one or more current sensing devices 310, such as one or more toroidal current transformers, can be installed on the wires of the compressor power line 112. The output of the one or more current transformers 310 is then provided to a power parameter meter 312, which can be any commercially available power meter or meter that can measure the current (e.g., RMS current) flowing through the power line 112. Some models of the power parameter meter 312 can also include a measurement of line voltage, such as a model that measures real power and apparent power (volt-amperes) in single-phase or multi-phase form. An example of a commercially available power meter that can be used as power parameter meter 312 is any of the PM8xx series power meters manufactured by Schneider Electric, which have associated circuitry for measuring active power. In systems where the line voltage remains constant or at least repeatable relative to the configuration of compressors 104 in the system, a simple clamped current transformer that can measure the current in one branch of compressor 104 may also be sufficient.

[0085] For embodiments using the CIPP relationship to estimate compressor input current, the apparatus may include one or more current transformers and other current measuring devices. Current measuring devices are available that can provide an indication of the RMS current flowing through power line 112 within a specified current range. In these embodiments, the RMS current delivered to compressor 104 alone may be sufficient as a compressor input power parameter measurement. An example of a current measuring device suitable for some HVAC&R applications is the Veris H923 split-core current sensor from Veris Industries, which can provide a 0-10 volt signal in response to 0-10 amps RMS current. Other similar current measuring devices or systems may be employed, as appropriate for the expected current levels in the system.

[0086] In some embodiments, instead of (or in addition to) measuring the compressor input power parameter, the process of learning the CIPP relationship described herein can be performed using an indication of the power consumed by the HVAC&R system 100 as a whole via the branch feeder circuits 114, as provided by the AC mains 118. As previously mentioned, many branch feeder circuits have current or power measurement capabilities built into their circuit breakers or otherwise embedded, which can provide a signal indicating the input power being used by the system. Some auxiliary devices 116 (such as electrical disconnect boxes, etc.) include similar current or power measurement capabilities. Therefore, although the present disclosure primarily describes the CIPP relationship learning process with respect to compressor input power parameter measurements, those skilled in the art will appreciate that the relationship can also be learned in a similar manner using the above-mentioned alternative (or additional) input power indicators.

[0087] The measured current or other compressor input power parameter can then be used to measure the evaporator's inlet or outlet fluid temperature (T ei or T ed ) and the condenser inlet or outlet fluid temperature (T ci or T cd ) to establish the CIPP relationship. In some embodiments, and by way of example only, the specific fluid temperature measurement used may be the evaporator inlet fluid temperature T ei and the condenser inlet fluid temperature T ci This is the measurement of Figure 3 In other embodiments, the fluid temperature measurement used may be the evaporator discharge fluid temperature T ed and the condenser discharge fluid temperature T cd In other embodiments, a combination of condenser entry temperature and evaporator discharge temperature may be used, or a combination of condenser discharge temperature and evaporator entry temperature may be used.

[0088] The two fluid temperature measurements (one from sensor 302 or 304 and one from sensor 306 or 308) and the compressor input power parameter measurement (from power parameter meter 312) can then be provided to an HVAC&R monitoring application or agent 314 for use in determining an expected compressor input power based on the CIPP relationship. The HVAC&R monitoring agent 314 can then compare the expected compressor input power with the observed (i.e., measured) compressor input power to detect potential system degradation and problems. The fluid temperature and compressor input power measurements can be provided to the monitoring agent 314 via any suitable signal connection, including wired (e.g., Ethernet, etc.), wireless (e.g., Wi-Fi, Bluetooth, etc.), and other connections. For example, the measurements from sensors 302, 304, 306, and / or 308 can be provided to the monitoring agent 314 as part of the Internet of Things (IoT).

[0089] In some embodiments, the monitoring agent 314 can be implemented as a cloud-based solution or a fog-based solution, wherein a portion or all of the monitoring agent 314 resides or is hosted on a network 316. The network 316 can be a remote network such as a cloud network, or it can be a local network 316 such as a fog network. Such a monitoring agent 314 (or portions thereof) can also be integrated into a so-called "smart" thermostat for an air conditioning system or HVAC&R controller. A "smart" thermostat or HVAC&R controller can include any programmable device that can be configured to input multiple data signals (e.g., analog signals, digital signals, etc.), execute an algorithm or software routine based on these data signals, and output one or more data signals (e.g., analog signals, digital signals, etc.). Other examples of commercially available devices that may be suitable for use with the monitoring agent 314 include commercially available programmable logic controllers (PLCs) and building management systems (BMSs), both manufactured by Schneider Electric.

[0090] In such Figure 3 In the HVAC&R system shown, it has been observed that once the system has changed the compressor state by turning the compressor "on" in a single compressor system, or more generally by changing a combination of compressor on / off states in a multiple compressor system, an accurate prediction of the compressor input power parameter can be obtained by waiting until the system has been running long enough so that the refrigerant state has stabilized in the system (referred to herein as having achieved a "steady state" of refrigerant operation). Figure 4A The meaning of "refrigerant steady state" operation of the VCC cycle with respect to the compressor input power parameter is shown, dividing a single VCC cycle into three operating intervals. Similarly, Figure 4BThe "steady state" of the refrigerant operation of the VCC cycle relative to the temperature drop across the evaporator is shown, again dividing a single VCC cycle into three operating intervals.

[0091] exist Figure 4A In FIG. 4 , a graph 400 of current (amperes) versus time (seconds), labeled “Icomp,” is shown for a typical “on” cycle of a single compressor system similar to system 100 described above. Graph 400 also shows the predicted compressor input current, labeled “Iest,” for that compressor cycle using the CIPP relationship learned for that system. From this graph, three distinct operating regions can be identified within the compressor cycle, including the region defined by the parameter t Plead The power parameter leading blanking interval 402, the power parameter prediction interval 404 and the parameter tPla indicated g Indicated by the hysteresis blanking interval 406, during the power parameter prediction interval 404, the power parameter value should be predictable from the learned compressor input power parameter relationship described above. In practice, only observations during the power parameter prediction interval 404 are useful for training the agent to learn the CIPP relationship and use it to predict equipment conditions. Observations during this power parameter prediction interval are the previously mentioned "refrigerant steady-state" observations.

[0092] The power parameter lead blanking interval 402 refers to the interval immediately after the compressor has been turned on. In a single compressor system, when the compressor has been turned off and then turned on, there is a transition period in which the compressor current represented by line 408 is a function not only of the temperature and mass flow rate, but also of the time that has elapsed since the compressor was turned on. In a multi-compressor system, this transition period occurs any time the combination of compressor on / off states changes. This transition period is highly system dependent. While the transition behavior can be repeatable, it is generally unpredictable using a time-invariant CIPP relationship. The power parameter lead blanking interval 402 is preferably required to ensure that observations made during this interval are discarded. Generally, the power parameter lead blanking interval 402 should be set long enough to allow the refrigerant circuit to reach "steady state" operation, which can vary depending on the size and type of system. For example, in a residential refrigerator, the lead blanking interval may be set to as little as 20-30 seconds and the entire compressor cycle may last only a minute or two, whereas in a large rooftop unit, a lead blanking interval 402 of about 5-10 minutes may be required and the compressor may run for hours or even over the course of a day. In some large chillers, blanking intervals of up to 30 minutes and longer are appropriate and the chiller may run uninterrupted for days. Figure 4A As shown in the figure, the interval is 250 seconds.

[0093] The power parameter prediction interval 404 refers to the interval during which the agent has declared that the HVAC&R system has reached a steady state for power parameter prediction. Observations made during the power parameter prediction interval 404 can be used to inform the CIPP relationship, and the learned CIPP relationship can then be applied to predict the power parameters, indicated by line 410, which should support the temperature and mass flow rate of the condenser and evaporator fluids. In the simplest HVAC&R system, the condenser and evaporator inlet temperatures are sufficient to accurately predict the compressor input power parameters, assuming there are no physical changes in the system. From Figure 4A As can be seen in FIG4 , when the system is operating normally, the predicted current (Iest) 410 effective within the power parameter prediction interval 404 tracks the measured current 408 very accurately. The power parameter dynamic prediction interval 404 lasts until just before the compressor is turned off again. The power parameter hysteresis blanking interval 406, shown greatly exaggerated in FIG4 , refers to the interval when the compressor is turned "off" again and is primarily included to facilitate the needs of the sampled data system, as will be discussed later.

[0094] Figure 4B Shown for Figure 4A A similar graph 420 of evaporator temperature drop (degrees Celsius) versus time (seconds) for the same compressor "on" cycle is shown in FIG. Figure 4B In the graph, the measured evaporator temperature drop is labeled "ETD Actual" (428) and the estimated evaporator temperature drop is labeled "ETD Est" (430). Figure 4A Three different operating ranges are identified on the same compressor cycle, including the range determined by the parameter t Elead The indicated evaporator temperature drop advance blanking interval 422, the evaporator temperature drop prediction interval 424 (wherein the evaporator temperature drop value should be predictable from the evaporator temperature drop relationship known above) and the parameter tEla g The indicated evaporator temperature drop hysteresis blanking interval 426. In practice, only the observations within the evaporator temperature drop prediction interval 424 are useful for training the agent to learn the ETD relationship and use it to predict equipment conditions. Observations within this evaporator temperature drop prediction interval are referred to as "thermal steady-state" observations.

[0095] The evaporator temperature drop leading blanking interval 422 is the interval immediately after the compressor has been turned on and can be timed to coincide with the evaporator temperature drop leading blanking interval 422. Figure 4AThe power parameter lead blanking interval 402 is different. This transition period 422 is highly system dependent. While the transition behavior may be repeatable, it is unpredictable using a time-invariant ETD relationship. It is desirable to have an evaporator temperature drop lead blanking interval 422 to ensure that observations made during this interval are discarded. Typically, the evaporator temperature drop lead blanking interval 422 should be set long enough to allow not only the refrigerant circuit to reach "refrigerant steady state" operation, but also to allow the bulk condenser and evaporator coil temperatures to stabilize. As shown above Figure 4A In the discussion of Figure 4B The value in the middle is 500 seconds, as shown in the figure.

[0096] The evaporator temperature drop prediction interval 424 refers to the interval in which the agent has declared that the HVAC&R system has reached a steady state for the purpose of evaporator temperature drop prediction. Observations made during the evaporator temperature drop prediction interval 424 can be used to inform the EDT relationship, and the subsequently learned EDT relationship can be applied to predict the evaporator temperature drop that is effective within the evaporator temperature drop prediction interval 424 (indicated by line 430). As described above, in the simplest HVAC&R system, the condenser and evaporator entering temperatures are sufficient to accurately predict the evaporator temperature drop within the evaporator temperature drop prediction interval 424, provided that there are no physical changes in the system. The evaporator temperature drop prediction interval 424 lasts until just before the compressor is about to become off again. The evaporator temperature drop hysteresis blanking interval 426 (in Figure 4B is shown greatly exaggerated and is represented by the parameter tEla g indication) refers to the interval when the compressor becomes "off" again and is mainly included to facilitate the needs of the sampled data system as will be subsequently amplified.

[0097] Figure 5A and 5B Conceptually illustrating how an HVAC&R monitoring agent (such as agent 314) can use the previously mentioned CIPP and ETD learned relationships according to aspects of the disclosed embodiments. These figures illustrate how agent 314 can use previously learned CIPP or ETD relationships to generate a time series of normalized residuals, which can then be used as a metric to detect potential performance degradation and problems early in the HVAC&R system 100. As used herein, the term "metric" can be conceptually understood as a type of "distance" between how the system currently behaves and how the system should behave. A preferred mechanism for agent 314 to learn relationships (i.e., via a relationship learner) will be discussed later.

[0098] refer to Figure 5A, P(k) is the observed compressor input power parameter of the system 100 for the kth observation in a series of observations. In some embodiments, the evaporator inlet fluid temperature T ei (k) and the condenser inlet fluid temperature T ci (k) Making Observations. The term "simultaneously" refers to individual power parameter and temperature measurements being made rapidly in time relative to the thermal time constant of the system 100, and when assembled together as an observation, these measurements can be assumed to represent the "state" of the HVAC&R machine at that instant or within a short time window. Preferably, the temperature and compressor input power parameter measurements for a given observation are obtained within a time window of a few seconds, and preferably by a PLC (Programmable Logic Controller)-based process. Such an automated measurement process can generally obtain measurements at a sufficiently high rate for the monitoring purposes herein. The HVAC&R system 100 should also be in a refrigerant steady state, i.e., using only refrigerant steady state observations as described above, meaning that the system has been operating for a sufficient period of time such that the refrigerant in the system is in the appropriate physical state (i.e., liquid or vapor) throughout the system and that heat transfer is occurring at a substantially constant rate (e.g., within 1%-2%) in both the condenser and evaporator (i.e., within the power parameter prediction interval 404).

[0099] Using the above setup, agent 314 can calculate the power parameter value The predicted value sequence and the corresponding normalized residual sequence R corresponding to the kth observation falling within the power parameter prediction interval 404 p (k). Figure 5A For each observation k in the power parameter prediction interval 404, the evaporator inlet fluid temperature and condenser inlet fluid temperature tuple (T ei (k),T ci (k)) is applied to the learned CIPP relationship block 500, where the agent 314 uses the observations and the previously learned CIPP relationship to predict the power parameters that represent the system 100 in the new maintenance condition From the learned CIPP relationship block 500, the agent 314 generates the compressor input power parameter according to the learned CIPP relationship The predicted value of is shown in equation (1):

[0100]

[0101] The predicted compressor input power parameters are then combined at summing node 502 and the observed value of the compressor input power parameter P(k) included in the kth observation. The summing node 502 generates the differential compressor input power parameter value ΔP(k) according to equation (2):

[0102]

[0103] Thereafter, the agent 314 normalizes the compressor input ΔP(k) power parameter difference at a normalization block 504 to produce a normalized residual compressor input power parameter R p (k), as shown in equation (3):

[0104]

[0105] As shown in Equation (3), the normalized residual R corresponding to the kth observation is P (k) is the ratio of the difference between the measured and predicted values ​​of the compressor input power parameter ΔP(k) to the predicted value of the power parameter. Then, the normalized residual R p (k) can be expressed as a percentage by multiplying by 100 to show the percentage difference between the expected value of the compressor input power parameter and the observed value of the compressor input power parameter according to equation (4):

[0106] %R P (k)=100*R P (k) (4)

[0107] In a similar way, Figure 5B A conceptual block diagram is shown that illustrates how the agent 314 can use the learned ETD relationship to generate a time series of normalized evaporator temperature drop residuals to detect potential performance degradation and problems early in the HVAC&R system 100. The details of learning this relationship will be discussed later. The three temperatures corresponding to the kth observation of the system are used as inputs to the process. Again, the evaporator inlet fluid temperature T is observed simultaneously. ei (k), evaporator discharge temperature T ed (k) and the condenser inlet fluid temperature T ci (k) Likewise, the term "simultaneously" means that the measurements are taken rapidly in time relative to the thermal time constant of the HVAC&R system 100. As described above, preferably, the temperature measurements for a given observation are obtained within a time window of a few seconds, and preferably by a PLC-based process.

[0108] When learning and using the ETD relationship, the HVAC system 100 should also be in a thermally stable state, i.e., using only the thermally stable state observations described above, this means that the system has been operated for a sufficient period of time such that not only is the refrigerant in the system in the appropriate physical state (i.e., liquid or vapor) throughout the system, but also that heat transfer occurs at a substantially constant rate (e.g., within 1%-2%) in the condenser and evaporator. This means that the system should have been operated for a sufficient period of time such that the external temperatures of the heat exchangers, i.e., those surfaces of the heat exchangers that come into contact with the external fluid, are also substantially stable, i.e., within the evaporator temperature drop prediction interval 424. In practice, it has been observed that the time required to reach a thermally stable state as described above can be significantly longer than the time required to achieve a refrigerant steady state, e.g., typically about 8-15 minutes for a simple residential air conditioner.

[0109] refer to Figure 5B , the agent 314 calculates the kth observed evaporator temperature difference, or equivalently the evaporator temperature drop E(k), which is defined as the difference between the observed evaporator entry and discharge temperatures:

[0110] E(k)=T ei (k)-T ed (k) (5)

[0111] Evaporator inlet fluid temperature T ei (k) and the condenser inlet fluid temperature T ci (k) is provided to the learned ETD relationship block 506. The learned ETD relationship block 506 uses the learned ETD relationship (learned via a relationship learner, described later) to predict the corresponding expected evaporator temperature drop It is the VCC cycle in the heat (T ei (k),T ci (k)) is a function of the kth observed evaporator entering fluid temperature and condenser entering fluid temperature tuple at steady state, the prediction representing the expected evaporator temperature drop of the HVAC&R system 100 under the new maintenance conditions:

[0112]

[0113] The predicted evaporator temperature drop is then combined at summing node 508 and the observed evaporator temperature drop E(k), which is calculated from the kth observed temperature measurement using equation (5). The summing node 508 generates the evaporator temperature drop difference ΔE(k) according to equation (7):

[0114]

[0115] The normalized temperature drop residual R is then formed in the normalization block 510E (T), as shown below:

[0116]

[0117] The result can again be multiplied by 100% to show the percentage difference between the expected evaporator temperature drop and the evaporator temperature drop calculated using the observed values ​​of the evaporator entry and discharge temperatures:

[0118] %R E (T)=100*R E (T) (9)

[0119] It has been observed empirically that Figure 5A and 5B The normalized residuals in have the property of promoting continuous learning of the CIPP and ETD relationships, even when the system experiences performance degradation. Although the power consumed by the compressor and the evaporator temperature drop are the observed temperature tuples (T ei ,T ci ), but the normalized residuals of both are approximately or quasi-temperature independent. This means that the normalized residual of the compressor input power parameter or the evaporator temperature drop calculated at one temperature tuple is observed to have approximately the same value as the corresponding normalized residual value at any other temperature tuple within the system's typical operating temperature range, while the physical conditions of the system remain unchanged.

[0120] The observed normalized residual sequence R P (k) and R E The quasi-temperature independence of (k) serves two useful purposes in the examples herein. First, the temperature-independent normalized residual R P (k) and R E (k) can be used directly as a metric to detect system degradation. If the system is in the new maintenance condition and there are no measurement errors, the measured value of the power parameter and the measured value of the power parameter correctly predicted by the learned CIPP relationship block 500 should be consistent, and the corresponding normalized residual should be zero. In some embodiments, the deviation from the new maintenance condition due to degradation can be inferred from the non-zero normalized residual of the compressor input power parameter, where the magnitude of the deviation is used as an indication of the severity of the degradation, i.e., interpreted as a measure of "distance" from normal operation. The sign of the residual, indicating whether the measured power parameter value is greater than or less than the predicted value, can be used to infer the possible cause of the observed degradation and to issue an appropriate alarm signal. Due to the quasi-temperature independence of the normalized residual, this "distance" does not vary substantially with the measured inlet temperature T ee and T ciThe normalized residual of the evaporator temperature drop, or the deviation of a sequence of normalized residuals, can be used to indicate system degradation in the same manner. In other embodiments, a power parameter and the normalized residual of the evaporator temperature drop can be combined to serve as an indicator of system degradation. In other embodiments, the relative COP of the system, discussed later, can be included in the detection of system degradation and provide another metric indicating the resulting loss in system efficiency. As more information is added to the degradation detection process, the agent's ability to infer costs and possible causes of degradation improves.

[0121] The observation that the normalized residuals calculated above are at least quasi-temperature independent also allows the relationship learner to "correct" for degradation of the power parameter measurements and the evaporator temperature drop measurements in order to learn the CIPP or ETD relationship in the manner described subsequently. It is clear from equation (5) that the evaporator temperature drop E(k) is the product of the temperature drop from T ei and T ed is uniquely determined in the measurement and can be equivalently used Figure 5B T in ed The learning relationship of evaporator temperature drop is replaced by the learning relationship of ei Subtract T from the measured value ed The normalized residual of the evaporator temperature drop is predicted using the predicted value of evaporator temperature drop. However, while the residual of the evaporator temperature drop is quasi-temperature independent, the residual of the evaporator discharge temperature is not. Using a learned relationship for the evaporator temperature drop greatly simplifies the process of learning the relationship when the system degrades, which is an important feature of the present invention. And as will be seen, the relative COP can be calculated directly from the predicted evaporator temperature drop. Therefore, while either relationship can be implemented and achieve equivalent results, in the embodiments herein, the learned ETD relationship is superior to the learned T ed relation.

[0122] Figure 6 Shown from Figure 36 and 614 (and subcomponents) can be hardware-based (e.g., executed by an ASIC, FPGA, etc.), software-based (e.g., executed over a network, etc.), or a combination of both hardware and software (e.g., executed by at least one microcontroller having at least one onboard and / or separate storage / memory device storing non-transitory computer-readable instructions). Furthermore, while the functional components 600, 606, and 614 (and subcomponents) are shown as discrete blocks, any one of these blocks may be divided into several constituent blocks, or two or more of these blocks may be combined into a single block, within the scope of the disclosed embodiments. The following is a description of the operation of the various functional components 600, 606, and 614 (and subcomponents).

[0123] The data acquisition processor 600 operates to continuously acquire and store fluid temperature and power parameter values, and pre-processes and assembles these values ​​into time series of observations that can be used by the prediction processor 606. These time series, referred to herein as "observations," are Figure 6 The prediction processor 606 accepts the observation sequence and, in some embodiments, may selectively use the observations to learn the CIPP relationship and generate a normalized power parameter residual sequence R P (n), presenting the sequence to the degradation detection processor 614 for analysis. In some embodiments, the prediction processor 606 may also optionally use the observations to learn the ETD relationship and generate a normalized ETD residual sequence R E (n), presenting the sequence to the degradation detection processor 614 for analysis. In systems that include both the CIPP relationship and the ETD relationship, the prediction processor 606 can use the results from the CIPP relationship and the ETD relationship to selectively generate a sequence of relative COP values, rCOP(n), which can be presented to the degradation detection processor 614 for analysis.

[0124] The degradation detection processor 614 operates to interpret the time series of normalized residuals and the relative COP series and generate a message Msg(n) which may be issued or issued as a warning signal or message or audio-visual display, or sent as information via a news feed, as generally indicated at 616, to notify of a potential problem with the HVAC&R system. In some embodiments, the degradation detection processor 614 may also (or alternatively) interpret the normalized residual series R P (n) and RE (n) and the relative COP sequence rCOP(n) along with other observation and status information are sent as messages Msg(n) to be interpreted by systems external to the HVAC&R monitoring agent 314 for further analysis.

[0125] As discussed, the data acquisition processor 600 operates to continuously acquire and store fluid temperature and power parameter values, and based on these values ​​and optional other inputs, assembles and pre-processes them into a time series of observations, denoted by O(k), which can be used by the prediction processor 606. While there are many ways to accomplish the above, as previously mentioned, a programmable logic controller, such as the Model M251 manufactured by Schneider Electric, is well suited for this task. In the example shown, the data acquisition processor 600 includes a system temperature acquisition processor 602 that operates to continuously or periodically acquire and store fluid temperature measurements for the agent 314. The data acquisition processor 600 also includes a power parameter acquisition processor 604 that continuously or periodically acquires and stores the fluid temperature measurements generated by the power parameter meter 312 (see Figure 3 ) measurements of one or more compressor input power parameters measured by the agent 314. In some embodiments, these one or more compressor input power parameters may include active power, reactive power, apparent power, and current and voltage consumed by the compressor 104. Alternatively, as described above, where the agent 314 is used to predict the compressor input current, a measurement of the RMS current delivered to the compressor 104 by itself may be sufficient.

[0126] Temperature measurements and power parameter measurements are generally referred to herein as "observed" temperature and power parameters. In some embodiments, the data acquisition processor 600 collects multiple sets of measurements of fluid temperature and power parameters and assembles them into "observations." The temperature and power parameters in an observation are represented by a single number that represents the corresponding temperature or power parameter at a particular moment or over a period of time. The number representing the corresponding temperature or power parameter can be a single measurement, or it can be derived as a function of multiple measurements, such as an average of multiple measurements taken over the interval to be represented by the observation. Of course, other functions are possible using well-known digital signal processing techniques.

[0127] Table 1 below shows exemplary “observations” that may be provided by the data acquisition processor 600 to the prediction processor 606, as shown in Table 1. Figure 6 The O(k) in is indicated where “k” is an index indicating the kth such observation provided in the time series.

[0128]

[0129] Table 1: Example observations

[0130] In Table 1, exemplary observations include T ci Data, T ee Data and T ed Data, which includes condenser inlet temperature measurement, evaporator inlet temperature measurement and evaporator discharge temperature measurement, respectively, or a signal-processed batch of such temperature measurements, which represents the external temperature of the system at a certain point in time or over a certain time interval. These fluid temperature measurements are obtained from temperature sensors 302, 304 located at or near the evaporator and condenser inlets and a temperature sensor 306 located at or near the evaporator discharge outlet, such as Figure 3 In some embodiments, the condenser discharge temperature T cd The condenser inlet temperature T may be substituted for the fluid temperature measurement acquired and pre-processed by the system temperature processor 502. ci Alternatively, a room temperature measurement (e.g. from a thermostat) can be used as the evaporator inlet fluid temperature T ei Instead of directly measuring the evaporator entering fluid temperature in direct exchange air conditioning applications, or used as a proxy for the condenser entering fluid temperature T in heat pump applications and many refrigeration systems ci In systems where only the CIPP relationship and the power parameter residual series are implemented and used to detect system degradation, the evaporator discharge temperature is not required. In refrigeration applications (including freezers), the temperature of the interior compartment directly cooled by the evaporator can be used as a proxy for the evaporator entry temperature. Other temperature proxies that track or appropriately respond to the various entry and discharge temperatures discussed herein may also be used within the scope of the disclosed embodiments. These include measured outdoor temperature or temperature estimates obtained from weather services or forecasts.

[0131] Additionally, in some embodiments, an observation may also include power parameter data, including measurements of one or more power parameters, or functions thereof, measured by the power parameter meter 312 at the same or similar time as the temperature measurement. An example of a power parameter that may be included in an observation as power parameter data is compressor input current.

[0132] Also shown in Table 1 above are optional timestamps or tags that indicate the date and time or interval represented by the measured temperature and power parameter values ​​included in the observations. In some embodiments, including a timestamp or tag in an observation or data frame can be beneficial to embodiments, from which the date and time intended to be represented by each measurement in the observation can be inferred. This timestamp or tag is particularly useful when individual observations are stored in a database for future retrieval, or when a group or batch of observations is assembled into a data frame that can then be transmitted over a network communication link. For example, a data frame of observations can be sent over the Internet to a web service, where agent 314 (or a portion thereof) reads the data frame, processes the observations within the data frame (using time tags as needed to maintain order), and provides the results for the HVAC&R monitoring and early problem detection system 300 to take appropriate action. In other embodiments, such as in building management systems, PLCs, and dedicated controllers, observations are made continuously directly through the system without intermediate storage beyond the delay lines required to determine steady-state operation. In these systems, observations generally do not need to be associated with timestamps.

[0133] It should be understood that since the evaporator inlet temperature T ei , evaporator discharge temperature T ed and the evaporator temperature drop E are related through equation (5), so it can be obtained by replacing T in Table 1 above with the evaporator temperature drop E. ei or T ed To make an alternative and mathematically equivalent representation. For observation, given T ei or T ed For any two of E and E, the third can be immediately derived as needed using equation (5). In the following, the representation of the observations provided in Table 1 is to be considered as exemplary only.

[0134] As described above, the time series of observations, denoted as O(k) and comprising the measurements and optional time stamps of Table 1 above, are forwarded one at a time or in batches of data frames from the data acquisition processor 600 to the prediction processor 606. According to the disclosed embodiment, the prediction processor 606 is operable to derive or learn the CIPP relationship and the ETD relationship from the observations O(k) provided by the data acquisition processor 600 and use these relationships to create normalized residual sequences of both the power parameter and the evaporator temperature drop, and further create relative COP sequences, each of which can be used by the degradation detection processor 614 to monitor performance degradation of the system.

[0135] 6A to 6I Shown from Figure 6 Additional details of an exemplary embodiment of the HVAC&R monitoring agent 314 are provided below.

[0136] refer to Figure 6A , showing the Figure 6 6, which illustrates additional details and information flow therein. In some embodiments, the prediction processor 606 includes a VCC state generator 608 that receives an observation sequence O(k) from the data acquisition processor 600 and enhances the sequence with state information derived from the sequence O(k) in a manner described later. The output of the VCC state generator is the enhanced sequence O(k). a (n), where the new index "n" indicates that the VCC state generator 608 has delayed the timing of the original observation sequence O(k) to accomplish the enhancement.

[0137] The state enhancement observation sequence O provided by the VCC state generator 608 a (n) is used as input to the CIPP processor 610 (or other input power parameter relationship processor), the ETD processor 612 (or other temperature parameter relationship processor), and the rCOP processor 613. The CIPP processor 610 uses the enhanced observation sequence O a (n) to learn the CIPP relationship discussed above and output the power parameter prediction sequence described by equation (3) (as will be described later) and the normalized residual sequence R P (n). Power parameter value sequence Serves as input to the relative COP processor 613. The normalized residual R P The sequence of (n) is provided directly to the degradation detection processor 614 as the output of the prediction processor 606.

[0138] Similarly, the ETD processor 612 uses the enhanced observation sequence O a (n) to learn the ETD relationship discussed above and output the evaporator temperature drop value prediction sequence (as will be described later) and the normalized residual R described by equation (8) E (n)Sequence. Predicted evaporator temperature value The sequence serves as the input to the relative COP processor (or rCOP processor) 613. The evaporator temperature drops R E The normalized residual sequence of (n) is directly provided to the degradation detection processor 614 as the output of the prediction processor 606.

[0139] The relative COP processor 613 receives and processes the state enhancement observations from the VCC state generator 608. a (n), power parameter prediction sequence from CIPP processor 610 and the evaporator temperature drop prediction sequence from the ETD processor 612 And optionally calculate the corresponding to Oa (n) is then provided to the degradation detection processor 614.

[0140] As described above, the VCC state generator 608 derives certain timing information about the state of the VCC cycles in the system and utilizes this information to enhance observations therefrom to inform and control downstream processing, such as via the CIPP relation processor 610, the ETD relation processor 612, and the rCOP processor 613. a The enhanced observation of (n) is delayed relative to the original observation due to the enhancement.

[0141] Figure 6B is a diagram illustrating an exemplary information flow through the VCC state generator 608. In this diagram, O(k) represents the kth original observation including the information from Table 1 above received by the VCC state generator 608 from the data acquisition processor 600. In order to derive timing information from the observation sequence O(k) and enhance each observation with appropriate state information, in this embodiment, it is necessary to delay the observation sequence by a specified number of sampling periods N db +1, where N db is an integer machine constant called the compressor state debounce count (discussed later in this document). The delay is represented here by the delay function 620 to indicate that N db +1 sample delay in the time series. The resulting delayed observation is represented as O(n), where the new index "n" is related to the original index "k" by:

[0142] n=kN db -1 (10)

[0143] In some embodiments, the delayed observations, together with the three new time series of information derived from the sequence O(k), become the three new time series generated by O(k). a (n) is part of the “augmented observation” indicated by the three new information sequences that augment the original but delayed observation: (i) S c (n), representing the on / off state of a simple single-compressor system, or the integer-encoded compressor state of a multi-compressor system, (ii) the power state variable S p (n), which indicates the delayed observation O(n) for learning and predicting the associated compressor state S c (i) The power parameter value and relative COP applicability of the compressor considered as “on” in (ii) and (iii) the evaporator temperature drop state variable S e (n), which indicates the suitability of delayed observation O(n) for learning and predicting the evaporator temperature drop and relative COP of the compressor considered "on" in the relevant compressor state. p (n) and S e(n) indicates observation O when true a (n) represent the operation in the power parameter prediction interval 404 and the evaporator temperature drop prediction interval 424 of the compressor in the ON state. When these state variables are generated by the VCC state generator 608, the generated sequence is appropriately delayed so as to be consistent with the delayed observation sequence O in the manner described later. a (n) Alignment. Although there are many ways to implement these state variables, the embodiment herein may be considered preferred because it is easily scalable to multi-compressor systems.

[0144] In order to enhance the observation of a (n) generates a sequence S c (n), the compressor state debounce function 622 implicitly uses the machine constant N db To determine and encode the on / off status of each compressor in the system. The output of the debounce function 622 is the sequence S c (kN db ), then delayed by one sample in the delay function 624 and applied directly to the enhanced observation O a (n) as S c (n). Sequence S c (kN db ) is also used as a parameter input N pl The input of the power stability function 626, and the input with parameter N el The input of the ETD stability function 628 generates the enhanced observation O a (n) sequence S p (n) and S e (n). It should be noted that in this example, for example Figure 3 For a single compressor system, "observation" includes a single power parameter. Those skilled in the art will understand that the principles and teachings of this document are also applicable to a system that obtains multiple power parameters for multiple compressors.

[0145] Figure 6C An exemplary flow chart 630 is shown that illustrates exemplary debounce logic that may be used with the debounce function block 622 in some embodiments. The debounce logic shown in flow chart 630 is applicable to embodiments having a single compressor or multiple compressors. The purpose of the debounce function is to defer declaration of a compressor state change until the compressor state is determined by the debounce machine constant N. db The delay in declaring the compressor state is to ensure that the debounced state of the sample represents the "true" state of the system at the time of observation, where N db Typical values ​​are in the range of three to five samples. The debounce function provides this internal value as output S c(k-Ndb).

[0146] The debounce logic maintains two internal state variables. The internal compressor state variable Sdb holds the current debounce state of the compressor, as declared by the debounce logic. An internal counter DBC is used to facilitate the delay of transitions in declared compressor states. The internal compressor state variable S db Typically initialized to a value of 0, indicating that all compressors in the system are declared off (including the single compressor in the simple HVAC system 100 described above), and DBC is initialized to the value N described above. db .

[0147] Entering the flowchart 630 begins by receiving the kth observation O(k) at box 631, where the power parameter of the kth observation is extracted from the observation. One power parameter is extracted for each compressor, preferably, each power parameter is uniquely responsive to the current flowing to the individual compressor, meaning that the power parameter representing one compressor does not include the other compressor currents, or is a function of the other compressor currents. This is the case when a single current transformer is used to measure the compressor current of an individual compressor. At box 632, in one embodiment, the "instantaneous state" of each compressor is determined by comparing the power parameter value with a threshold value unique to each compressor or a system-wide threshold value to determine whether the individual compressor is in the on state (as indicated by the power parameter value of the compressor exceeding the threshold value), or if the power parameter value of the compressor is less than or equal to the threshold value, the "instantaneous state" of each compressor is determined. In one embodiment, the instantaneous on / off state of each compressor is encoded into a number, such as a binary number. For an M compressor system, an example of such an encoded number is shown in Table 2 below:

[0148]

[0149] Table 2: Compressor on / off status encoding for M compressor systems as M-bit binary representation

[0150] The result of this encoding is represented as S t (k) which can be observed to change as the state of each compressor changes. For example, in a single compressor system, this number will range between "0" (compressor off) and "1" (compressor on). In an M-compressor system, the number will range between 0 (all compressors off) and 2M (all compressors on). The internal representation of the debounced compressor state uses the encoding format of Table 2.

[0151] Once the instantaneous state of each compressor is determined and encoded as S t (k), then at block 633, the debounce logic then sets S t (k) and the stored state Sdb If the observed state S t (k) and the declared status S db If the compressors are different, it is possible that one or more compressors have changed state, but as mentioned above, it is desirable to defer declaring the change of state until N consecutive compressors in a row have changed state. db Thus, if the instantaneous compressor state has changed at block 633, then in this example, the debounce counter is loaded with a value N at block 634. db The flow chart then proceeds to block 635 where the debounce counter DBC is decremented by one or some other predefined decrement. On the other hand, if at block 633 the transient state S t (k) and the previously stored state S db If yes, control passes to block 635 where the debounce counter DBC is decremented.

[0152] At block 636, a determination is made as to whether the debounce counter DBC is less than zero. If so, control passes to block 637, where the internal state variable S db Set to S t (k), and the debounce counter DBC is set to a zero value in anticipation of the next observation. Control then passes to block 638. If at block 636, it is determined that the debounce counter DBC is greater than or equal to zero, control passes directly to block 638. At block 638, the control logic returns to S db The current value of the compressor state S c (kN db ).

[0153] The output of the compressor state debounce function 622 is then declared to represent the original observation O(kN db ) is the “true” or “derived” state of the compressor of the kth observation. By noting that if one or more of the compressor states changes in the kth observation, the output of the compressor state debounce function 622 is subsequently changed to match that observation N db The resulting output of the compressor state debounce function 622 (labeled S c (kN db ) is delayed by one sample in delay block 624, thereby c (n) is incorporated into the enhanced observation O a (n) The compressor state sequence S is generated before c (n).

[0154] Reference again Figure 6B , providing a power stability function 626 to calculate Figure 4AA separate ETD stability function 628 is provided to calculate the power parameter leading blanking, power parameter lagging blanking and power parameter prediction interval. Figure 4B The evaporator temperature drop leading blanking, evaporator temperature drop lagging blanking and evaporator temperature drop prediction interval are first considered. Figure 3 The need and function of the power stability function 626 and the ETD stability function 628 can be best understood with respect to the simple single compressor system 100 shown. When the compressor is off for an extended period in such a system 100, the pressure of the refrigerant throughout the system tends to equalize so that the refrigerant is always in a pure vapor state and the temperatures of the evaporator and condenser coils tend to those of the ambient conditions in which they operate according to the laws of thermodynamics. Turning the compressor on causes the system to migrate toward another quasi-steady state in operation, wherein the refrigerant state is driven by the mechanical action of the compressor to the liquid and vapor states described in the discussion of the VCC cycle above, and the corresponding power consumed in moving the refrigerant through the system stabilizes as the level of liquid refrigerant in the condenser also stabilizes. The time required to achieve this was previously referred to as Figure 4A "Power parameter advance blanking interval" 402.

[0155] In practice, the system in question is typically implemented as a sampled data system, where the sampling period may be long relative to the time required to shut down the compressor. For example, the time required to completely remove power from the compressor and thus disrupt the VCC cycle may be on the order of 10 to 50 milliseconds, while in systems that have been operating for some time, the sampling period of a sampled data system may be even longer, on the order of seconds or even minutes in some embodiments. Furthermore, as described above, the value of the power parameter provided by the data acquisition processor 600 typically represents an average of the power parameter values ​​over the sampling period. In such a system, the last sampled power parameter value of the system in a compressor cycle where the compressor changes state at some time within a sampling interval may vary between nearly zero, if the compressor is shut down at the very beginning of the interval, and full value, if the compressor is shut down at the very end of the interval. The average value of the power parameter obtained over this last sampling interval may not represent the true value of the power parameter within that interval, which may result in learning inappropriate values ​​or generating inappropriate normalized residuals. Therefore, in particular in some embodiments, when constructing the observed power parameter component, where multiple power parameter measurements are averaged over a period of several seconds, it is important to ignore the last sample of the compressor cycle, i.e., the sample before the compressor is first detected in the off state by the compressor state debounce function 622 in the case of a simple single compressor system, or more generally when one or more compressors change state in a multi-compressor system. This was previously referred to as Figure 4A The power parameter in the hysteresis blanking interval 406.

[0156] The same problem exists for evaporator temperature drop. In order for evaporator temperature drop to be stable, not only must the stability conditions for the power parameter be met, but the rate of heat transfer from the evaporator coil to the evaporator ambient fluid must also be stable. In fact, in some cases, the evaporator temperature drop lead interval for evaporator temperature drop can be significantly longer than the time required for the power parameter to stabilize, while the requirements for the evaporator temperature drop hysteresis interval 426 are the same as the requirements for the power parameter hysteresis interval 406. Therefore, for the purpose of learning the relevant attribute (power parameter or evaporator temperature drop), some state logic is required to determine when the observation is valid and stable. The form of this stability logic can be the same for both attributes (power parameter or evaporator temperature drop), differing only in the lead blanking time.

[0157] Stability logic indicating the suitability of observations for power parameter or evaporator temperature drop learning and prediction is implemented in the VCC state generator 608 via the power parameter stability function 626 and the ETD stability function 628. These stability functions have a state sequence S p (n) and S e (n) as output. State sequence S p (n) indicates that when it is true, the enhanced observation is located in the power parameter prediction interval 404 of the VCC process, indicating that the observation is suitable for learning and predicting the power parameters of the compressor in the ON state, and is false when the observation is not. Similarly, the state sequence S e (n) indicates that when it takes a Boolean value of true, the enhanced observation is within the evaporator temperature drop prediction interval 424 of the VCC process, indicating that if at least one compressor is on, the observation is considered suitable for learning and predicting the evaporator temperature drop, and if not, the observation is considered suitable for learning and predicting the evaporator temperature drop.

[0158] Figure 6D An exemplary flow chart 640 is shown that illustrates exemplary stability logic that may be used with the power parameter stability function 626 and the ETD stability function 628 according to one embodiment. The flow chart 640 represents one implementation of the stability function and may be used with both the power parameter stability function 626 and the ETD stability function 628, differing only in that the corresponding leading blanking intervals are represented as N xl Parameters, where N xl In the case of the power parameter stability function, the value Npl is taken, and in the case of the evaporator temperature drop stability function, a separate, possibly larger value N el In some embodiments, N pl and N el is stored as a machine constant.

[0159] Return Reference Figure 6B The inputs to the power parameter stability function 626 and the ETD stability function 628 are the debounced compressor state S c (kN db )(provided by the compressor state debounce function 622). Figure 6D As shown in the flowchart 640, S c (kN db ) is calculated before executing the stability function on the kth observation. Internally, each instance of the stability function maintains an internal counter SLCnt, an internal state variable S cl , which is intended to represent the delayed state of the compressor from the previous observation, i.e. S c (kN db -1). In some instances, the counter SLCnt is initialized to a value of N xl , where N xl are parameters that depend on the individual functions implemented, and the internal state variable S cl Initialized to indicate all compressors are off to promote repeatable behavior on system startup.

[0160] When flow chart 640 is entered at block 641, it is determined at block 642 whether the result of executing compressor state debounce function 622 on the kth observation has changed from the result generated for the previous observation, as described above by S cl If yes, control passes to block 643, which uses the value N xl Load counter SLCnt, where N xl is the expected lead blanking interval for the particular stability function (i.e., the lead blanking intervals 402 and 422 for the power parameter and evaporator temperature drop, respectively), expressed as an integer number of sampling periods. Control then passes to block 644, which decrements the counter SLCnt by one or another predefined decrement, which may result in a negative number in SLCnt. If it is determined in block 642 that the compressor state has not changed since the previous observation, control passes directly to the counter decrement block 644.

[0161] From block 644, control passes to block 645, which checks whether the counter SLCnt is less than zero. If so, control passes to process block 646, where the counter SLCnt is loaded with the integer value "0" for the next observation, and the temporary internal state variable S is set to zero. x is set to true, indicating that the observation in question is within the relevant prediction interval (i.e., the appropriate power parameter prediction interval 404 or the evaporator temperature drop prediction interval 424). Control then passes to block 647 for further operation. If the counter SLCnt is greater than or equal to zero in block 645, then control instead passes to block 647, where S xis set to logical false, indicating that the observation of the utterance is not within the relevant prediction window. Control is then passed again to block 648 for further operations.

[0162] In block 648, the internal state variable S c l is then set to the value S c (kN db ) in anticipation of use in the next observation. In block 649, the S determined as described above is x The value of the power leading stability function 626 is used as the output S p (n) is returned and in the case of the ETD lead stability function 628 is output as S e (n)Return.

[0163] In some embodiments where only power parameters are used to detect system degradation, there is no need to implement the ETD stability state variable S e (n) Required logic. In other embodiments where both power parameters and ETD are used to detect system degradation, N pl and N el A single stability function is achieved by maximizing , where the resulting value indicates that the system is stable.

[0164] In other systems where the power parameter is not available and only the evaporator temperature drop is used to detect system degradation, the evaporator temperature drop itself can be used as a proxy for the power parameter input P(k) in the compressor state debounce function 622. In this case, the threshold value for the evaporator temperature drop will be selected to be applied in block 632 ( Figure 6C ), where an evaporator temperature drop greater than the threshold provides an indication that the compressor may be "on," and a drop less than the threshold provides an indication that the compressor may be "off." In this case, only the evaporator temperature drop stability logic will be implemented.

[0165] The enhanced observations obtained from the foregoing are shown in Table 3 below. a (n) It can be seen that Table 3 is similar to Table 1 except that the system status value is additionally included.

[0166]

[0167] Table 3: Enhanced Observation

[0168] In some embodiments, the VCC status generator 608 uses the enhanced observation discussed above. a (n) Provided to CIPP processor 610, ETD processor 612 and rCOP processor 613, as described above Figure 6Those processors 610, 612 and 613 can then use enhanced viewing. a (n) to use internal logic to determine if the enhanced observation is suitable for further processing as needed. The means for making these decisions will be discussed as part of the following discussion. Alternatively, the VCC state generator 608 can select only observations that are considered stable with respect to the power parameters, i.e., S p The state variable has been declared true and at least one compressor such as S c The VCC state generator 608 may also select only those observations for which the VCC state generator 608 has declared the HVAC&R system 100 stable with respect to the evaporator temperature drop, i.e., the evaporator temperature drop state variable S. e has been set to true and as S c The observation of at least one compressor on indicated by the state variable is provided for processing by the ETD processor 612. In other embodiments, using the state information provided by the VCC state generator 608, other components in the prediction processor 606 can determine which enhancement observations are relevant to their respective functions as needed.

[0169] Enhanced Observation a (n) Allows the CIPP processor 610 to learn the relationship between the inlet temperatures and the compressor input power parameter values ​​associated with those temperatures (i.e., the CIPP relationship), selectively generate a prediction of the power parameter value representing the HVAC&R system in the new maintenance condition for a given observation, and, if appropriate, generate a normalized power parameter residual value for that observation. Similarly, the ETD processor 612 uses the enhanced observation. a (n) to learn the relationship between the incoming temperatures and the evaporator temperature drops associated with those temperatures (i.e., the ETD relationship), thereby selectively generating a prediction of the evaporator temperature drop value representing the new maintenance condition of the HVAC&R system for a given observation, and, when appropriate, generating a normalized evaporator temperature drop residual value for that observation. The CIPP processor 610 and the ETD processor 612 are essentially identical in functionality, differing only in their parameter focus, and both follow the basic signal processing described above for generating the normalized residual. Hereinafter, the term "discourse prediction" will be used to describe the power parameter or evaporator temperature drop, as appropriate.

[0170] Previously about Figure 5A and 5BReference is made to learning CIPP relationships and ETD relationships with reference to learned CIPP relationships block 500 and learned ETD relationships block 506. The purpose of learned CIPP relationships block 500 is to learn the relationship between the evaporator inlet temperature, the condenser inlet temperature, and the compressor power parameter and provide a prediction of the power parameter value representing the operation of the system under the new maintenance condition. The purpose of learned ETD relationships 506 is to learn similar relationships between the evaporator inlet temperature, the condenser inlet temperature, and the evaporator temperature drop of the HVAC system representing the system under the new maintenance condition. CIPP processor 610 and ETD processor 612 implement the purpose of learned ETD relationships block 506 for CIPP relationships block 500. Now, the following is a more detailed explanation of how CIPP relationships and ETD relationships can be learned in some embodiments.

[0171] As background, existing solutions for learned CIPP relationships and learned ETD relationships use so-called lumped regression approaches, where a large set of observations is acquired, assuming the system is operating under new maintenance conditions, collected over a long period of time and intended to represent the entire operating "range" of the equipment in terms of temperature. The large dataset is intended to be acquired when the system is in "new maintenance" conditions and assembled into a training dataset, a test dataset, and, in some cases, a validation set. Machine learning in the form of a linear regression algorithm is used to create a model of the system from the entire training set, where those observations of the training set meet the criteria of the VCC state generator 608, and logic is applied to select those observations that represent "stable" operation with respect to utterance prediction.

[0172] The ensemble linear regression algorithm is well established and understood in the discipline of machine learning—utterance prediction is “predicted” using a linear combination of functions of the explanatory variables; in this application, T ei and T ci , so-called parametric models have parameter coefficients that are chosen (learned) to minimize a linear combination of cost functions, usually the mean squared error between observed and predicted utterances on the training set. Once the parameter coefficients are determined, T ei and T ciThe resulting parametric model associated with utterance predictions represents the operation of the system under all conditions. To make predictions about observations based on the parametric model, the evaporator entry temperature and the condenser entry temperature are substituted into the parametric model with the learned coefficients and the calculated results. A test dataset is then applied to the parametric model to confirm that the model can indeed represent the characteristics of the actual system, and not just the training set for which the parameter coefficients were determined. Often, it is necessary to "tune" the algorithm parameter coefficients to ensure that the parametric model represents not only the training set, but also the test set with sufficient joint accuracy to be actually used in an application. In some cases, once the parametric model is appropriately tuned, it is applied to a third validation set with observations that the parametric model has never "seen", with the goal of establishing a level of confidence that the tuned parametric model represents the underlying mechanisms of the process to be modeled, and not just the details of the training and test sets.

[0173] As described above, existing solutions have practical considerations that limit the usefulness and marketability of HVAC&R degradation detection systems. One limitation of existing solutions is the large datasets required, which often take a long time to assemble, particularly where the training is customized for an individual HVAC&R system. Accurate predictions of expected power parameters or evaporator temperature drop values ​​are delayed until training is complete. For example, for an air conditioning system operating in a mild climate, data for the entire cooling season may be required to ensure that all expected external conditions are observed, e.g., because in most parts of the United States, average and peak outdoor temperatures in May are typically much colder than average and peak outdoor temperatures in August. The availability of the necessary degradation detection system is delayed until the dataset is available.

[0174] Another limitation of existing solutions is that the HVAC&R system needs to be maintained in a "fresh maintenance" condition throughout the period during which the various datasets are acquired to build an accurate parametric model. This is impractical when the training period takes weeks or months to complete due to the large training datasets required and fully anticipating degradation due to condenser and evaporator fouling (which can be caused by dirty filters).

[0175] Another practical limitation is that collecting and storing large numbers of observations for training data may not be feasible except for cloud-based solutions with large storage capacities, as solutions residing near the HVAC&R system typically have much smaller storage capacities.

[0176] Another practical limitation of existing technical solutions is the manual "tuning" of the model's parameter coefficients that is often required to establish accurate predictions, and the need to verify the validity of these predictions against test and validation sets.

[0177] Yet another practical limitation of prior art solutions is the difficulty in knowing whether a given observation that the learned parametric model is to predict lies within the operating range for which there are sufficient observations in the training set. Lumped regression algorithms are prone to producing inaccurate predictions when operating outside the range of the large number of explanatory variables for which they were created. Accepting the results of predictions from a lumped linear regression model without additional information about the distribution of the training set may result in inaccurate predictions without knowing it occurred, causing the system to generate false positives (i.e., declaring degradation when it is not present) or false negatives (i.e., indicating that the system is in a relatively good state when it is not present). It would be better if the agent simply ignored the observation rather than making a poor prediction or a series of poor predictions that result in a false positive or false negative.

[0178] Thus, one aspect of the embodiments herein is a novel learning method for signal processing and signal flow by which the agent 314 learns the corresponding relationships (CIPP or ETD) described in functional blocks 500 and 506, respectively (see Figure 5A and 5B ), which addresses the limitations described above. While particularly well-suited to the task of learning CIPP and ETD relationships, reliably predicting power parameters and evaporator temperature drop values ​​when a physical HVAC&R system is experiencing degradation, and doing so without the need to a priori collect large datasets, it should be clear that the embodiments disclosed herein are applicable to many other types of applications.

[0179] With the above background, we now refer to Figure 6E , which shows an exemplary relationship learner process 650 that may be operated or otherwise employed by the CIPP processor 610 and the ETD processor 612 to use the enhanced observations provided by the VCC state generator 608. a (n) Learn the CIPP relationship and the ETD relationship, respectively, and use these relationships to predict parameter values. In some embodiments, the relationship learner 650 uses a machine learning process to learn the CIPP relationship and the ETD relationship. The machine learning used by the relationship learner 650 adopts a novel approach that specifically exploits certain characteristics of the physics and design of HVAC&R equipment, but is more generally applicable to a wider range of applications.

[0180] In the embodiment of this article, two relation learners 650 are assigned to Figure 3 A relationship learner 650 assigned to learn the CIPP relationship and a second relationship learner 650 assigned to learn the ETD relationship. These relationship learners 650 are respectively Figure 5A CIPP Relationships in Learning 500 and Figure 5BThe internal mechanism behind the learned ETD relation 506 in

[15] . In a multi-compressor system, for each possible state of Sc(n) where the compressor is on, a separate pair of relation learners is assigned to each compressor. However, currently, according to Figure 3 The operation of the relational learner is best understood for a simple single compressor system, where extension to multiple compressor systems is postponed until the principles of the relational learner are understood as follows.

[0181] exist Figure 6E In the example of FIG, the relationship learner 650 uses several modules to learn the CIPP and ETD relationship and make predictions, including a relationship builder 652, a temperature map 654, a neighborhood extractor 656, and a parameterized predictor 658. In general, the temperature map 654 relates incoming temperatures to the parameters of interest (as received from the VCC state generator 608) that may indicate that the HVAC&R system is in a new maintenance condition related to these temperatures, while the relationship builder 652 operates to compile and maintain the temperature map 654. The neighborhood extractor 656 defines the observed temperature tuple (T ei (n),T ci (n)) around a range or “neighborhood” of acceptable temperature points, and the parameterized predictor 658 operates to make an utterance prediction for that temperature tuple based on the values ​​in the temperature map 654 predictions.

[0182] A benefit of using the temperature map 654 is that it allows the neighborhood extractor 656 to detect when a temperature tuple of a steady-state observation in the temperature map 654 is outside the range that can be confidently predicted. This means that the agent 314 can choose not to make a prediction via the parameterized predictor 658, rather than risk predicting an incorrect value for the corresponding prediction of the utterance. This arrangement can be used to greatly reduce the chance of creating a "false positive" condition, declaring the system to be degraded when no problem exists, or a "false negative" condition, declaring the system to be in good condition when it is actually degraded.

[0183] In some embodiments, the parameterized predictor 658 makes predictions using a parameterized model of a predetermined form, wherein, in contrast to the lumped regression model described above, each time a prediction is to be made, the parameter coefficients of the model are derived from the data in the temperature map(s). Figure 6G Discusses parameters and examples of parameterized functions.

[0184] In some embodiments, the relationship builder 652 uses a 2-stage guided learning strategy combined with a reference degradation estimator function to, in some cases, modify the predictions of steady-state observations before using the modified observations to populate the temperature map 654. Figure 7 The details of these two functions are described in more detail.

[0185] for Figure 3 For a simple single compressor system, in some embodiments, the relationship builder 652 uses only those enhanced observations provided by the VCC state generator 608. a (n) to construct a temperature map 654, for which the compressor is connected via the state variable S c (n) is declared to be in the "on" state and is only determined when observations indicate that the vapor compression cycle is in a steady state relative to the predictions discussed above, in the case of the power parameter, via the state variable S p (n) and in the case of a drop in evaporator temperature via S e (n). Hereinafter, for the purposes of the relation builder 652 and neighborhood extractor 656, an observation that meets the above criteria is called a steady-state observation, and the associated relation learner is declared to be "active" for that observation. Observations that do not meet these criteria for utterance prediction are ignored by the relation builder 652. The concept of active relation learners will be subsequently extended to multi-compressor systems.

[0186] Each steady-state observation O provided by the VCC state generator 608 a (n) includes temperature tuple (T ei (n),T ci (n)) and the corresponding observation of the “utterance prediction” X, or the power parameter P, or the temperature from which the evaporator temperature drop E can be calculated. Each temperature tuple (T ei ,T ci ) is used as a two-dimensional index in the temperature map 654 when quantized. For each indexed temperature tuple, the agent 314 "learns" by updating the summary data for the "cell" corresponding to the temperature tuple from the predicted sequence of utterance values. The details of the cell contents will be described later. When an observation that meets certain requirements is encountered, the relationship builder 652 updates the summary data for a given cell in this manner until enough observations have been applied to be considered to represent an utterance prediction of the machine under the new maintenance condition under that temperature tuple, as described later in this document, after which the relationship builder 652 stops updating the summary data for that cell, and the summary data for the cell can be used to predict a power parameter value representing the system under the new maintenance condition. In some cases, once the requisite number of observations have been made for that cell, power parameter and EDT predictions can be derived directly from the summary data for a single cell indexed by a tuple of steady-state observations. In other cases, a parameterized prediction function can derive power parameter predictions for a tuple of steady-state observations by first using a neighborhood extractor 656 and a parameterized predictor 658 to construct a local parameter model with parameter coefficients derived using summary data from nearby tuples, as will be described later.

[0187] Using the above approach, the relation builder 652 can quickly build useful relations, the neighborhood extractor 656 can be used to determine whether a prediction should be made for a given tuple (i.e., within the "neighborhood"), and the parameterized predictor 658 can begin making predictions for utterance predictions almost immediately. This allows the CIPP processor 610 and the ETD processor 612 (via the relation learner 650) to begin making corresponding parameter predictions and generating useful normalized parameter residual sequences shortly after the HVAC&R system is commissioned (in some cases within the same day), provided that the system is operating and in its new maintenance condition, and allows the correlated COP processor 613 (using the parameter predictions) to create the correlated COP sequence rCOP(n).

[0188] Using the temperature map 654 described herein, the neighborhood extractor 656 can evaluate whether the prediction of the power parameter corresponding to a given temperature tuple is likely to represent the behavior of the HVAC&R system under the new maintenance condition and decide whether to issue a prediction. The ability to evaluate the reliability of the prediction significantly reduces the likelihood of the agent providing false positives and false negatives. Additionally, because the normalized residuals of both the CIPP relationship and the ETD relationship can be assumed to be quasi-temperature independent (as described above and further discussed herein), the agent 314 can continue to learn the behavior of the HVAC&R system under the new maintenance condition as the HVAC&R system degrades, thereby compensating for the degradation so that the prediction better represents the system under the new maintenance condition.

[0189] Continuous learning of relationships by the relationship builder 652 can be achieved by updating the temperature map 654 as additional temperature and measurement predictions of speech data become available in the form of observations. In some embodiments, the temperature map 654 is updated in batches, whereby a set of observations is assembled into one or more data frames of steady-state observations (i.e., observation sets) and presented to the prediction processor 606 of the agent 314 by the data acquisition processor 600 as a batch of time-ordered observations. Multiple batches of observations can be acquired on an hourly, daily, or other time basis and presented to the agent as a time series using the above-mentioned timestamp (TS) or another means to sort the time series. In some embodiments, observations can also be provided on an individual observation basis, one at a time as they are received.

[0190] In some embodiments, by using an evaporator within a specific temperature range of interest, the temperature T ei and the condenser inlet temperature T ci To construct the temperature map 654. Assume that for each T ei and T ci, quantized to 0.1°C (other quantization levels can of course be used) and a temperature range of 10 to 40°C, the resulting temperature map will be a 300×300 table (with 90,000 cells). A partial example of an exemplary temperature map 654 is shown in Table 4 below, where the cells of the map contain, for each temperature tuple (T ei ,T ci ) Summary values ​​of the compressor input power parameter observed. Although the table is shown as mostly populated, in general, only those T ei and T ci The cells with the value of will contain the summary value.

[0191]

[0192] Table 4: Example Temperature Map

[0193] As mentioned above, each cell in the temperature map (e.g., C00, C01, C02, etc.) contains a temperature tuple (T ei ,T ci ) corresponding to the observations in the dataset. These summary values ​​(also called summary statistics or sample statistics in some cases) provide summary information about the steady-state observations represented by the cell. For example, summary values ​​can provide information about the data in the dataset, such as the sum, mean, median, average, variance, deviation, distribution, etc.

[0194] As previously mentioned, the steady-state observed power parameter values ​​are calculated from measurements made with power or current meters designed specifically for this purpose, and the evaporator temperature drop is calculated from readings from the temperature sensor. However, real-world measurements may still be noisy due to operational and / or environmental variability. Therefore, the temperature map 654 inherently incorporates real-world conditions, whereby the values ​​used to update the cells may be corrupted by noise. These real-world conditions can be described as a stationary zero-mean additive random noise process NoiSe(0,σ x 2 ), where σ x 2 is the variance, which may depend on the noise process of the utterance prediction. Then, each measurement of the steady-state prediction X of the utterance can be expressed as shown in equation (10):

[0195] X=X o (T ei (n),T ci (n))+Noise(0,σ x 2 ) (10)

[0196] Among them, X o (T ei (n),Tci (n)) is the predicted potential value of the observed utterance.

[0197] In one embodiment, the relationship builder 652 applies one of two functions of parameter values ​​from steady-state observations to populate and update the summary values ​​of the cells in the temperature map of Table 4. In the following, the term f will be used x (X,n) describes the application of an appropriate function to the nth steady-state augmentation observation O used to update a particular unit. a The measurement of the value of the utterance (n) predicts the result X. One of the functions applied is the identity function, where the value predicted by the measurement of the utterance itself is the result of the function. In this case, f x (X,n) is given by:

[0198] f X (X,n)=X (11)

[0199] When compensating for system degradation during the learning process, the agent can apply a second time-varying compensation function based on previously learned characteristics of the system, the details of which will be described later. To reduce measurement noise present in real systems, the agent constructs and maintains summary data for each cell, which can be stored in the cell and used to calculate sample statistics for predicting utterances corresponding to indexed temperature tuples. In some embodiments, the summary data for each cell includes the following summary values:

[0200] The sum of the observed values,

[0201] The observed sum of squares,

[0202] Where N is the total number of observations stored in the sum, which is also stored as an element of the summary data in the cell. In other words, each time the relationship builder 652 updates the summary data in a cell, it does the following:

[0203] a. Apply the appropriate function to the value in steady-state operation, expressed by equation (11), to obtain the value f x (X,n);

[0204] b. Set the value f x (X,n) is added to the sum of the observed values, as described in equation (12);

[0205] c. Calculate f x The square of (X,n) gives the value

[0206] d. Value is added to the observed sum of squares, described by equation (13); and

[0207] e. Increment the value of N associated with the cell to reflect the update.

[0208] The parameterized predictor 658 can use these summary values ​​to calculate predicted values ​​for the utterance value corresponding to the unit as needed, and also to compensate subsequent observations for utterance predictions, both in a manner discussed subsequently.

[0209] Additionally, for each cell of the temperature graph, in some embodiments, the relationship builder 652 maintains two pieces of metadata: (1) an indication of whether sufficient observations have been made at the particular temperature tuple represented by the cell such that the summary statistic represented by the cell can be designated as valid for prediction purposes; and (2) an indication of whether one or more observations used to form the summary statistic for the cell have been modified to compensate for system degradation.

[0210] The first metadata may be stored as a Boolean variable, such as "observations," where the variable is set to true to indicate that enough observations have been made and false otherwise. The entries in the temperature map are populated as quickly as possible with enough observations so that the average of the stored observations can be used to reliably predict the power parameter, while stopping populating the entries in the map when the number of observations is sufficient so that, under normal noise conditions, additional observations are unlikely to significantly change the sample average of the cell. Thus, in some embodiments, a temperature tuple (T ) is added to the cell when a minimum of four observations have been made and the relationship builder 652 stops adding information to the cell at that point. ei ,T ci ) is defined as being observed, and the "observed" metadata variable is set to true. This approach has the effect of limiting the data stored in the cell to that which most likely reflects the new maintenance condition of the system, and serves as an aid in allowing the parameterized predictor 658 to quickly begin predicting system conditions.

[0211] The "observation" metadata variable is optional in the sense that it is derived from the already stored summary data value N. However, maintaining this variable so that it is only "set" once can reduce processing time and aid in understanding the principles and teachings herein.

[0212] The second metadata may also be stored as a Boolean variable, such as "compensation", where true indicates that the time-varying compensation function has been applied to at least one of the steady-state observations used in forming the summary data for the unit, and false indicates that the compensation function has not been used to compensate for system degradation in the steady-state observations used in forming the summary for the unit. Figure 8 The discussion provides further details.

[0213] Thus, in some embodiments, each cell in the temperature map includes the following exemplary variables and their corresponding data: "SV" {summary data}, "Compensation" {true / false}, and optionally "Observation" {true / false}.

[0214] An estimate of the average predicted utterance value of an entry in a cell of the temperature map can be calculated from the summary quantity using equation (14):

[0215]

[0216] in, is the average prediction of the utterance value, and the variance of the cumulative utterance value prediction σ 2 An estimate of can be calculated using the equation:

[0217]

[0218] When the methods subsequently taught herein are applied, equation (14) can be used to predict the utterance prediction value that most likely represents the HVAC&R system in the new maintenance condition at the corresponding steady-state observed temperature tuple value. In some embodiments, equation (15) can be used as an indicator of the "fidelity" of the prediction, where a low variance indicates that the values ​​forming the sum are all nearly identical, while a high variance indicates otherwise.

[0219] We now discuss how predictions for utterance prediction can be extracted from a temperature map (e.g., temperature map 654) that is unique to the utterance prediction used by the CIPP processor 610 and the ETD processor 612, respectively, to compute the corresponding sequence presented to the rCOP processor 613. and present to Figure 6 The corresponding residual sequence R of the degradation detection processor 614 in P (n) and R E (n). Reference Figure 6F and 6G The method by which predictions for utterance predictions can be extracted from the temperature map 654 is best understood.

[0220] Figure 6F A flowchart 660 is shown illustrating an exemplary process that may be used by or in conjunction with the neighborhood extractor 656 to determine whether to make a prediction and to provide a table of temperature tuples for cells in the temperature map 654 pointing to utterance predictions sufficient to construct a local parameter model via the parameterized predictor 658 to predict utterance parameters when appropriate. Figure 6G A flow chart 670 is shown illustrating an exemplary process that may be used by or with the parameterized predictor 658 to predict what the utterance prediction values ​​(i.e., compressor input power parameters, evaporator temperature drop) should be if the HVAC&R system is in a “new maintenance” condition for degradation detection purposes.

[0221] As described above, the purpose of the neighborhood extractor 656 is to determine whether and when to make a prediction based on the observed temperature tuples pointing to cells within the temperature map 654 for the parameter in question to assemble and provide a temperature tuple (T ei (n),T ci (n)), denoted herein as N(n), whose aggregate values ​​the parameterized predictor 658 can use to make the prediction. Figure 6F , the flowchart 660 generally begins at 661, where the neighborhood extractor 656 receives or is presented with an augmented observation (i.e., the nth observation of the sequence) provided by the VCC state generator 608 as a temperature tuple (T ei (n),T ci (n))O a (n) The neighborhood extractor 656 operates on the individual enhancement observations received from the VCC state generator 608, either one at a time as they are generated, or continuously in a frame of data.

[0222] In some embodiments, the neighborhood extractor 656 can simply ignore any observations from the VCC state generator 608 that do not meet the criteria for a steady-state observation with respect to utterance prediction, i.e., the relation learner is not active, as defined above. Therefore, at 662, the neighborhood extractor 656 determines whether the relation learner 650 for the utterance is active for the current observation. If the relation learner is not active, then in process step 665, the neighborhood extractor 656 immediately assigns a null value to the set N(n) of the observation, and processing of the observation is complete, with the null value indicating that no prediction should be made.

[0223] Assuming that the relation learner 650 is determined to be active for utterance prediction in 662, the neighborhood extractor 656 searches for ei and T ci The temperature tuple (T ei (n),T ci (n)) within a "neighborhood" of temperature tuples within + / - δ degrees, with a typical δ being 0.5°C. Thus, for example, if the nth steady-state observation of the system results in a temperature tuple (T ei (n),T ci (n)), the neighborhood extractor 656 searches for all temperature map cells (points) that satisfy equations (16) and (17):

[0224] T ei -δ≤T ei (n)≤T ei +δ (16)

[0225] T ci -δ≤Tci (n)≤T ci (n)+δ (17)

[0226] And the temperature tuple is within the established range of the temperature map 654.

[0227] For the above search, the neighborhood extractor 656 considers only "observed" temperature map cells, i.e., cells for which the "observed" metadata variable has been set to true in some embodiments, as discussed above or otherwise tested against a condition. Whenever the neighborhood extractor 656 finds a cell within the above neighborhood that is determined to be "observed" according to the above, it adds the corresponding temperature tuple (T ei (n),T ci (n)) are added to the initial empty or null set N(n). The neighborhood extractor 656 then allows (or recommends) making a prediction if and only if two criteria are met. First, a certain absolute minimum number of observed cells is mathematically required to determine the parameter coefficients of the parameterized predictor 658, but a larger number of observed cells can be used, and is preferred. Therefore, the minimum number of cells, Nmin, is determined by a system-dependent predefined constant, which must be at least the absolute minimum number of tuples required by the parameterized predictor 658, with larger numbers being preferred. In some embodiments, the parameterized predictor 658 requires an absolute minimum number of three observed cells, and in those embodiments, Nmin can be set to five cells.

[0228] To ensure that this first requirement is met, when the search is completed in process step 663, the set N(n) contains the number of tuples represented as Size(N(n)). In decision step 664, a test is performed to determine whether the number of tuples in the set N(n) is greater than or equal to the minimum number defined by the predefined constant Nmin described above. If this criterion is not met, the set N(n) is assigned the value null at 665, and the work of the neighborhood extractor 656 is completed for this observation.

[0229] If the neighborhood extractor 656 finds enough temperature tuples in the set N(n) at 664, the neighborhood extractor 656 proceeds to 666 to test the second criterion required for making a non-null prediction in the present invention, namely, the number of observed temperature tuples (T ei ,T ci ) lies within the convex hull formed by the subset of the set of temperature tuples denoted by N(n) collected as described above, where if a particular point of the utterance observation (T ei (n),T ci(n)) is a member of N(n), then the particular point of the observation is excluded for the purposes of this test. This criterion essentially means that the temperature tuple of the observation is "surrounded" by the temperature tuples of N(n). This allows the parameterized predictor 658 to perform local interpolation using the summary data contained in the cells corresponding to those tuples, rather than extrapolating outside the convex hull defined by the observed tuples, which can lead to inaccurate predictions, as is often observed in the lumped regression methods described above. Determining whether a point lies within the convex hull of a set of points is a common problem in the field of linear programming, and there are many "packing" solutions that can be used to make this determination. As an example, the packing function "linprog" included in the Python scipy.optimize library can be used in the determination, and there are many other packing functions in Python and other programming languages ​​that can make this determination. If it is determined in 666 that the observed tuple is not within the convex hull of the tuple of cells determined at 664 above, then the set N(n) is assigned the value null at 665 and the process is completed for that observation.

[0230] If the neighborhood extractor 656 determines at 666 that the observed temperature tuple is within the convex hull according to the minimum number of temperature tuples determined above, this can greatly improve the reliability of the prediction compared to prior art solutions. If both criteria at 664 and 666 are met, then at 668 the neighborhood extractor 656 provides the set of tuples N(n) found as above to the parameterized predictor 658, as described below with respect to Figure 6G discuss.

[0231] Figure 6G A flowchart 670 is shown that illustrates an exemplary process that may be used by or with the parameterized predictor 658 to make predictions. The flowchart 670 begins at 671, where the parameterized predictor 658 receives a set of temperature tuples N(n) and corresponding enhanced observations O provided by the neighborhood extractor 656 as described above. a (n) (or observation sequence). At 672, the parameterized predictor 658 determines whether N(n) has been set to null. If so, the prediction Set to empty and no prediction is returned for the given observation. If the set of temperature tuples N(n) is non-empty, then in process step 674, the parameterized predictor 658 builds a value table from the summary data in the cells of the temperature map 654 of the utterance prediction. The value table includes the T from each entry in N(n). ei ,T ciThe values ​​of and the average value of the parameter in question corresponding to each tuple in N(n) are presented as rows, which have been determined by the neighborhood extractor 656 to be the tuples whose corresponding cells in the corresponding temperature map meet the criteria for "observed" cells as described above. The average value is calculated from the summary data in the cell using equation (14). Assuming that there are m temperature tuples in the temperature map N(n), the resulting value table is given by

[0232] Table 5 Description:

[0233]

[0234] Table 5: Table of values ​​extracted from the temperature map

[0235] Using Table 5 constructed as above, the parameter coefficients of the parameter predictor in question can be determined from the data in the table in step 675. In some embodiments, the parameter model of the parameterized CIPP predictor 658' is in the form of a simple hyperplane of the following form:

[0236]

[0237] Among them, the parameter K x0 ,K xei ,K xci is only for the current observation value O a (n) Parameters of the effective predictor function, the current observation value O a (n) including T ei (n) and T ci In some embodiments, the parameter coefficients are calculated from the values ​​in Table 5 using an optimization program such as scipy.optimize.lsq_linear developed for the Python programming language or many equivalent packages in other programming languages.

[0238] When applied according to the above, these optimization procedures select the parameter K in equation (18) x0 ,K xei ,K xci The value of , which provides a "best fit" to the data in Table 5. Many optimization programs allow the values ​​of the parameters to be selected as needed to better represent the thermodynamics of the system. An example of this is when the power parameter is predicted using equation (18) above, the parameter K xei and K xci can be constrained to be non-negative to reflect that any increase in temperature should result in an increase in compressor power.

[0239] Once the parameter values ​​are established in process step 675 above, then in process step 676 the prediction extractor 656 uses the values ​​obtained in step 675 at the observation value O to a (n) Temperature tuple (Tei (n),T ci (n)) to evaluate the resulting equation (18) to calculate the predicted value of the utterance prediction and assign this value to And the process completes the utterance prediction for that observation.

[0240] above Figure 6F and 6G The process described in allows the CIPP processor 610 and the ETD processor 612 to provide predictions for utterance prediction. like Figure 6 As shown, in the case of CIPP processor 610 is designated as and in the case of the ETD processor 612 is designated as Based on these predictions of the utterance prediction values, the CIPP processor 610 and the ETD processor 610 can Figure 5A and 5B Generate normalized residual sequence R respectively P (n) and R E (n), when the predicted value is not empty, use equation (3) to calculate R P (n), and when Assign the value null to R when assigned the value null P (n), and when If not empty, use equation (8) to calculate R E (n), and when Assign value null to R when it is empty E (n). In addition, the CIPP processor 610 and the ETD processor 612 also provide the relative COP processor 613 with and Parameters are used to determine the corresponding relative COP sequence, expressed as rCOP(n), where or When either or both are empty, rCOP(n) is assigned a null value, and when neither is empty, rCOP(n) is assigned a null value. Details of the generation of the sequence rCOP(n) will be discussed later. One or more or all of these outputs (i.e., the normalized residual sequence R P (n) and R E (n) and the sequence rCOP(n)) are provided to the degradation detection processor 614 for further analysis.

[0241] As mentioned, the degradation detection processor 614 operates to interpret the normalized residual sequence and the sequence rCOP(n) to detect performance degradation and may issue an alarm signal or message or audio-visual display, or send information via a news feed, as generally indicated at 616, to notify of a potential problem with the HVAC&R system. Figure 6H The general operation of the degradation detection processor 614 is described.

[0242] refer to Figure 6H , a block diagram 680 is shown, which illustrates an exemplary operation of the degradation detection processor 614 according to an embodiment of the present disclosure. The purpose of the degradation detection processor 614 is to monitor the normalized residual R P (n) and R E (n) and the relative COP sequence rCOP(n), and issues alarms and warnings as needed when it detects potential problems via the degraded residual sequence. In the exemplary degradation detection processor 614 shown here, the normalized residual sequence R P (n) and R E (n) and the relative COP sequence rCOP(n) to be defined subsequently are received by a plurality of limit detection blocks, including a power parameter limit detector 682, an ETD limit detector 684, and an rCOP limit detector 686. These blocks operate in a similar manner to each other, such as Figure 6I and can be used by the degradation detection processor 614 to generate an alarm signal indicating that the power parameter normalized residual sequence R P (n) or the normalized residual sequence R of the evaporator temperature drop E (n), both provided by the prediction processor 606, or when the relative COP indicated by the sequence rCOP(n) and provided by the prediction processor 606 deviates significantly from 1.0, or equivalently, 100%, for any reason, where a 100% relative COP indicates operation as expected for the HVAC&R system under the new maintenance conditions.

[0243] Figure 6I An exemplary limit detector 690 and detection process that can be used is shown, showing how each of the power parameter limit detector 682, the ETD limit detector 684, and the rCOP limit detector 686 can be implemented in hardware or firmware. The input to each of the limit detectors 682, 684, 686 is shown symbolically as a "sequence of utterances" or SD(n), and can be a sequence R P (n),R E (n) or rCOP(n). Recall that whenever the requirement to provide a numerical value for an utterance sequence is not met, the prediction processor 606 ( Figure 6A) is processed into a speech sequence SD(n) (i.e., a normalized residual compressor input power parameter sequence R P (n) and the evaporator temperature drop parameter sequence R E (n) and the corresponding rCOP(n) sequence) are assigned null values. The non-null members of the sequence SD(n) are passed to a low-pass digital filter 692, which can be an EWMA (indexed weighted moving average) filter to reduce noise in the reference residual sequence. A general form of such a filter is shown in equations (19) and (20):

[0244] x(m+1)=βx(m)+(1-β)u(m) (19)

[0245] y(m)=x(m+1) (20)

[0246] Where x(m) is the mth updated internal state variable of the filter, u(m) is the mth value of the input sequence to the filter; the normalized residual y(m) is the mth output of the filter, and β is the EWMA filter time constant, which determines how quickly the filter responds to input changes. In some embodiments, a β value of 0.9996 can be used as the filter constant. The output of the filter 692 is the filtered utterance sequence SDf(n), where the resulting value of the filter output is used if the input sequence element is non-null, and the value null is used if the input sequence element is null.

[0247] The initial value x(0) for each of these EWMA filters is chosen to represent the expected value of the associated SD(n) for the system under fresh maintenance conditions. Since the values ​​of the power parameter residuals and the evaporator temperature drop residuals are expected to be zero for a freshly maintained system, x(0) is assigned a value of zero upon initialization. Assuming the relative COP series is expressed as a percentage, where 100% represents the coefficient of performance of a freshly maintained system, an appropriate value for x(0) is 100.

[0248] The filtered speech sequence SD from the low-pass filter 692 f (n) is provided to two threshold detectors, a high threshold detector 694 and a low threshold detector 696. The high threshold detector 694 operates to convert the filtered SD f (n) The non-zero sequence elements of the sequence and the preset high threshold T xh Compared, and when the filtered sequence SD f The value of (n) exceeds the high threshold T xhWhen SD_High_Alert is set, the logical variable SD_High_Alert is declared to have a Boolean value of true. Otherwise, the high threshold detector 694 declares the variable SD_High_Alert to have a Boolean value of false. Typical threshold values ​​will be discussed later. Similarly, the low threshold detector 696 generates a logical variable SD_Low_Alert as an output. When SD_High_Alert is set, the logical variable SD_High_Alert is declared to have a Boolean value of true. Otherwise, the high threshold detector 694 declares the variable SD_High_Alert to have a Boolean value of false. f (n) is less than the lower threshold T xl , the logical variable is assigned the Boolean value true; otherwise, it is assigned the Boolean value false.

[0249] In some embodiments, the high threshold detector 694 and the low threshold detector 696 may be implemented similarly to those previously described. Figure 6C The debounce logic 630 described in the debounce logic 630 is used to ensure that when one of the two alarm variables SD_High_Alert and SD_Low_Alert has been set to true, it is because the sequence SD f (n) has exceeded the threshold for a certain number of non-zero observations in a row, not just due to corrupted observations.

[0250] The filtered signal sequence SD generated within the limit detection process 690 f (n) may be of interest among other functions, especially when combined with enhanced observation sequences. a In some embodiments, the limit detection function 690 may set a null value for any observation where the input sequence SD(n) is null, so that the external function can determine which sequence elements have actually been updated by the limit detection function. Trend analysis may be applied to such sequences to estimate the sequence SD f (n) Dates and times when predefined thresholds will be crossed, indicating the rate of degradation and the sense of urgency of the service. Of particular interest is the trend analysis that can be performed on the SD f (n) to predict when the predicted subsequence will exceed the threshold T xh or T xl As an HVAC&R service provider, it would be helpful to know when, for example, the limit is expected to be exceeded within the next 30 days. f (n) is provided as the output of the limit detection function so that it can be subjected to external analysis.

[0251] Return to reference Figure 6H , the various threshold detectors operate as described above to provide an alert when the normalized residual becomes larger or smaller relative to what would be considered "normal" operation of the HVAC&R system. Figure 6I In the limit detection process 690, the threshold T xp and T xlThe selection of various thresholds is part of the field of HVAC&R degradation detection. For example, when the limit detection process 690 is applied to Figure 6H The power parameter limit detector 682, for most residential and commercial air conditioning systems, has adopted the T xp and T xl Typical values ​​for are +.05 and -.05, respectively. Experience has shown that when these limits are used and the output variable PP_High_Alert or PP_Low_Alert is asserted by the power parameter limit detector 682, if a maintenance technician is sent to the equipment site, the technician can almost always find something to repair, which results in the subsequent normalized residual power parameter sequence R P (n) and its filtered counterpart R Pf (n) tends toward zero again. PP_High_Alert typically indicates that something in the system is causing the pressure in the system to be greater than normal, and it can be inferred that condenser fouling or degraded condenser fan operation, a clogged expansion valve, or overcharge of refrigerant are possible causes, while PP_Low_Alert typically indicates that something in the system is causing the refrigerant pressure to be less than normal, and it can be inferred that refrigerant loss or undercharge, evaporator fouling or a dirty filter, other types of air flow obstructions, degraded evaporator fan operation, or possibly a stuck-open expansion valve are possible causes. Thus processed, the residual sequence R generated using the teachings of the embodiments herein is P (n) can be used not only to indicate degradation but also to infer the possible causes of degradation and to issue appropriate alarm signals.

[0252] It has been observed empirically that the evaporator temperature drop residual R E (n) tends to be larger than the power parameter residual R P (n) is more sensitive to system degradation, so T has been adopted in the ETD limit detector 684 xp and T xl Typical values ​​are +.1 and -.1, respectively. An ETD low alarm signal indicates system degradation due to causes that may include loss of refrigerant, reduced refrigerant flow to the evaporator due to expansion valve problems, or condenser problems (such as condenser fouling or condenser fan degradation), while an ETD high alarm signal indicates system degradation due to causes that may include evaporator or filter fouling, evaporator fan degradation or failure, or excessive refrigerant flow to the evaporator due to expansion valve problems.

[0253] T has been adopted in the rCOP limit detector 686 xp and T xlTypical values ​​are +115% and 90%, respectively. An upper rCOP alarm signal can be caused by a severe reduction in air flow through the evaporator, which can be caused by a fouled evaporator coil, a fouled air filter in the evaporator fluid stream, a malfunctioning evaporator fan, or excessive refrigerant loss resulting in frost on the evaporator. An expansion valve problem can also generate an upper rCOP alarm. A lower rCOP alarm signal can be caused by degradation sources including refrigerant loss, condenser fouling or condenser fan degradation, and expansion valve problems.

[0254] Furthermore, in some instances, the corresponding rCOP(n) sequence can provide an indication of the severity of the degradation in terms of wasted energy and can be used to infer the urgency of required service. For example, a PP_High_Alert combined with an average rCOP value of 0.98 would indicate that the equipment is operating at 98% efficiency even though a condition requiring attention exists, and the user might schedule maintenance for the next week or two or even wait until the upcoming scheduled maintenance to resolve the issue, understanding that the issue should be resolved in the near future, whereas an average rCOP value of 0.7 would indicate that the equipment is operating at 70% of its expected efficiency and should be serviced soon.

[0255] Similarly, and in a very common scenario, an air conditioner may begin to lose refrigerant due to a leak, and if this condition goes undetected, it may begin to lose cooling capacity. While the unit may provide adequate cooling on cooler days, on very hot days, it may no longer be able to maintain the appropriate temperature in the conditioned space. The PP_Low_Alert can be used to infer possible refrigerant loss, and if accompanied by a low rCOP value of 0.8, it can indicate that repairs should be performed very quickly, as the unit is operating at 80% of its expected efficiency. As mentioned above, several possible causes can be inferred. A PP_Low_Alert accompanied by an rCOP value above or near 100% can be used to infer that a filter replacement within the next few days will likely resolve the issue. Appropriate alarm signals can then be issued accordingly.

[0256] In some embodiments, multiple limit detectors of the type described above may be incorporated to indicate different "levels" of alarms. As an example, with reference to the detection of degradation via the power parameter normalized residual, one embodiment may create three instances of the power parameter limit detector 682, one instance having upper and lower limits (or thresholds) of +.03 and -.03, respectively, and a second instance having upper and lower limits (or thresholds) of +0.1 and -0.1, respectively. The first instance above may be used to indicate the need to schedule maintenance on the equipment, while the alarm generated by the second instance may be used to indicate the need for immediate repair. The third instance (perhaps set to +0.2 and -0.2) may be used to automatically shut down the HVAC&R equipment (or portion thereof) by, for example, remotely opening the branch feeder circuit 114 when corresponding via appropriate control or trip signals to its circuit breaker. Similar limits and controls may be established and implemented using similar thresholds for any limit detector of the degradation detection processor 614, including the ETD limit detector 684 and the relative COP detector 686. Furthermore, these limit detectors 682, 684, 686 do not necessarily have upper and lower thresholds that are symmetrical about a center point, ie, the magnitude of the lower threshold may be different from the upper threshold.

[0257] If the physical HVAC&R system could be maintained in the newly maintained condition long enough to obtain observations over the full range of temperature tuples that the system might encounter during one or more operating weather seasons, then the temperature map thus constructed using only the recognition function would be sufficient to fully characterize the system. As previously discussed, this is generally unlikely, so an apparatus is now described that allows system characteristics of a "newly maintained" system to be learned as system performance degrades.

[0258] It should be recalled here that in some embodiments, each observation provided by the data acquisition processor 600 includes a timestamp indicating the date and time when the observation was obtained, while in other embodiments, the VCC state generator 608 of the prediction processor 606 may implicitly track the date and time of a given observation, or simply track the time elapsed from a reference time. Learning involves the relation builder 652 of the relation learner 650 of the utterance prediction, which converts the appropriate function f x (X,n) is applied to the utterance prediction and the sample statistics of the cells of the temperature map 654 are constructed using the condenser and evaporator entry temperatures, which can be used to predict the power parameters or ETD values ​​of the equipment under the "new maintenance" condition as described above and can be used with the help of Figure 7 This is best illustrated by the exemplary timing diagram of FIG.

[0259] Reference Figure 7, the timing diagram 700 typically begins when the HVAC&R system, including the agent 314 and its relationship builder 652, has been commissioned or otherwise deployed, and assumes that the HVAC&R equipment is in a "new maintenance" condition when learning begins. Once these conditions are met and learning is enabled, learning of the prediction of utterance features begins at 702 by receiving an initial valid observation for which the relationship learner 650 for the utterance parameters is active (i.e., the compressor is on and, in the case of a single compressor system, the appropriate state variable S p (n) or S e (n) is set to true). The steady-state observation is presented to the relationship builder 652 and is preferably the first steady-state observation received after the above considerations are satisfied. Learning continues by receiving additional steady-state observations over a learning interval 704, which is defined by the learning interval system constant. After the learning interval 704 is completed, the relationship builder 652 is considered to have sufficiently learned the characteristics of the utterance prediction of the HVAC&R system, and once learned, these characteristics should not change over time in the absence of system degradation. If the system has degraded and subsequently returned to a newly maintained state, the relationship should again reflect the newly maintained characteristics of the system without further training.

[0260] like Figure 7 As shown, the learning interval 704 includes two component intervals, a "bootstrap" interval 706 and a compensation learning interval 708. As the name implies, the "bootstrap" interval 706 jumps in and starts the learning process of the relationship builder 652. It is assumed that the physical HVAC&R system starts and remains in a new maintenance condition during the bootstrap interval, and during this interval, the relationship builder 652 applies the identity function of the above equation (11) to predict the utterance value of the steady-state observation to update the sample statistics of the corresponding unit. In other words, during the bootstrap interval, when the steady-state observation is within the bootstrap interval, the relationship builder 652 uses the predicted unmodified value of the utterance item of the steady-state observation to update the sum of the SV part of the corresponding unit (i.e., f x (X,n)=X).

[0261] The bootstrap interval 706 begins with the receipt of the initial steady-state observation at 702 and ends at 710 after a predefined duration indicated by the bootstrap interval system constant. The bootstrap interval 706 can be as short as a few days, but may actually need to be set as high as the first 30 days of system operation, depending on the specific HVAC&R system. During the bootstrap interval, the compensation metadata for each cell is false. If enough observations are made during the bootstrap interval for a cell to be marked as "observed" by setting the cell's "observed" metadata to true according to the logic described above, the sum in the cell will no longer be updated and will effectively remain constant "forever." These cells can be identified by the metadata of Observed = True and Compensated = False and are designated as "reference cells" below because they are the cells most likely to represent the system in its new maintenance condition.

[0262] Following the introductory interval is a compensation learning interval 708, during which the assumption that the system remains in the new maintenance condition is relaxed. During this compensation learning interval 708, the relationship builder 652 may modify the value of the power parameter in the steady-state observation using the time-varying compensation function mentioned above to compensate for the estimated degradation before updating the sample statistics for the unit. When the relationship builder 652 updates the unit during the compensated learning interval 708, because the observation metadata variable is not already set to true, it sets the compensation metadata variable for the unit to true, indicating that at least one of the predictions of the utterance value used to update the sample statistics for the unit has been modified using the compensation function. The compensation learning interval 708 begins at the end of the introductory interval at 710 and continues until the end of the learning interval at 712, thereby completing the learning interval 704. In some embodiments, a typical value for the learning interval 704 is approximately 120 days, although fewer or more days may be used.

[0263] Once the learning interval 704 is complete, the learning of the relationship builder 652 is deemed sufficient for purposes herein, and the temperature map 654 is deemed fully representative of the expected operation of the HVAC&R system, such that no further learning is required by the relationship builder 652. Thereafter, the relationship builder 652 is idle unless it is determined by other means that it needs to be restarted, such as when the HVAC&R equipment has been replaced with new or different equipment.

[0264] Compensating the power parameter values ​​prior to updating the sample statistics during the compensation learning interval 708 is facilitated by the time-varying reference degradation generator function described next. The reference cells of the map that were declared as "observed" during the pilot interval 706 as described above (i.e., observed = true, compensated = false) are likely to be most representative of the system in the new maintenance condition because a) they represent observations that are closest in time to the time when the system was placed in the new maintenance condition, and b) enough observations have been made so that the sample statistics of the cell are likely to represent the actual characteristics of the system at that temperature tuple. For these cells, the predicted mean of the discussion given by equation (14) is an estimate of the prediction of the discussed value of the equipment under the new maintenance condition for the corresponding temperature tuple. Since the observed = true data variable indicates that the relationship builder 652 will no longer update the summary statistics for that cell, the prediction of the discourse estimate of the mean of the cell so generated via equation (14) is now a constant.

[0265] During the above-mentioned pilot interval 706, the relationship builder 652 assumes that the HVAC&R system remains in the new maintenance condition, which is a reasonable assumption if the pilot interval duration is short. It has been observed that in practice, any normalized residual of the temperature from the prediction in question (R in the case of the power parameter) P (Equation (3)) or R in the case of a drop in evaporator temperature E The relationship between (Eq. (8)) is quasi-temperature independent, at least for degradation levels that are not generally considered extreme. As used herein, the term "quasi-temperature independent" means that the predicted normalized residuals discussed above are approximately independent of the observed temperature tuple (T) over the operating temperature range of the HVAC&R system as long as the physical conditions of the equipment do not change. ei (n),T ci (n)). Experience has shown that this is true in practice, at least for relatively small magnitude normalized residuals within the temperature range considered "normal," and begins to be violated as the system degrades to a level where a service call for maintenance would be recommended.

[0266] Consider an HVAC&R system where the above assumptions hold and the characteristics of the system have been learned, and the temperature map 654 has acquired multiple reference cells during the pilot interval 704, but not all cells in the temperature map 654 meet the conditions for being a reference cell. Furthermore, assuming that a sufficient number of reference cells have been acquired, the prediction extractor 656 can use these cells when encountering them in subsequent steady-state observations to use the average of the utterance predictions of the indexed reference cells calculated according to the above equation (14) as the prediction To predict the "new maintained" value of the utterance prediction for at least some of the time observations. For the steady-state observations for which the relation builder 652 indexes the reference unit, the relation builder 652 may then calculate the normalized residual R according to equations (3) or (8), as appropriate. x to be suitable for utterance prediction, where X is the predicted value of the observed utterance and As calculated above. Due to the quasi-temperature independence assumption, the normalized residual R calculated under these conditions is x The value should be independent of the temperature tuple, as described above, and therefore independent of the cells in the temperature map 654 used to make the prediction. In other words, as long as the physical conditions of the HVAC&R system do not change, its temperature tuple (T ei (n),T ci (n)) Any steady-state observation corresponding to one of the reference cells should produce (approximately) the same R x value.

[0267] In the absence of system degradation and measurement noise, the residual R x should be zero or close to zero, since the prediction of the predicted utterance value should be equal to the prediction of the observed utterance value. As understood in the art, system degradation manifests as R x The bias in , and it has been shown to be beneficial in detecting system degradation. Designated as R x The individual residual R obtained for (m) x A sequence of (where index m indicates the temperature tuple (T ei (n),T ci The mth such residual computed by the relationship builder 652 when (n)) indexes the reference cell may be used to infer the evolution of the degradation of the system for the purpose of compensating for the observation.

[0268] Ideally, the normalized residual R x (m) will represent the true normalized difference between the measured value of the utterance prediction and the value of the utterance prediction, where the device is in the new maintenance condition, but assumed to be used to calculate the reference residual value R x (m) The utterance predictions for steady-state observations are corrupted by additive noise, as described above (Equation (10)). As a result, the sequence of reference normalized residuals may be somewhat noisy. By appropriate signal processing (e.g., filtering), the estimation of the sequence of normalized residuals can be performed such that the effects of the noise in the observations are relatively insignificant.

[0269] In some embodiments, the agent uses a simple filter, such as the EWMA filter discussed above via equations (19) and (20), to reduce noise in the reference residual sequence. In the computation of the reference residual estimate, the input sequence u(m) is a sequence of residuals R calculated by the agent's residual estimator function according to the above. s(m), and the output sequence y(m) represents R Xsys (m) System degradation sequence, where the subscript "Xsys" implies the specific system residual sequence used for speech prediction. In some embodiments, an exemplary value of β is 0.98. A suitable initial value for x(0) is 0.0.

[0270] As a next creative step, suppose R Xsys Express the coefficient R in the form of normalized residual Xsys (m) The latest estimate of the system degradation sequence. It is also assumed that a steady-state observation with the predicted value X is made within the compensation learning interval 708, for which the cell in the temperature map 654 represented by the temperature tuple does not meet the requirements of the observed cell, that is, the observed metadata variable of this second cell is set to false. Since R Xsys represents the entire system, so according to equations (3) or (8) above, depending on the prediction in question, it can be calculated from R Xsys and X define an adjusted value f of the observed power parameter that more closely represents what would be observed in the absence of system degradation x (X,n), as follows:

[0271]

[0272] Equation (21) can then be solved for the adjusted value of the power parameter:

[0273]

[0274] The adjusted observation f from equation (22) above x (X,n) represents the best estimate of the relationship builder 652 for what the observation X(n) would be in the absence of system degradation, and is based on the R Xsys The value of , and is the above-mentioned second time-varying compensation function applied to the observed utterance prediction before updating the summary statistics of the cell corresponding to the temperature map 654. During the compensation learning interval 708, the "correction" value f x (X,n) updates the summary statistics of the unit corresponding to this observation, rather than the original measurement prediction X(n) made during the pilot interval 706, which should better represent the operation of the equipment under the new maintenance conditions. The relationship builder 652 uses this value to update the sample statistics of the unit during the compensated learning interval 708.

[0275] The above discussion provides ways in which the agent 314 uses the relationship builder 652 to extend the temperature map 654 beyond the cells that can be fully learned during the guidance interval 706. Figure 8 The process of maintaining temperature map 654 for individual observations is described in further detail in .

[0276] refer to Figure 8 , a flowchart 800 is shown illustrating a method that can be used by or with the relationship builder 652 of the agent 314 and the active relationship learner 650 as described above to maintain a temperature map 654 for individual observations. The method generally begins at 802 when the relationship builder 652 receives a temperature tuple (T ei (n),T ci (n)) New steady-state enhancement observation O a (n). At 804, the relationship builder 652 checks whether the time of the observation is within the learning interval 704. If not, the observation is not used to maintain the temperature map 654, and the control flow proceeds to 822, where no further action is taken with respect to the observation with respect to the temperature map 654. If it is determined at 804 that the observation was obtained within the learning interval 704, the relationship builder 652 determines at 806 whether a sufficient number of observations have been obtained (e.g., with the observation-temperature tuple (T ei (n),T ci (n)) The observation metadata variable of the corresponding unit is true).

[0277] If the determination at 806 is yes, then at 808, the relationship builder 652 determines whether to update the residual sequence estimator for the observation being processed (e.g., whether the compensation metadata variable is set to false). If not, the observation being processed is not used to update the residual sequence estimator R Xsys , and the relationship builder proceeds to 822 where no further action is taken with respect to the temperature map 654 regarding this observation. If the determination at 808 is yes (e.g., the compensation metadata variable is true), the relationship builder 652 proceeds at 810 to update the residual sequence estimate R referenced above. Xsys . Reference below Figure 9 The estimator update function described further provides two details that are useful for maintaining the temperature map 654 during the compensation learning interval 708. First, the function updates the residual sequence estimator R Xsys Secondly, it provides an indication of whether system degradation observed subsequently within the compensation learning interval 708 should be compensated for before being used to update the temperature map 654. In some embodiments, this indication may be in the form of a Boolean system state variable, such as COMPENSATION_ENABLED, the generation of which will be subsequently Figure 9 In R Xsys Following the update of the estimate, flow proceeds to 822 where no further action is taken on the temperature map update with this observation.

[0278] Referring back to 806, if a sufficient number of observations have not been obtained for the cell (e.g., the observed metadata variable is false), the relationship builder 652 continues to process the observation as a candidate for updating the temperature map 654 by determining whether the observation was obtained during the bootstrap interval 706 at 812. If the time of the observation lies within the bootstrap interval 706, the relationship builder 652 uses the observation to update the cell corresponding to the observed temperature tuple at 820 by updating the cell's summary data using the identity function of equation (11) above (and also updating the in-process observation metadata variable).

[0279] If the determination at 812 is no, meaning that the observation is not within the guidance interval (706) but is within the compensation interval (708), the relationship builder 652 determines at 814 whether the observation should be compensated for degradation (e.g., the COMPENSATION_ENABLED state variable is true) before updating the summary data for the unit. If not (e.g., the COMPENSATION_ENABLED state variable is false), the agent takes no further action on the temperature map at 822. If observation compensation is enabled for the unit (e.g., the COMPENSATION_ENABLED state variable is true), then at 816, the relationship builder 652 compensates for degradation of the observed value (i.e., the compressor input power parameter or the evaporator temperature drop) of the utterance prediction included in the observation by calculating fX(X,n) using equation (22) above, and indicates at 818 that the observation has been compensated (e.g., by setting a compensated metadata variable to true). Thereafter, the relationship builder 652 uses the observed utterance prediction fX(X,n) at 820 to determine the observed value. x The adjusted value of (X,n) is used to update the summary data for the cell (and also update the observation metadata variables in the process). At this point, no further action is taken on the temperature map with respect to this observation 822.

[0280] Figure 9 Shown Shown Figure 8 R cited in sys Functional diagram 900 of additional details of the estimator update process 900. The estimator update process 900 provides the latest updated value of the system degradation level, the residual sequence estimator R sys , and updates the value of the COMPENSATION_ENABLED state variable. The process generally begins at 902, where the agent calculates the value of the COMPENSATION_ENABLED state variable by following equation (14). ei ,T ci ) cell calculation index The normalized residual for the current observation is calculated using the relationship learned from the temperature map 654, resulting in the calculated residual R shown x (m). Recall that according to Figure 8, the temperature tuple corresponding to the observation (T ei ,T ci ) has the observation metadata variable set to true, and the compensation metadata variable of the cell is set to false. Based on the aggregated data for this unit, the predicted value is the average of the utterance predictions As given in equation (14) above. Based on the predicted value and the observed value of the utterance in the observation, the normalized residual R x This can be calculated by equation (3) or (8) above as appropriate for utterance prediction. The agent then feeds this normalized residual (via the relation builder 652) at 904 into R sys In the estimator, R sys An estimator can calculate and output R sys A simple filter for estimation, such as the EWMA filter mentioned above.

[0281] R from 904 sys The concept of estimating the suitability of compensation for system degradation relies on the assumption that the residuals are quasi-temperature independent. This assumption has been observed to be reasonable when the magnitude of the residual series is small. However, the assumption begins to break down when the condition of the equipment deteriorates to the point where repairs are required to restore the equipment to normal operation. In practice, it has been shown that in HVAC&R applications, service is typically guaranteed when the magnitude of the normalized residuals of the power parameters consistently exceeds approximately 4% to 5%. Typical limits for evaporator temperature drop are approximately 10%, and the quasi-temperature independence assumption begins to break down before these limits are reached. Once the equipment is restored to its newly maintained condition, attempts to compensate for degradation under these conditions can have uncertain effects.

[0282] Thus, in some embodiments, the agent (via the relationship builder 652) maintains a Boolean system state variable COMPENSATION_ENABLED for each relationship managed by the relationship builder 652 to Xsys The R calculated by the estimator 904 Xsys In one embodiment, the current value of R Xsys The R calculated by the estimator 904 Xsys The value of is the input of the absolute value function 906, the output of which is shown as |R Xsys Absolute value |R Xsys| is then fed into the compensation threshold function 908, which operates based on preset compensation limits and a composite hysteresis. These parameter inputs are system-dependent and, in some embodiments, may be represented by the variables "CompensationLimit" and "CompensationHysteresis." Typical values ​​for these parameters are 0.02 and 0.002, respectively, for the power parameter residual, and 0.05 and 0.005, respectively, for the evaporator temperature drop residual. These two parameters work together to create two thresholds, labeled T according to the following equation: low and T high :

[0283] T low =CompensationLimit-CompensationHysteresis (23)

[0284] T high =CompensationLimit+CompensationHysteresis (24)

[0285] The output of the compensation threshold function 908 is the Boolean system state variable COMPENSATION_ENABLED mentioned above, which is used to indicate the system residual R to the relationship builder 652. Xsys Is it within the range assumed to be valid for applying degradation compensation. In some embodiments, when the system is initialized, the state variable COMPENSATION_ENABLED is set to true. If sys and subsequently |R Xsys |Afterwards,|R Xsys The mth value of | is less than T low , then the mth value of the COMPENSATION_ENABLED state variable is always set to true. Similarly, if |R Xsys The mth value of | is greater than T high , the COMPENSATION_ENABLED status variable is always set to false. low ≤|R sys |≤T high |R Xsys |, the value of the COMPENSATION_ENABLED status variable remains unchanged.

[0286] So far, the embodiments herein have primarily focused on Figure 3A basic HVAC&R system 100 is shown and described in the accompanying drawings, which utilizes a single compressor, a single evaporator, and a single condenser. More complex VCC-based systems than the basic HVAC&R systems discussed thus far can also benefit from the principles and teachings herein. Many commercial and industrial HVAC&R systems, for example, have multiple compressors rather than a single compressor. Multiple compressors are housed within a single mechanical package and operate individually or in parallel in response to heat load conditions.

[0287] Figure 10 An example of an HVAC&R system 1000 having multiple (e.g., two) compressors, equipped with the early problem detection system 300 discussed herein, is shown. The early problem detection system 300 otherwise operates in a manner similar to that described above with respect to the HVAC&R system 100 of FIG. 1 using similar components, except that the early problem detection system 300 predicts compressor input power parameters for two compressors 1002 and 1004 rather than a single compressor. As can be seen, each compressor 1002, 1004 is driven by a respective motor 1002a, 1004a, with the input power of each motor 1002a, 1004a measured by respective current sensing devices 310a, 310b and power parameter meters 312a, 312b. In this arrangement, it has been observed that when both motors are operating, each motor 1002a, 1004a individually consumes less power than if either motor were operating alone. The input power measurements from each power parameter meter 312a, 312b are then provided to an agent 314 which processes the measurements to derive a CIPP relationship for each compressor 1002, 1004 using separate temperature maps of the CIPP relationship for each compressor when operating individually or in series, respectively.

[0288] In other HVAC&R systems, there may be multiple refrigerant circuits, each supported by one or more compressors. In many of these systems, each refrigerant circuit has its own condenser coil (and fan assembly in the case of direct exchange), and the condenser coils may be physically separated in space so that they can experience significantly different inlet temperatures. This is often the case, for example, with rooftop units, where for certain parts of the day, one condenser coil and the nearby roof are directly in the sun, while the other side is shaded. For this reason, each condenser assembly may require a condenser inlet temperature sensor. Many of these multiple refrigerant circuit systems share interleaved evaporator coils, where the refrigerants of the various circuits remain separate from each other, but all circuits cool the same fluid flowing through the interleaved evaporators. In this case, a single evaporator inlet temperature sensor and a single evaporator discharge sensor can be employed, even though multiple condenser inlet temperature sensors are present.

[0289] In some chilled water systems, each refrigerant circuit has its own condenser coil, which may be physically separated in space, and its own evaporator coil, which may be separated in space. In these systems, each refrigerant circuit cools its own fluid, and the fluids mix upstream. In this type of system, there may be more than one evaporator inlet temperature sensor and more than one evaporator discharge temperature sensor. From a practical design perspective, it is preferable to construct the system so that each compressor has its own virtual condenser and evaporator inlet temperature sensor in order to manage the various CIPP and ETD relationships that may be required. Each CIPP and ETD relationship requires a separate relationship learner for each of the above-mentioned state values ​​Sc(n), with the compressor in the on state encoded according to Table 2 above.

[0290] Consider the case of staggered evaporator coils in a direct exchange system. For a given inlet air flow temperature and rate (mass flow rate) on the evaporator function, the power required by one compressor in a multi-compressor system will depend on the state of the other compressors. If two compressors are used to cool the air, the power consumed by either compressor operating in series is expected to be less than the power of the same system under the same conditions if only a single compressor were operating. From a CIPP perspective, the important point is that the operating characteristics of a given compressor in a system may depend on the state of the other compressors in the system. Therefore, for each combination of compressors that the compressor is operable with, a CIPP relationship is preferably maintained for each compressor.

[0291] Similarly, multiple versions of the ETD relationship may be needed in a multi-compressor system. In a staggered evaporator system with multiple refrigerant circuits, different EDT relationships can be expected, depending on which compressors are "on" and "off". Intuitively, for a fixed condenser inlet temperature, when both compressors are on, the ETD relationship is more accurate than when only one compressor is on. Figure 10 A system with a larger evaporator temperature drop would be expected.

[0292] It should be noted that in the foregoing embodiments, the agent has little control over the condenser entry temperature, as the entry temperature may depend on many factors, including weather, time of day, orientation of the condenser, etc. In operation, the agent is simply presented with the entry temperature as observations of the HVAC&R system to be monitored, each observation including one or more condenser entry temperatures T ci , one or more evaporator inlet temperature T ei and the minimum value of the compressor input power parameter P of each compressor in the system. The compressor input power parameter P can be compressor current, active power, volt-ampere, etc.

[0293] As a matter of learning or debugging the configuration, each compressor is assigned an appropriate condenser inlet temperature measurement or combination of compressor inlet temperature measurements, an evaporator inlet temperature measurement or combination of evaporator inlet temperature measurements, and a measured power parameter for that compressor. In some systems, a single condenser inlet temperature may suffice for all compressors, but in some systems it may be advantageous to have different condenser inlet values, particularly when there is more than one condenser that may be oriented differently from one another. Similarly, in a chiller system, each chiller compressor unit has its own evaporator function, and it may be advantageous to assign a separate temperature to each inlet. In other systems, a staggered evaporator assembly may be employed, in which case a single temperature measurement may be sufficient for all compressors in all refrigerant circuits containing staggered evaporators.

[0294] In some systems, multiple compressors may be employed in a single refrigerant circuit, while in other systems incorporating condenser and evaporator units that are staggered or in close proximity to each other, the characteristics of a given compressor learned by the agent may be a function of the system's "compressor state" (i.e., which compressors are on or off at a given time).

[0295] Furthermore, the fluid at the aforementioned inlet does not need to be air. Water or a chemical mixture (such as ethylene glycol and water or a salt solution) can be used as the evaporator ambient fluid or the condenser ambient fluid. In so-called chilled water systems, a liquid evaporator ambient fluid is circulated through the system as a liquid. This cooled liquid fluid can be circulated through the building to various radiators, where it can be used for remote cooling. This can be used to cool large areas, such as schools, hospitals, and commercial buildings, as well as more common spaces, such as supermarket refrigerators and freezers, where the chemical mixture can be cooled to well below the freezing point of water. The condenser ambient fluid can also be a liquid. This is useful in large chilled water systems, where the condenser fluid can be circulated over the condenser coils of the system located inside the building, and heat is transferred to a heat exchanger located outdoors. Such a system can have advantages over direct exchange systems because it does not require long runs of refrigerant lines operating at high pressure to and from the outdoor heat exchanger. A very common chilled water system, known as an air-cooled chiller, uses direct heat exchange with air as the condenser ambient fluid while simultaneously cooling a liquid as the evaporator ambient fluid. This allows the entire mechanical system, including the compressor and condenser fans, to be located outdoors or outside the building.

[0296] In a heat pump system operating in heating mode, a reversing valve reverses the roles of the condenser and evaporator, as depicted in Figure 1, with the condenser function located inside the conditioned space and the evaporator function drawing heat from the outdoor environment. The physical heat exchangers do not move, but their actions are reversed. The evaporator function (now outside) absorbs heat from the outdoor ambient air and rejects this heat to the conditioned space air via the condenser function (now inside). In this case, frost typically condenses onto the evaporator coil function (outside), which must occasionally be defrosted as part of normal operation.

[0297] Thus, the disclosed monitoring and early problem detection system offers numerous benefits when expanded to more complex HVAC&R systems. Due to the potential for interaction between compressors in a multi-compressor system, in some embodiments, the agent allocates and maintains separate relationship learners 650, one for each compressor in the system and each compressor state Sc(n) in which the compressor is operational or in the on state. For example, in a three-compressor system with a total of eight possible combinations of compressor on / off states, a total of 12 pairs of relationship learners are required to learn the CIPP and ETD relationships, one pair per compressor for each individual compressor on state Sc(n).

[0298] In some embodiments, for a given enhanced observation O a (n), a relationship learner assigned in this manner is encoded according to Table 2 above if a) it has been assigned by the agent to a particular compressor in the encoded state Sc(n), and b) the appropriate "stable" state Sp(n) (in the case of the CIPP relationship) or Se(n) (in the case of the ETD relationship) is true. A test for the active relationship learner is performed for each observation, and all such relationship learners operate as described above. A result of this embodiment is that even if multiple relationship learners are assigned to a given compressor to represent different combinations of compressors, at a given observation time, at most one pair of relationship learners are active for each compressor. By selecting only the non-zero residuals and relative COP values ​​of the compressors, a single series of residuals for each compressor can be presented to the degradation processor 614, appropriately expanding the scope to monitor each compressor.

[0299] While direct, isolated measurement of the compressor power parameter can produce the most accurate prediction of the compressor power parameter as described herein, and the methods have been described in these terms, simply responsive to a signal of the compressor power parameter can similarly provide useful information, and such instrumented systems can be valuable in detecting HVAC&R system degradation. In particular, in many HVAC&R systems, it is simpler to monitor the power parameter of the electrical feed to the entire unit or a portion of the unit rather than directly measuring the compressor. Many, if not most, HVAC&R units are driven by isolated branch feeder circuits, which may have current or power measurement capabilities built into the circuit breakers. Many of these circuit breakers provide remote activation capabilities, and many residential split systems, modular units, and commercial rooftop units have a disconnect physically located near the unit to allow an HVAC&R technician to electrically isolate the unit for maintenance purposes. The power fed to the entire unit typically includes power to the condenser fan and multiple compressors, which increases the power consumed by the compressor.

[0300] The above embodiment of whole or part unit feeding is Figure 3 and Figure 10 An alternative embodiment is shown by dashed lines in FIG. Figure 3 and 10 As shown, in some embodiments, instead of (or in addition to) a power parameter meter such as the power parameter meter 312, the input to the power parameter processor 604 can be provided by an energy meter embedded in the branch feeder circuit 114 or included in the electrical disconnect box or other auxiliary device 116. The energy meter can be a discrete meter that forms part of the branch feeder circuit 114, or it can be integrated into the feeder circuit 114, such as in a circuit breaker of the feeder circuit 114. In either case, the power measured by the energy meter reflects the entire or partial unit power input to the HVAC&R system 100. This feeder circuit power input can then be provided to the proxy power parameter processor 604 to detect HVAC&R system degradation in a manner similar to that described for the power parameter meter 312.

[0301] Those skilled in the art will appreciate that other embodiments may be used within the scope of the present disclosure. From a practical perspective, a desirable characteristic of a learning system for monitoring HVAC&R systems for developing problems is that it becomes functional quickly and does not require long training intervals during which equipment degradation is not monitored. That is, to the extent practical, agent 314 should learn time-invariant CIPP and ETD relationships on the fly.

[0302] Now go to Figures 11A to 11C , recall from above that, in some embodiments, the agent only observes the temperature tuple (T ei(n),T ci (n)) lies within the convex hull of the set of observed tuples. In these embodiments, a newly observed temperature tuple must lie within the convex hull formed by the previously observed tuples (points) in the original set that the agent used to learn the CIPP relationship. This ensures that the agent interpolates between tuples (points) that have been "seen" by the agent, rather than extrapolating from points that have never been seen. In some embodiments, the convex hull can be defined as follows. Given a set of training points {X} in Euclidean space, the convex hull H(X) of the set {X} is the minimal set containing the points in {X} for which every point on any line between any two points in H(X) lies entirely within H(X).

[0303] Figures 11A to 11C An example of hull convexity according to some embodiments is shown graphically. Figure 11A , an exemplary convex hull 1100 is created by a set {X} containing five two-dimensional tuples labeled P1 to P5. The line segments P1→P2, P2→P3, P3→P4, and P4→P1 form the edges of the convex hull 1100 defined by the set {X}. In this example, the tuples P1 to P5 that define the edges of the convex hull 1100 are included in the convex hull. The hull is "convex" because any line segment in the hull, including those formed by tuples on the edges of the hull, lies entirely within the hull. Tuple P5 also lies within the hull. Visually, it can be seen that the convex hull 1100 is the smallest set of tuples that contains all tuples in the set {X} and is convex.

[0304] Figure 11B An example of a tuple P is shown that lies within the convex hull 1100. If an interpolation model made from the set of tuples {P1 ... P5} is applied to the tuple P, the model interpolates between the values ​​of the tuples within the set.

[0305] Figure 11C An example of a tuple P located outside of convex hull 1100 is shown. In this example, a line drawn between P and, for example, P5, includes points located within convex hull 1100 as well as points located outside of it. If an interpolation model constructed from the set {P1…P5} is applied to tuple P, the model extrapolates from the values ​​of tuples within the set. In general, extrapolation is typically less accurate than interpolation. Therefore, this agent requires that any tuple for which a predicted compressor input power parameter value is to be determined must be within the convex hull of the observed tuples.

[0306] As discussed, embodiments of the monitoring agent herein use the CIPP relationship to predict compressor input power parameter values ​​for an HVAC&R system under "fresh maintenance" conditions and compare these values ​​to observed compressor input power parameter values ​​for early detection of performance degradation. Embodiments of the monitoring agent can similarly learn the ETD relationship and calculate a sequence of normalized temperature drop residuals that can similarly detect performance degradation early. The agent can also use a combination of the CIPP and ETD relationships to not only detect the presence of a problem, but also indicate the likely nature of the problem. As explained, the process of learning the ETD relationship via a separate temperature map is almost identical to the process of learning to predict the expected power parameter and the normalized power parameter residual, only requiring the evaporator temperature drop in place of the compressor input power parameter. The mechanism for learning the evaporator temperature drop when the system degrades is also the same, but optionally, the compensation limits and compensation lags in the above equations (23) and (24) can be selected separately for the CIPP and ETD relationships as needed or desired, as well as the lead blanking intervals 402 and 422. The resulting ETD normalized residual sequence is then presented to the degradation detection processor 614 along with the CIPP normalized residual sequence as discussed.

[0307] In addition to the above, the monitoring agent simultaneously predicts the expected power parameter sequence and the expected evaporator temperature drop sequence given the observations and uses Equation (5) from T ed and T ei The ability to calculate the power parameter and the measured value of the evaporator temperature drop E allows the agent to calculate the observed relative COP, as discussed. This provides a number of additional benefits. On the one hand, the relative COP can be used not only to detect degradation, but in some cases also to quantify the energy usage and cost attributable to the degradation. A monitoring agent such as the one described above that can quantify degradation in the form of relative COP relative to the learned normal conditions of the system provides significant advantages. For example, knowing that a system has degraded in performance to a certain percentage (X%) of its original efficiency means that, under the observed conditions, the system operating cost is approximately 100 / (X%) more than when the system was newly maintained. Since the agent can monitor the compressor input power parameter (as described above) and the evaporator temperature drop to calculate the relative COP, the agent can also provide an estimate of the power consumed, and therefore the operating cost of the system degradation, and thus declare a problem not only when it is detected, but also to declare a sense of urgency when the degradation becomes too expensive.

[0308] As about Figure 6 As discussed above, the monitoring agent 314 (and the prediction processor 606 therein) uses the relative COP processor 613 (as well as the CIPP processor 610 and the ETD processor 612) to calculate the relative COP. In some embodiments, the relative COP processor 613 may use the VCC state generator 608 provided by the enhanced observation.a The measured power parameter measurement P(n) and the measured evaporator temperature drop measurement E(n) and the power parameter prediction provided by the CIPP processor 610 and the evaporator temperature drop prediction provided by the ETD processor 612 To calculate the relative COP, use the relative coefficient of performance or rCOP sequence, as used herein.

[0309] As background, the “rate” form of the instantaneous coefficient of performance (COP) for an HVAC&R system is typically defined as:

[0310]

[0311] Among them, Pwr system is the total electrical power delivered to the system including one or more compressors, fans, controllers, pumps, etc., and is h t The total rate at which heat is removed from the air passing over the evaporator coil (or in addition to the air passing over the evaporator coil in the case of a heat pump). "h" is often used to refer to the heat transfer rate in Watts or J / S. This total heat rate includes the sensible heat rate, h s (i.e., can be sensed), which manifests itself as a temperature drop across the evaporator, and the latent heat rate h l , which is the rate at which the air loses heat as moisture condenses on the evaporator coil:

[0312] h t =h s +h l (26)

[0313] The COP defined by equation (25) above captures the efficiency of the device, but is difficult and expensive to measure. As might be expected, it is a sensitive function not only of the device conditions but also of the inlet temperature tuple (T ei ,T ci ), humidity, and the mass air flow rate of the heat transfer fluid at the inlet and discharge of the evaporator. In order to measure the input power to the total system, a power meter needs to be applied at the power feed to the equipment, which can be expensive, as briefly discussed above. Measuring the total heat removed from the conditioned space by the evaporator typically involves employing expensive mass air flow sensors at both the inlet and discharge of the evaporator, both of which are impractical and very expensive in most commercial systems. There are other ways of estimating the COP of a system by measuring certain internal conditions of the VCC cycle, involving measuring the actual refrigerant temperature and pressure, but these approaches can also be expensive and impractical for all but the most sophisticated HVAC&R systems.

[0314] HVAC&R systems are often characterized by equipment manufacturers to provide an indication of system performance relative to other similar equipment. For example, air conditioning systems are often given a "Seasonal Energy Efficiency Rating," or SEER rating, where a weighted average of the equipment's COP, measured under carefully controlled conditions in a laboratory under several predefined sets of indoor and outdoor environmental conditions, provides an indication of the equipment's expected efficiency and operating costs. While this rating can be useful in selecting one piece of equipment over another, it does not address equipment degradation, i.e., how the equipment performs under conditions it experiences "now" compared to when it was in a new, maintained condition. This measure is useful in determining when an HVAC&R system may require repair or maintenance.

[0315] To address this situation, the relative COP (or rCOP) can be defined as follows:

[0316]

[0317] Among them, COP measured The COP is the instantaneous COP measured or estimated from the conditions of the device and the current environment in which the device is located. ref is the reference COP, which is calculated when the equipment is in the new maintenance condition and placed in the same environment. For the purpose of detecting system degradation and its impact on the performance of a piece of HVAC&R equipment that has already been selected and installed, this is a more practical measure because it can show how inefficient it has become relative to the new maintenance condition. According to the definition in equation (27) above, an rCOP of 0.8 means that the equipment in its current physical condition and current operating environment is removing heat at 80% of the rate it should be in the new maintenance condition. Approximately, the system needs to consume about 1 / 0.8 = 1.25 times the energy to remove the same amount of heat from the conditioned space, which means that operating in the current conditions is about 25% more expensive than operating in the new maintenance conditions. Knowing this cost factor (calculated at 1.25) and the current usage rate, if the energy cost is known, the degree or extent of system degradation can be determined based on both the wasted energy and the cost.

[0318] The relative COP defined above using the classical definition of COP, while useful, suffers from many practical problems addressed by the embodiments herein. First, in order to be useful, the classical definition of relative COP requires the construction of a reference model of COP that can be used at the aforementioned inlet temperature (T ei ,T ci) operating within the entire expected operating range of the device. If the reference model is not provided by the device manufacturer, it must be learned on-site by some method. The relationship learner process 650 of the embodiments herein can be used to learn the actual COP of the system, but in most applications, the instrumentation required to measure the actual COP of the device is still very expensive. If regression or other typical machine learning techniques are used to create a mathematical model of the relative COP, there is an additional concern about whether the model adequately represents the current external operating conditions of the system.

[0319] An additional aspect of the embodiments herein is that the instrumentation is much simpler than the embodiments herein and provides a very useful relative COP proxy for the purpose of detecting degradation and estimating wasted energy and associated costs. These embodiments use the power parameter prediction of the CIPP processor 610 and the evaporator temperature drop value of the ETD processor 612 When both, and measured values ​​of P(n) and E(n) are available (i.e., obtained when the system is operating at refrigerant steady state and thermal steady state as described above), since the relationship learner 650 learns to quickly and accurately predict these residual sequences, a system based on these residual sequences can quickly provide an approximate rCOP estimate and will also know when the estimate is likely to be close to the rCOP value and when it is likely not, rejecting estimates when the confidence level is low.

[0320] The rCOP processor 613 is operable to provide a sequence of approximate relative COP values ​​rCOP(n) to the degradation detection processor 614 using the following equation:

[0321]

[0322] For those quantities corresponding to the nth observation, and The output of the rCOP processor 613 (designated as the sequence rCOP(n) calculated above) is fed directly to the degradation detection processor 614.

[0323] For background, the form of the rCOP defined in equation (28) above can be explained as follows. The derivation of the rCOP approximation begins with a version of equation (27) defining the coefficient of performance, denoted as COPC, that is more suitable for the instrumentation commonly available at this time:

[0324]

[0325] Among them, W c In this more general definition, the latent heat h in Eq. (26) is neglected.l , only considering the sensible heat h s The effect of neglecting latent heat will be discussed below.

[0326] By the above assumptions, the evaporator removes sensible heat h from the air. s The rate is given by:

[0327]

[0328] Where E is the temperature drop measured across the evaporator coil, is the mass air flow rate over the evaporator, and Cpe is the specific heat of the fluid flowing through the evaporator. If each of these parameters is assumed to be constant, the above Figure 2 The air mass flow rate mentioned in and specific heat C pe Combined into a single COPC constant K copc :

[0329]

[0330] And the equation for COPC becomes:

[0331]

[0332] As we will see later in this article, the COPC constant K copc Useful for deriving relative COP.

[0333] By timely measuring the temperature drop E on the evaporator m (n) and observe the compressor power W at n mc (n), the same constant K can be used copc To calculate the instantaneous measurement COPC of the system, that is, COPC m :

[0334]

[0335] Now, using this same definition, assume that there is a reference COPC value, COPC r , which means that if the equipment is operated under the new maintenance conditions, the measured temperature (T ei (n),T ci What should the COPC be for the current observation of (n)? The relative COP of this observation, rCOP(n), can be calculated based on COPC m (n) and COPC r (n) is written as:

[0336]

[0337] In the same manner as in equation (33) above, reference COPC r (n) may take the following forms:

[0338]

[0339] Among them, W rc (n) is a theoretical reference electric power value in some embodiments, but is not defined but will be eliminated from discussion later, and E r (n) is the corresponding reference evaporator temperature drop. Substituting the equation for the measured COPC (Equation (33)) and the equation for the reference COPC (Equation (35)) into the equation for the relative COP (Equation (34)) and arranging the terms gives:

[0340]

[0341] Under the above assumptions of constant line voltage and power factor, the following relationship holds for any two measurements or estimates of power W c1 and W c2 And the corresponding power parameters (such as power, current, volt-ampere, etc. P1 and P2) are completely established:

[0342]

[0343] Substituting equation (37) into the above equation (36) yields:

[0344]

[0345] Among them, P rc (n) corresponds to W in equation (36) rc (n) is the value of the power parameter, and P mc (n is the value corresponding to W in equation (36) mc (n) The value of the measured power parameter. Figure 6A , for the nth enhanced observation provided by the VCC state generator 608, for the nth enhanced observation O a (n), the term E in equation (38) m (n),E r (n),P rc (n) and P mc (n) immediately from Figure 6A Identify them as E(n), and P(n). When these values ​​are substituted into equation (38) above, equation (28) immediately follows. Therefore, the rCOP processor 613 performs the following operations for which and For each observation where both are not null, the above equation (28) is evaluated to compute the sequence rCOP(n), where or Observations that are both assigned a non-null value will have the value null assigned to rCOP(n).

[0346] Several explicit assumptions are made in the calculation of rCOP given by equation (28). First, the mass air flow rate is roughly constant at a given evaporator inlet temperature. In addition, the air flowing through the evaporator is "dry", meaning that the latent heat component of heat removal from the air is dominant. This is equivalent to saying that all heat removed from the air is sensible (i.e., can be sensed). In addition, the air is modeled as having a specific heat C pe The refrigerant is an ideal gas and is dry enough that the latent heat involved in the condensation of water on the evaporator coil is insignificant. Finally, the line voltage and compressor power factor are approximately constant in all cases, and the equipment operates in a quasi-steady state, meaning that the refrigerant is in the correct state everywhere in the refrigerant circuit and that condenser and evaporator temperature transients have subsided.

[0347] In the real world, one or more of the above assumptions may not hold true, such as the assumption of a constant mass flow rate through the evaporator, which may be greatly affected by a dirty air filter that is typically inserted in line with the evaporator inlet to prevent particulate matter from fouling the evaporator to a significant extent, and by the fact that moisture entering the evaporator inlet typically causes condensation, reducing the humidity at the exhaust port and thus reducing the density. However, despite any real-world limitations, the embodiments of the present disclosure presented herein have proven to be both practical and beneficial.

[0348] Mass air flow rate across the evaporator with a dirty air filter The cooler evaporator causes the measured evaporator temperature to drop more than expected. Second, the cooler evaporator can cause the pressure in the evaporator to drop, thereby reducing the power required to move the refrigerant through the evaporator. The increase in evaporator temperature drop combined with the reduction in compressor power results in rCOP values ​​given by Equation (28) that are greater than 1.0 or 100%. Therefore, the rCOP value can be confusing unless the condition manifests itself in a combination of a positive evaporator normalized residual and a negative power parameter residual. In this case, the rCOP value is questionable, but the combination can be used to infer a dirty filter condition and / or other type of air flow obstruction and issue an appropriate alarm signal. Simply replacing the dirty filter with a clean filter can eliminate the condition, exposing a more accurate rCOP value.

[0349] Next go to Figure 12A and Figure 12B, further techniques for calculating relative COP in addition to (or as an alternative to) the techniques described so far are now described. As discussed, using the relationship builder 652 ( Figure 6E ) The CIPP relationship 500 and the ETD relationship 506 learned above ( Figure 5A and 5B ) temperature map 654( Figure 6E ) represents the compressor current and evaporator temperature drop of the HVAC&R system under the new maintenance conditions. In some embodiments, the information contained in these corresponding CIPP and ETD temperature maps 654 can be combined in novel ways to model a virtual HVAC&R system based on the actual HVAC&R system under the new maintenance conditions, but which can predict "correct" operation under conditions other than the new maintenance conditions, thereby constructing a surrogate normalized residual and relative COP "score." This model of the virtual HVAC&R system can be software-based, such as a virtual HVAC&R system executed on a network or cloud computing system.

[0350] In the form presented so far, the nth amplification observation O a (n) Temperature tuple (T ei (n),T ci (n)) is used as an index element of the temperature map, thereby generating the predicted power parameters as described above and the predicted evaporator temperature drop Each parameter represents the operation of the physical HVAC&R system under the new maintenance conditions. A virtual HVAC&R system that can be constructed according to the subsequent teachings of this article allows the observed triples (T ei (n),T ci (n), P(n)) to construct a new evaporator temperature drop prediction

[0351]

[0352] Equation (39) is based on the physical system under the new maintenance condition and is suitable when the observed power parameter P(n) is exactly the learned value of the physical system under the new maintenance condition. When the same (approximate) value is predicted according to equation (6) and the teaching provided above But it allows predictions that differ from the case when P(n) is not a learned value of the physical system. The equivalent evaporator temperature drop value is as if for a given observation tuple (T ei (n),T ci(n)), the system can be operated "normally" at different power parameter values, in particular observing the power parameter value of P(n), where the expected (or predicted) temperature drop is given by equation (39). In some embodiments, the function g E (T ei (n),T ci (n), P(n)) is a parametric predictor whose coefficients are uniquely determined for a particular observation by the contents of the above CIPP temperature map and ETD temperature map in a manner to be discussed.

[0353] According to this new evaporator temperature drop, it can be determined by noting that in this case P(n) and The observed and predicted values ​​are equal and the calculated value Substitution To construct an alternative relative COP value (designated as rCOP) from equation (28) above E (n)), we get:

[0354]

[0355] In a similar manner, a second virtual HVAC&R system can be constructed according to the subsequent teachings herein, which allows for the observation of triples (T ei (n),T ci (n),E(n)) to construct a new power parameter prediction Symbolically:

[0356]

[0357] Similar to equation (39) above, equation (41) is based on the physical system at the new holding condition and is adapted so that the observed evaporator temperature drop E(n) is exactly the learned value of the physical system at the new holding condition. According to equation (1) and the above teaching, the same (approximate) value is predicted when But it allows predictions that differ from when E(n) is not exactly the learned value of the physical system. The equivalent power parameter value is as if for a given tuple (T ei ,T ci ), the system can operate "normally" at different values ​​of evaporator temperature drop, in particular the observed E(n), with the resulting expected (or predicted) power parameter value given by equation (41). In some embodiments, the function g P (T ei (n),T ci (n), E(n)) are parametric predictors whose coefficients are uniquely determined for a particular observation by the contents of the above CIPP temperature map and the ETD temperature map in a manner to be discussed.

[0358] In this case, we can do this by noting that in this case E(n) and The observed and predicted values ​​are equal and the calculated value Substitution From equation (28) above, we construct the COP which is designated as rCOP P (n) by replacing the relative COP value, we obtain:

[0359]

[0360] Figure 12A and Figure 12B The above alternative techniques for calculating relative COP using a virtual HVAC&R system are conceptually illustrated. These diagrams are similar to those of a physical HVAC system. Figure 5A and 5B The corresponding figure in the figure, except that the compressor power parameter P(k) and the evaporator temperature drop E(k) are used to calculate the prediction and the evaporator inlet fluid temperature and condenser inlet fluid temperature tuples (T ei (k),T ci (k)). For example, the triple (T ei (k),T ci (k), E(k)) may be provided to the joint CIPP and ETD relation block 500'( Figure 12A ) to predict the compressor power parameters, and the triplet (T ei (k),T ci (k), E(k)) may be provided to the joint CIPP and ETD relation block 506'( Figure 12B ) to predict the evaporator temperature drop. Each of these joint CIPP and ETD relationship blocks 500' and 506' is similar in form and function to its counterparts 500 and 506 ( Figure 5A and 5B ) in the same manner to learn the observed temperature tuple (T ei (n),T ci (n)) and the corresponding power parameter and evaporator temperature drop, and the temperature diagram of the CIPP relationship and the ETD relationship 654 is also identical in form and function. The learned CIPP and ETD relationships can then be used in conjunction with the observed temperature drop E(n) to generate new power parameter predictions and its corresponding normalized residual, and are applied together with the observed power parameter value P(n) to generate a new evaporator temperature drop prediction and its corresponding normalized residual, to be consistent with the above Figure 5A and 5BThe differences between the learned CIPP relation 500 and the learned CIPP relation 506 and their corresponding joint CIPP relation 500′ and joint ETD relation 506′ are: (a) the joint CIPP relation 500′ and the joint ETD relation 506′ respectively use the CIPP temperature map and the ETD temperature map 654, and (b) the observed temperature tuple (T ei (n),T ci (n)) the manner in which the neighborhood is extracted, and (c) the form of the parametric predictor used to compute the predicted value.

[0361] Figure 13 1 shows an exemplary embodiment of a prediction processor 606' suitable for use by or in an HVAC&R monitoring agent 314 to monitor an alternative technique for a virtual HVAC&R system. Generally, the prediction processor 606' accepts observations O(k) from the data acquisition processor 600 and can optionally use the observations to learn various CIPP and ETD relationships in the same manner as described above. The prediction processor 606' can then generate a normalized power parameter residual sequence and a normalized ETD residual sequence, which are presented to the degradation detection processor 614 for analysis. As can be seen, the prediction processor 606' is similar to its counterpart ( Figure 6A The prediction processor 606 in FIG. 1 is similar in that there is a VCC state generator 608′ having the same form and function as 608, which can receive and accept the observation sequence O(k) from the data acquisition processor 600 and enhance the sequence with system state information to generate an enhanced observation sequence O(k). a (k).

[0362] However, in the example shown, the prediction processor 606' includes a joint CIPP / ETD processor 610' rather than a separate CIPP processor and a separate ETD processor. The joint CIPP / ETD processor 610' can then be used to learn the various CIPP and ETD relationships mentioned above and use the resulting temperature map in conjunction to generate new power parameter predictions. and the corresponding normalized residuals, and the new evaporator temperature drop prediction and the corresponding normalized residuals. The normalized residuals are then provided to the degradation detector processor 614 of the HVAC&R monitoring agent 314 for use in detecting degradation in the manner described above. and / or new evaporator temperature drop prediction The measured values ​​of E(n) and P(n) are provided to the rCOP processor 613' for generating a power parameter-derived relative COP (designated as rCOP P(n)) and / or ETD-derived relative COP (designated as rCOP E (n)), as will be described later.

[0363] Figure 14 An exemplary embodiment of a joint relationship learner 650' is shown, which may be used by or in conjunction with the joint CIPP / ETD processor 610' to learn the CIPP and ETD relationship and predict from both. and The joint relation learner 650′ operates in a similar manner to its counterparts from Figure 6E The (single) relationship learner 650 and CIPP relationship builder 652′, CIPP temperature map 654′, ETD relationship builder 652″ and ETD temperature map 654″ are identical in form and function to their “single relationship learner” counterparts and thus may also be used by the prediction processor 606 ( Figure 6A ) is used by the CIPP processor 610 or the ETD processor 612. However, while the (single) relational learner 650 has a neighborhood extractor 656 operating on a single temperature map, the joint relational learner 650′ includes a joint neighborhood extractor 656′ that operates on two temperature maps simultaneously to extract a set of tuples N′(n) from the neighborhood where N′(n) is the number of tuples in the neighborhood of the observation tuple (T ei (n),T ci (n)) in both the CIPP and ETD temperature maps within a defined neighborhood, where corresponding cells in both the CIPP and ETD temperature maps are "observed" as described above—these tuples and their corresponding cells are defined herein as "joint observations." The joint neighborhood extractor 656' performs the same test as the single neighborhood extractor, but on the set of jointly observed tuples, requiring a certain minimum number of jointly observed tuples within the defined neighborhood, the required minimum number then being consistent with the requirements of the parameterized CIPP predictor 658' and the parameterized ETD predictor 658", and the observed temperature tuples (T ei (n),T ci (n)) lies in the convex hull of the subset of set N'(n) that does not include the temperature tuple (T ei (n),T ci (n)), if it is a member. As above, if either test fails, the joint neighborhood extractor provides a null value for N'(n).

[0364] Again in a manner similar to the relational learner 650, the resulting set N'(n) provides input to a parameterized CIPP predictor 658' and a separate parameterized ETD predictor 658" and operates to perform power parameter predictions It extracts summary data from the cells corresponding to the tuples in N'(n) to create a parameterized model for evaporator temperature drop prediction. The parameterized predictors 658' and 658" are different in form from their counterpart 658 above. In a single relational learner 650, and The parameter model 658 is simply the measured temperature tuple (T ei (n),T ci (n)), the parameter model of the parameterized CIPP predictor 658' is constructed to calculate the triplet (T ei (n),T ci (n),E(n)), as if the measured value of E(n) represents the operation of the system under the new maintenance conditions and is independent of (T ei (n),T ci (n)). Similarly, the parametric model of the parameterized ETD predictor 658" is constructed to calculate different estimates of the evaporator temperature drop, As a measurement triplet (T ei (n),T ci (n), E(n)), as if the measured value of P(n) represents the operation of the system under the new maintenance conditions and is independent of (T ei (n),T ci (n)).

[0365] A dashed line 659 is placed around the joint neighborhood extractor 656′, and a parameterized CIPP predictor 658′ and a parameterized ETD predictor 658″ are placed around these elements to indicate that they each have access to observations when performing their functions. a (n).

[0366] Figure 15A A flowchart 1500 is shown illustrating an exemplary process that may be used by or with the joint neighborhood extractor 656′ to determine whether to make a prediction and to provide a set N′(n) of temperature tuples pointing to cells jointly observed in the CIPP temperature map 654′ and the ETD temperature map 654″ that are sufficient to construct a local parametric model to predict the parameter in question via the parameterized CIPP predictor 658′ or the parameterized ETD predictor 658″ or both, as appropriate. Referring first to Figure 15A , flowchart 1500 generally begins at 1501, where the joint neighborhood extractor 656' receives or is presented with an augmented observation (i.e., the nth observation of the sequence) provided by the VCC state generator 608' as a tuple with an observation temperature (T ei (n),T ci (n))O a(n) The joint neighborhood extractor 656' operates on the individual augmented observations received from the VCC state generator 608', either one at a time as they are generated, or continuously across a frame of data.

[0367] In some embodiments, the joint neighborhood extractor 656' may simply ignore any observations from the VCC state generator 608' that do not meet the criteria for a steady-state observation of the power parameter and evaporator temperature drop for a given compressor in the on state, a condition previously described as an "active" relationship learner 650'. Thus, at 1502, the neighborhood extractor 656' determines whether the joint relationship learner 650' is active, as defined by the following conditions: a) the compressor to which the joint relationship learner is applied is on (indicated by Sc(n) = true in a single compressor system or Sc(n) > 0 in a multi-compressor system), b) the observation from the VCC state generator 608 is taken when the HVAC&R system is in a steady state with respect to the power parameter prediction, indicated by Sp(n) = true, and (c) the observation is taken when the HVAC&R system is in a steady state with respect to the evaporator temperature drop, i.e., Se(n) = true. If the joint relation learner 650' is not active in 1502, the neighborhood extractor 656' immediately assigns the value NULL to the set N'(n) for the observation in processing step 1505, and processing of the observation is complete, with the value NULL indicating that no prediction should be made for the observation.

[0368] Assuming that the joint relationship learner 650' is determined to be active in step 1502, then in step 1503, the joint neighborhood extractor 656' is activated at T ei and T ci The observed temperature tuple (T ei (n),T ci (n)) within a "neighborhood" of + / - γ degrees. So for example, if a temperature tuple (T ei (n),T ci (n)), then the joint neighborhood extractor 656′ searches for all temperature map cells (points) in both the CIPP relationship temperature map 654′ and the ETD relationship temperature map 654″ that satisfy both equations (43) and (44):

[0369] T ei -γ≤T ei (n)≤T ei +γ (43)

[0370] T ci -γ≤T ci (n)≤T ci (n)+γ (44)

[0371] The neighborhood in which this search occurs, specified herein by parameter γ, is typically chosen to be larger than the neighborhood used to determine the predicted power parameters and evaporator temperature drop predictions for rCOP realization described above, which is specified by parameter δ with a typical value of γ of approximately 1-2 degrees Celsius.

[0372] For the above search in step 1503, the joint neighborhood extractor 656' considers only those cells within the neighborhood established above for which both the CIPP relationship temperature map 654' and the ETD relationship temperature map 654" cells are designated as "observations," i.e., cells for which, in some embodiments, the "observation" metadata variable has been set to true, as described above or otherwise tested for that condition. Such cells are referred to as "jointly observed" cells. Each time the joint neighborhood extractor 656' finds a joint observation cell within the above neighborhood, it replaces the corresponding temperature tuple (T ei (n),T ci (n)) is appended to the initial empty set N'(n). Set N'(n) is the result of processing step 1503 and can be used for subsequent processing.

[0373] Based on the contents of the set N'(n), the joint neighborhood extractor 656' then allows (or recommends) making a prediction if and only if two criteria are met. First, a certain absolute minimum number of observed cells is mathematically required to determine the parameter coefficients of the CIPP parameterized predictor 658' and the ETD parameterized predictor 658", but a larger number of observed cells can be used and is preferred. Therefore, the minimum number of cells, N'min, is determined by a system-dependent predefined constant, which must be greater than the absolute minimum number of tuples required by the parameterized predictors 658' and 658", with larger numbers being preferred. In some embodiments, the parameterized predictors 658' and 658" require an absolute minimum number of 4 observed cells, and in those embodiments, N'min can be set to 8 cells.

[0374] To ensure that the first requirement is met, when the search is completed in processing step 1503, the set N'(n) contains a number of tuples represented as Size(N'(n)). In decision step 1504, a test is performed to determine whether the number of tuples in the set N'(n) is greater than or equal to the minimum number defined by the predefined constant N'min described above. If this criterion is not met, the set of temperature tuples N'(n) is assigned the value null at 1505, and the work of the joint neighborhood extractor 656' is completed for this observation.

[0375] If the joint neighborhood extractor 656′ finds enough temperature tuples in N′(n) at 1504, the joint neighborhood extractor 656′ proceeds to 1506 to test the second criterion required for non-null prediction in the present invention, namely, the number of observed temperature tuples (Tei (n),T ci (n)) lies within the convex hull formed by the set of temperature tuples represented by N'(n) collected as described above, where if a particular point (T ei (n),T ci (n)) is a member of N'(n), then that particular point is excluded from the test. If it is determined in 1506 that the tuple of the observation does not lie within the convex hull of the tuples of the cell determined above at 1504, then at 1505 the table for the tuple N'(n) is assigned a null value and the process is completed for that observation.

[0376] If the joint neighborhood extractor 656' determines at 1506 that the observed temperature tuple is within the convex hull of the minimum number of temperature tuples determined as described above, this can significantly improve the reliability of the prediction compared to prior art solutions. If both criteria at 1504 and 1506 are met, the joint neighborhood extractor 656 provides the table of tuples N'(n) found above to the CIPP parameterized predictor 658' or the ETD parameterized predictor 658", or both, as described below with respect to Figure 15B Flowchart discussion.

[0377] Figure 15B The flowchart of FIG. 6 illustrates the predictions performed by the parameterized CIPP predictor 658′. or the predictions of the parameterized ETD predictor 658 (or both) processes. Figure 14 It can be noted that the same set N'(n) of temperature tuples computed by the joint neighborhood extractor 656' is applied to the parameterized CIPP predictor 658' and the parameterized ETD predictor 658", and that a The flowchart 1570 starts at 1571, where the parameterized CIPP predictor 658' or the parameterized ETD predictor 658" (or both) receives the set of temperature tuples N'(n) and the corresponding enhanced observations O provided by the joint neighborhood extractor 656' described above. a (n) (or observation sequence). Step 1572 determines whether N'(n) has been set to null by the joint neighborhood extractor 656'. If so, then in step 1573 or (or both) are assigned the value NULL, and no prediction is returned for the given observation.

[0378] If it is determined in step 1572 that the set of temperature tuples N'(n) is not empty, then in process step 1574, a joint value table is constructed based on the summary data in the cells of the CIPP relationship temperature map 654' and the ETD relationship temperature map 654". The value table includes the T value of each tuple in N'(n). ei ,Tci The values ​​of , the average values ​​of the power parameters from the CIPP temperature map 654', and the average values ​​of the ETD from the ETD temperature map 654" are used as rows, and have been determined to be tuples whose corresponding cells in the corresponding temperature maps meet the above-mentioned criteria for "observed" cells. The two average values ​​are calculated from the summary data in the cells using equation (14). Assuming that there are m' temperature tuples in the temperature map N'(n), the resulting value table is described by Table 6:

[0379]

[0380] Table 6: Table extracted from the joint ClPP and ETD temperature maps

[0381] With Table 6 populated appropriately, in step 1575, the parameter coefficients for the parametric CIPP predictor 658′ or the parametric ETD predictor 658″, or both, may be determined. In some embodiments, the form of this parametric model for the parameterized CIPP predictor 658′ is a simple hyperplane of the form:

[0382]

[0383] Among them, P0,K pei ,K pci and K e is specific to the current observation O a (n) Parameter coefficients of the determined predictor function, where in some embodiments the values ​​are calculated from the values ​​in Table 6 using an optimization program (such as scipy.optimize.lsq_linear developed for the Python programming language) or many equivalent packages in other programming languages.

[0384] In some embodiments, the parametric model of the parameterized ETD predictor 658" is in the form of a simple hyperplane of the form:

[0385]

[0386] Among them, E0,K eei ,K eci and K p Is to focus on the current observation O a (n) Parameter coefficients of the determined predictor function, where in some embodiments the coefficient values ​​are calculated from the values ​​in Table 6 using an optimization program (such as scipy.optimize.lsq_linear developed for the Python programming language) or many equivalent packages in other programming languages.

[0387] When applied according to the above, these optimization procedures select the parameters P0,K in equation (45) pei ,K pciand K e Or E0,K in equation (46) eei ,K eci and K p The value of , which provides a "best fit" to the data in Table 5. Many optimization programs allow the values ​​of the parameters to be selected as needed to better represent the thermodynamics of the system. An example of this is when the power parameter is predicted using equation (45) above, the parameter K pei and K pci can be constrained to be non-negative to reflect that any increase in temperature should result in an increase in compressor power.

[0388] It should be recognized that when necessary and In both systems, the coefficients for both predictors can be calculated from the same table constructed in step 1575 and symbolically represented by Table 6 above, and there is no need to construct separate tables.

[0389] Once the parameter coefficients for the parametric model of the parameterized CIPP predictor 658′ or the parameterized ETD predictor 658″ (or both) are determined in step 1575, the parameter coefficients for the parametric model of the parameterized CIPP predictor 658′ or the parameterized ETD predictor 658″ (or both) can be determined by applying the observation a To determine the appropriate measured values ​​of (n) via the parameterized CIPP predictor 658′ In step 1576, in the enhanced observation O a The value of (n) T ei (n),T ci Equation (45) with the parameter coefficients determined above is evaluated at E(n) and E(n). Similarly, to determine via the parameterized ETD predictor 658″ In step 1576, in the enhanced observation O a The value of (n) T ei (n),T ci Equation (44) with the parameter coefficients determined according to above is evaluated at P(n) and P(n).

[0390] Return Reference Figure 13 , once the prediction is made using the above method The predictions can be applied together with the observed values ​​of P(n), as Figure 12A As described above, by using the above equations (2) and (3) Replacement amount and using equation (42) to calculate the surrogate relative COP "score" rCOPP(n) for observation.

[0391] Similarly, once predictions are made using the above method The prediction can be applied together with the observed values ​​of E(n), as Figure 12B is described in the above equations (7) and (8). Replacement amount and the observed surrogate relative COP “fraction” rCOP using equation (40) E (n) to calculate the alternative normalized residual

[0392] Power parameter prediction sequence Typically the same predictions as those generated by the CIPP processor 610 described previously However, the degradation detection processor 614 can be used to detect the degradation of the P (n) The same method is used to monitor the limit detector 690 based on P * (n), The sequence of normalized power parameter residuals can be replaced by R P (n) or in addition thereto, wherein the resulting filtered sequence includes Figure 6H The message Msg(n) is in the format of

[0393] Similarly, the evaporator temperature drop sequence Typically the same predictions as those generated by the ETD processor 612 described previously However, the degradation detection processor 614 can be used to detect the degradation of the E (n) The same method is used to monitor the limit detector 690 based on the The series of normalized evaporator temperature drop residuals can be substituted for R E (n) or in addition thereto, wherein the resulting filtered sequence includes Figure 6H The message Msg(n) is in the format of

[0394] Similarly, the sequence rCOP P (n) and rCOP E (n) has been used in the same way as rCOP(n) above to estimate the cost of the degradation observed in the system. In some practical applications of the present disclosure, when the normalized power parameter residual R P When (n) is greater than zero, the sequence rCOP P (n) has been used to indicate wasted power and degradation costs, thereby indicating that the system is using more power than a newly maintained system, and when the normalized power parameter residual R P When (n) is less than or equal to zero, rCOP E (n) has been used. In addition, the relative COP sequence rCOP P (n) or rCOP E(n) or both may be monitored by the degradation detection processor 614 in the same manner as rCOP(n) using a limit detector for each sequence monitored at the appropriate limit to generate an alarm, and the resulting filtered sequence included in Figure 6H The message Msg(n) is in the format of

[0395] Next, refer to Figure 16 , shows a more general system parameter monitoring agent 1602 that can be used with other types of systems indicated at 1600 in addition to the HVAC&R systems described herein. As mentioned at the outset, the principles and teachings discussed herein are applicable to any deterministic system or device where, for a given parameter of interest, a certain parameter result or value will be consistently produced and, therefore, can be rapidly learned and predicted as described herein, given an index parameter or set of index parameters (and their values). Examples of parameters that can be used as parameters of interest and index parameters include flow control parameters (e.g., flow rate, viscosity, etc.), power control parameters (e.g., voltage, current, etc.), motion control parameters (e.g., speed, altitude, etc.), etc., and combinations thereof.

[0396] according to Figure 16 , the agent 1602 has similar functional components to the agents discussed previously, including a data acquisition processor 1604, a prediction processor 1614, and a degradation detection processor 1622 (and their respective subcomponents). The data acquisition processor 1604 operates to continuously acquire and store observations of parameters to be used as index parameters (indicated at 1610) and parameters of interest (indicated at 1612). These observations 1610, 1612 can be acquired in real time using appropriate sensors that measure these parameters, or can be obtained from a database of these observations, or a combination of both. Based on these observations 1610, 1612, the data acquisition processor 1604 assembles a time series of observations that can be used by the prediction processor 1614. The prediction processor 1614 operates to derive certain operational information from the time series of observations and selectively uses the observations to learn the relationship between the index parameters 1610 and the parameters of interest 1612. Thereafter, the prediction processor 1614 uses the learned relationship along with the observations to generate a time series of normalized residuals that contains information about the physical condition of the system 1600. This series of normalized residuals is passed to a degradation detection processor 1622, which interprets the time series of normalized residuals and may issue a warning signal or audio-visual display, or send information via the news feed 616 indicating a potential problem with the system 1600.

[0397] Table 7 below shows exemplary observations that may be provided by the data acquisition processor 1604 to the prediction processor 1614. In the table, the exemplary observations contain several parameters that may be used as indices 1610, including index parameter 1, index parameter 2, and so on, for a parameter of interest 1612, up to index parameter i. Consider an example in the context of HVAC&R, where compressor input power is a function of condenser inlet temperature, evaporator inlet temperature, and evaporator discharge temperature. Such an HVAC&R system would have a temperature map with three index parameters, namely the three temperatures mentioned, rather than the two index parameters discussed above. These index parameters and the parameter of interest, or more precisely their values, may be obtained from appropriate sensors strategically positioned to measure these values. Alternatively, a proxy may be used for one or more of these parameters, rather than measuring them directly. In some embodiments, an optional timestamp or tag indicating the date and time or interval represented by the measured parameter may be included in the observation.

[0398]

[0399] Table 7: Exemplary Observations

[0400] The time series of observations is forwarded from the data acquisition processor 1604 to the prediction processor 1614, either one at a time or in batches of data frames as described above. According to the disclosed embodiments, the prediction processor 1614 is operable to derive or learn a relationship between an index parameter and a parameter of interest and use this relationship to monitor performance degradation of the system 1600 from the observations provided by the data acquisition processor 1604. In some embodiments, the prediction processor 1614 includes a system state generator 1616 that operates to derive certain timing information from the observation sequence provided by the data acquisition processor 1604 and enhance the observations with this information to produce a steady-state observation sequence. A parameter relationship processor 1618 is provided to learn a relationship from the enhanced time series of steady-state observations provided by the system state generator 1616.

[0401] Also included is a degraded residual sequence generator 1620 that uses the learned relationship and the time series of steady-state observations to calculate a time series of normalized residuals, labeled as a degraded residual sequence, that indicates the condition of the system 1600. It should be understood that the version of the degraded residual sequence generator 1620 herein is only one embodiment. In general, the degraded residual sequence generator 1620 or its underlying principles and teachings can be used with any system 1600 in which there is a fixed, known, or learnable relationship "form" between the residuals and a set of index parameters.

[0402] The degradation residual sequence generated by the degradation residual sequence generator 1620 may then be provided to the degradation detection processor 1622. The degradation detection processor 1622 thereafter operates to analyze the degradation residual sequence generated by the degradation residual sequence generator 1620 to detect and report degradation.

[0403] As discussed, predictions of parameters of interest using the embodiments described herein are most accurate after the system has operated for a sufficient period of time to stabilize with respect to the parameters of interest, a time period that may vary depending on the device. To this end, system state generator 1616 may use appropriate logic or circuitry to detect whether the system has stabilized with respect to the parameters of interest and is in a stable state, and therefore likely stable, or is in a transient state and therefore likely unstable. The system state generator may then declare the system to be stable or unstable for relational purposes. In some embodiments, system state generator 1616 may augment observations obtained from data acquisition processor 1604 with system state information in the form of Boolean variables. The Boolean variables may take values ​​from the set {true, false} to represent the system state. VCC state generator 608 may set the Boolean variables to true to indicate that the system is stable and in the on state, respectively, as described above, and to false to indicate otherwise. In some embodiments, agent 1602 may associate system state information, such as that referenced above, with each observation to generate augmented observations.

[0404] The parameter relationship processor 1618 is responsible for learning the relationship between the values ​​of the index parameters 1610 and the parameters of interest 1612 from the steady-state observations described above. The parameter relationship processor 1618 includes three main functions that provide the capabilities required to build relationships that represent the system 1600 under new maintenance conditions. In some embodiments, the parameter relationship processor 1618 compiles and maintains a parameter map, similar to the temperature map discussed above, that relates the index parameters 1610 to the parameters of interest 1612. In some embodiments, a guided learning strategy similar to that discussed herein can be used in combination with a reference degradation estimator function to, in some cases, modify the parameter of interest values ​​of the steady-state observations before using the modified observations to populate the parameter map.

[0405] In some embodiments, the agent 1602 constructs a parameter map using steady-state observations provided by the system state generator 1616, each steady-state observation comprising at least an index parameter or set of index parameters and a corresponding parameter of interest. Each index parameter or set of index parameters forms an index into the parameter map of the parameter of interest, and the agent 1602 "learns" by updating the summary data for a cell from the parameter of interest values ​​corresponding to the steady-state observations of the index parameter value. The agent 1602 updates the summary data for a given cell in this manner until a sufficient number of observations have been applied, as described above. At this point, the agent stops updating the summary data for that cell, and the summary data for the cell can be used to predict the parameter of interest value representing the system under the new maintenance condition. In some cases, once the requisite number of observations have been made for that cell, a parameter value prediction can be derived directly from the summary data for a single cell indexed by a set of steady-state observations of the index parameter. In other cases, as described herein, the agent can derive a power parameter prediction for a set of steady-state observations of the index parameter by performing a local regression using summary data from nearby values.

[0406] Using the above approach, the agent can quickly collect data and begin making parameter value predictions almost immediately, assuming the system is operating and in a new maintenance condition. Using the parameter graphs described herein, the agent can evaluate whether a prediction of a parameter value corresponding to a given index parameter or set of index parameters is likely to represent the characteristics of the system under the new maintenance condition and decide whether to issue a prediction. The ability to assess the reliability of a prediction beneficially reduces the likelihood of the agent issuing false positives and false negatives. Furthermore, because in some systems the relationship can be assumed to be quasi-independent of the index parameters, the agent can continue to learn the characteristics of the system under the new maintenance condition as the system degrades, thereby compensating for the degradation and making the prediction better representative of the system under the new maintenance condition.

[0407] Furthermore, continuous learning of the agent's relationship can be achieved by updating the parameter map as additional observations of the index parameters and corresponding parameter data of interest become available. And as discussed, in some embodiments, the parameter map can be updated in batches, whereby a set of observations are assembled into one or more data frames of steady-state observations and presented as a batch of observations by the data acquisition processor 1604 to the agent's prediction processor 1614. Of course, in some embodiments, observations can also be provided on an individual observation basis, one at a time as they are received.

[0408] A partial example of an exemplary parameter map is shown below in Table 8, where the cells of the map contain the summary values ​​of the parameter of interest observed for each temperature parameter index. Although the table is shown as mostly populated, in general, only those parameters for which T ei and T ci The cells with the value of will contain the summary value.

[0409]

[0410] Table 8: Example parameter diagram

[0411] As previously described, each cell in the parameter map (e.g., C00, C01, C02, etc.) contains a summary value for the observation corresponding to the index value used as the cell index (e.g., IV0, IV1, IV2, etc.). These summary values, or summary statistics (or sample statistics), provide summary information about the steady-state observation represented by the cell. As an example, the summary values ​​can provide information about the data in the data set, such as the sum, mean, median, average, variance, deviation, distribution, etc. The agent can then use these summary values ​​to generate predictions for the parameters of interest as described above.

[0412] The prediction is then provided to the agent's degradation residual sequence generator 1620 to create a degradation residual sequence for each steady-state observation. This degradation residual sequence serves as input to a degradation detection processor 1622, which is configured to analyze the degradation detection sequence in a manner similar to that discussed above. The degradation detection processor 1622 monitors the sequence of degradation residuals and, in response to detecting a potential problem via the degradation residual sequence, issues a warning signal and / or audio / visual display or news feed, typically indicated at 1624.

[0413] While particular aspects, embodiments, and applications of the present disclosure have been shown and described, it should be understood that the disclosure is not limited to the precise construction and composition disclosed herein and that various modifications, changes, and variations may be apparent from the foregoing description without departing from the scope of the invention as defined in the appended claims.

Claims

1. A monitoring system for an HVAC&R system, the monitoring system comprising: at least one processor; A storage device coupled to the at least one processor and having stored thereon processor-executable instructions, the instructions comprising instructions that, when executed by the at least one processor, cause the at least one processor to instantiate: a data acquisition processor operable to acquire observations about the HVAC&R system, the observations comprising a condenser fluid temperature measurement and an evaporator fluid temperature measurement, the observations also comprising compressor input power parameter measurements corresponding to the fluid temperature measurements; a relationship builder operable to learn a compressor input power parameter (CIPP) relationship between the fluid temperature measurements of the evaporator entry temperature and the condenser entry temperature and the compressor input power parameter measurements, and further operable to learn an evaporator temperature drop (ETD) relationship between the fluid temperature measurements of the evaporator entry temperature and the condenser entry temperature and the evaporator temperature drop; as well as a temperature map comprising a plurality of cells, each cell corresponding to a temperature tuple consisting of a condenser entry temperature and an evaporator entry temperature, the temperature map being configured to receive from the relationship builder and store, for each cell, summary statistics of a measured compressor input power parameter or a measured-derived evaporator temperature drop, or both, corresponding to the temperature tuple of the cell; The processor-executable instructions further cause the at least one processor to calculate a predicted value of a compressor input power parameter and / or a predicted value of an evaporator temperature drop using the CIPP relationship or the ETD relationship, respectively, and to declare that performance degradation exists in the HVAC&R system using the predicted value of the compressor input power parameter or the predicted value of the evaporator temperature drop, or both.

2. The system according to claim 1, wherein: The processor-executable instructions further cause at least one processor to shut off power to the HVAC&R system in response to declaring that performance degradation exists in the HVAC&R system.

3. The system according to claim 1, wherein: The relationship builder uses a machine learning based learning process to learn the CIPP relationship and the ETD relationship respectively.

4. The system of claim 1 , further comprising a neighborhood extractor operable to define, for a given temperature tuple and its elements, a temperature tuple range surrounding the given temperature tuple, the temperature tuple range being usable by the monitoring system for calculating the compressor input power parameter and the evaporator temperature drop.

5. The system according to claim 4, wherein: The neighborhood extractor defines a temperature tuple range for the given temperature tuple by: constructing a set of observed temperature tuples from the temperature tuples in the temperature map; determining whether the set of observed temperature tuples satisfies a predefined minimum number of temperature tuples; and determining whether the given temperature tuple is within a convex hull of a subset of the set of observed temperature tuples.

6. The system of claim 5, further comprising a parameterized predictor operable to calculate predicted values ​​of the compressor input power parameter and the evaporator temperature drop using the set of observed temperature tuples.

7. The system according to claim 6, wherein: The parameterized predictor calculates predicted values ​​of the compressor input power parameter and the evaporator temperature drop using a summary value table constructed from the set of observed temperature tuples and the temperature map and using parameter coefficients derived from the summary value table.

8. The system according to claim 7, wherein: The relationship builder includes a CIPP relationship builder configured to learn the CIPP relationship and an ETD relationship builder configured to learn the ETD relationship.

9. The system according to claim 8, wherein: The temperature maps include a CIPP temperature map configured to receive and store summary statistics of measured compressor input power parameters from the CIPP relationship builder, and an ETD temperature map configured to receive and store summary statistics of evaporator temperature drops from the ETD relationship builder.

10. The system according to claim 9, wherein: The neighborhood extractor is a joint neighborhood extractor operable to define a temperature tuple range for a given temperature tuple using both a CIPP temperature map and an ETD temperature map.

11. The system according to claim 10, wherein: The parametric predictor includes a CIPP parametric predictor operable to calculate a predicted value of the compressor input power parameter and an ETD parametric predictor operable to calculate a predicted value of the evaporator temperature drop.

12. A method for monitoring an HVAC&R system, the method comprising: At a data acquisition processor, obtaining observations about the HVAC&R system, the observations including a condenser fluid temperature measurement and an evaporator fluid temperature measurement, the observations also including compressor input power parameter measurements corresponding to the fluid temperature measurements; At the relationship builder, learning a compressor input power parameter (CIPP) relationship between the fluid temperature measurements of the evaporator entry temperature and the condenser entry temperature and the compressor input power parameter measurement, and further operable to learn an evaporator temperature drop (ETD) relationship between the fluid temperature measurements of the evaporator entry temperature and the condenser entry temperature and the evaporator temperature drop; as well as At a temperature map comprising a plurality of cells, each cell corresponding to a temperature tuple consisting of a condenser entry temperature and an evaporator entry temperature, receiving and storing from a relationship builder a summary statistic of a measured compressor input power parameter or a measured derived evaporator temperature drop, or both, corresponding to the temperature tuple of the cell; The monitoring system calculates a predicted value of the compressor input power parameter and / or a predicted value of the evaporator temperature drop using the CIPP relationship or the ETD relationship, respectively, and the monitoring system declares that performance degradation exists in the HVAC&R system using the predicted value of the compressor input power parameter or the predicted value of the evaporator temperature drop, or both.

13. The method of claim 12, further comprising shutting off power to the HVAC&R system by the monitoring system in response to declaring that there is performance degradation in the HVAC&R system.

14. The method according to claim 12, wherein: Learning the CIPP relationship and the ETD relationship, respectively, is performed by the relationship builder using a machine learning based learning process.

15. The method according to claim 12, further comprising: At a neighborhood extractor, a temperature tuple range surrounding a given temperature tuple is defined for a given temperature tuple and its cells, the temperature tuple range being usable by the monitoring system for calculating the compressor input power parameter and the evaporator temperature drop.

16. The method according to claim 15, wherein Defining a temperature tuple range for the given temperature tuple is performed by the neighborhood extractor by: constructing a set of observed temperature tuples from the temperature tuples in the temperature map; determining whether the set of observed temperature tuples satisfies a predefined minimum number of temperature tuples; and determining whether a given temperature tuple lies within a convex hull of a subset of the set of observed temperature tuples.

17. The method according to claim 16, further comprising: Predicted values ​​of the compressor input power parameter and the evaporator temperature drop are calculated by a parameterized predictor using the set of observed temperature tuples.

18. The method according to claim 17, wherein The parameterized predictor calculates predicted values ​​of the compressor input power parameter and the evaporator temperature drop using a summary value table constructed from the set of observed temperature tuples and the temperature map and using parameter coefficients derived from the summary value table.

19. The method according to claim 18, wherein The relationship builder includes a CIPP relationship builder configured to learn the CIPP relationship and an ETD relationship builder configured to learn the ETD relationship.

20. The method according to claim 19, wherein The temperature maps include a CIPP temperature map configured to receive and store summary statistics of measured compressor input power parameters from the CIPP relationship builder, and an ETD temperature map configured to receive and store summary statistics of evaporator temperature drops from the ETD relationship builder.

21. The method according to claim 20, wherein The neighborhood extractor is a joint neighborhood extractor operable to define a temperature tuple range for a given temperature tuple using both a CIPP temperature map and an ETD temperature map.

22. The method according to claim 21, wherein The parametric predictor includes a CIPP parametric predictor operable to calculate a predicted value of the compressor input power parameter and an ETD parametric predictor operable to calculate a predicted value of the evaporator temperature drop.

23. A non-transitory computer readable medium containing program logic that, when executed through the operation of one or more computer processors, causes the one or more processors to perform the method of claim 12.

24. A monitoring and detection system comprising: at least one processor; A storage device coupled to the at least one processor and having stored thereon processor-executable instructions, the instructions comprising instructions that, when executed by the at least one processor, cause the at least one processor to instantiate: a data acquisition processor operable to acquire observations about the system, the observations comprising a specified system temperature measurement and an input power parameter measurement corresponding to the specified temperature measurement; a relationship builder operable to learn a relationship between a specified system temperature measurement and an input power parameter measurement, and to learn a relationship between a specified system temperature measurement and a specified system temperature drop; a temperature map comprising a plurality of cells, each cell corresponding to a temperature tuple consisting of a specified system temperature measurement, the temperature map being configured to receive from the relationship builder and store, for each cell, summary statistics of a measured input power parameter or a measured-derived specified system temperature drop corresponding to the temperature tuple of the cell, or both; The processor-executable instructions further cause at least one processor to use the relationship to calculate a predicted value of an input power parameter and a predicted value of a specified system temperature drop, respectively, and are further configured to use the predicted value of the input power parameter and the predicted value of the specified system temperature drop to declare that there is performance degradation in the system.

Images