Abnormality cause estimation system and abnormality cause estimation method

Abstract

Provided are an abnormality cause estimation system and an abnormality cause estimation method for estimating an abnormality cause with high accuracy in an air conditioning system. Comprises: a feature quantity (learning data) calculation means that detects a feature quantity during normal operation and obtains a feature quantity (learning data); a feature amount (abnormal data) calculation means that detects a feature amount during an abnormal operation and obtains a feature amount (abnormal data); and an abnormality cause estimation means for estimating the cause of abnormality during the abnormal operation. The abnormality cause estimation unit includes: an abnormality cause estimation map in which an abnormality feature amount state (map) is set in advance and an evaluation value including a conformity between an abnormality cause and a feature amount is set; and an abnormality cause index calculation means that, when the abnormality feature quantity state (abnormality data) during the abnormal operation coincides with the abnormality feature quantity state (map), adds the evaluation value when the abnormality cause and the feature quantity are suited, and subtracts the evaluation value when the abnormality cause and the feature quantity are not suited, thereby obtaining an abnormality cause estimated evaluation value.

Description

Technical Field

[0001] The present invention relates to an abnormality cause inference system and an abnormality cause inference method, and in particular to an abnormality cause inference system and an abnormality cause inference method which are effective for determining subsequent actions such as countermeasures for the abnormality, cause investigation, etc. after an abnormality in an air-conditioning system is detected. Background Art

[0002] In order to reduce the cost of inspection and maintenance of air conditioning systems, it is extremely important to detect abnormalities, failures (hereinafter, representatively referred to as abnormalities) and signs of the occurrence of abnormalities in air conditioning systems (abnormality detection). In addition, there is a recent trend to commercialize services that not only provide air conditioning equipment to customers, but also undertake the operation and maintenance of the provided air conditioning equipment.

[0003] In addition, the Ministry of the Environment has implemented the "CFC Emission Control Law" for commercial air-conditioning systems and refrigeration equipment since April 2015, making "simple inspections" and "regular inspections" mandatory for equipment using CFC gas owned by companies. Therefore, it is important to accurately detect abnormalities in air-conditioning systems at an early stage in the operation of service businesses.

[0004] In addition, from the perspective of maintenance and repair, it is important to detect an abnormality in the air-conditioning system composed of an outdoor unit (one or more units) and an indoor unit (more units) connected by piping for refrigerant circulation, determine the cause of the abnormality, and determine which device has caused the abnormality. By effectively utilizing such abnormality detection, the following "environmental value", "economic value", and "social value" can be provided.

[0005] That is, in terms of "environmental value", early detection of refrigerant (Freon) leakage can reduce the amount of leakage, which helps prevent global warming. In addition, in terms of "economic value", the customer's business losses (production stoppage, yield rate, food waste loss, etc.) caused by sudden failure of air-conditioning equipment can be suppressed, and by changing maintenance from a time basis to a condition basis, it can help reduce life cycle costs. Furthermore, in terms of "social value", it can help stabilize the operation of air-conditioning equipment in key businesses such as medical sites and eliminate the shortage of maintenance personnel (improvement of maintenance workers' work efficiency).

[0006] In particular, it is impossible to determine the next action such as countermeasures for the abnormality or cause investigation by detecting abnormal signs alone, so it is necessary to infer the cause of the abnormality, and this demand is increasing. In addition, when the characteristic quantity with sensitivity is repeated in the abnormal cause, there is a problem that the accuracy of the abnormal cause inference becomes low. Here, the low accuracy of the inference of the abnormal cause includes two meanings: low inference accuracy (many false inferences) and small recognition margin even in the case of correct inference. Therefore, it is necessary to develop a method for inferring the cause of the abnormality with high accuracy.

[0007] Regarding the inference of the cause of the abnormality, for example, the following method is disclosed in Japanese Patent Gazette No. 2020-16358 (Patent Document 1): using an abnormal cause inference map that represents the relationship between the cause of the abnormality and the state of the abnormal feature quantity (hereinafter referred to as the abnormal feature quantity state), when the abnormal feature quantity state of the abnormal cause inference map is consistent with the feature quantity state of the sampled abnormal data, a suitable (compliant) score of the abnormal cause (so-called evaluation value) is added, and the cause of the abnormality is inferred based on the size of the added score.

[0008] Prior art literature

[0009] Patent Literature

[0010] Patent Document 1: Japanese Patent Application Publication No. 2020-16358 Summary of the invention

[0011] Problems to be solved by the invention

[0012] However, the abnormality detection system described in Patent Document 1 selects a plurality of calculation formulas representing the air conditioner capacity corresponding to the refrigerant of the air conditioner to calculate the capacity value based on the acquired sensor signal, and performs abnormality detection based on the learned normal capacity value data. Furthermore, when an abnormality is detected, the system determines the abnormality-related capacity value (feature quantity) and outputs information related to the abnormal state.

[0013] Then, using an abnormality cause estimation map indicating the relationship between the abnormality cause and the abnormality feature quantity state, when the abnormality cause estimation map and the feature quantity state of the abnormal data are consistent, the score of the appropriate abnormality cause is added and the abnormality cause is estimated based on its magnitude.

[0014] In the abnormality detection system of patent document 1, an abnormality cause inference map is used to represent the relationship between the abnormality cause and the abnormal feature quantity state. When the feature quantity state of the abnormality cause inference map is consistent with the feature quantity state of the abnormal data, only the evaluation value (score) of the appropriate abnormality cause is added. When the feature quantities that are sensitive to each abnormality cause are repeated, there is a problem that the accuracy of abnormality cause inference deteriorates.

[0015] For example, when a certain feature quantity is focused on, another abnormality cause different from the original abnormality cause may appear as a feature quantity of the original abnormality cause. Therefore, it is necessary to reduce the influence of this unnecessary feature quantity.

[0016] An object of the present invention is to provide an abnormality cause estimation system and an abnormality cause estimation method for estimating the cause of abnormality in an air conditioning system with high accuracy.

[0017] Methods for solving problems

[0018] The present invention is an abnormal cause inference system, which monitors the operation of an air-conditioning device composed of multiple device types, and is characterized in that: the abnormal cause inference system includes: a feature quantity (learning data) calculation unit, which detects multiple feature quantities representing the operation status of the above-mentioned device type when the above-mentioned air-conditioning device is operating normally, and obtains the feature quantity (learning data); a feature quantity (abnormal data) calculation unit, which detects multiple feature quantities representing the operation status of the above-mentioned device type when the above-mentioned air-conditioning device is operating abnormally, and obtains the feature quantity (abnormal data); and an abnormal cause inference unit, which infers the abnormal cause of the above-mentioned abnormal operation based on the above-mentioned feature quantity (normal data) and the above-mentioned feature quantity (abnormal data), and the above-mentioned abnormal cause inference unit includes: an abnormal cause inference mapping, which pre-sets a feature quantity corresponding to the above-mentioned abnormal cause for each of the above-mentioned feature quantities. The state of the above-mentioned characteristic quantity related to the abnormal action (hereinafter referred to as the abnormal characteristic quantity state (mapping)), and an evaluation value including the consistency of the above-mentioned abnormal cause and the above-mentioned characteristic quantity is pre-set; an abnormal cause index calculation unit, which determines, for each of the above-mentioned characteristic quantities, whether the abnormal characteristic quantity state (abnormal data) calculated according to the above-mentioned characteristic quantity (abnormal data) during the above-mentioned abnormal action is consistent with the above-mentioned abnormal characteristic quantity state (mapping) of the above-mentioned abnormal cause inference mapping. If they are consistent, for each of the above-mentioned characteristic quantities, when the above-mentioned abnormal cause is consistent with the above-mentioned characteristic quantity, the above-mentioned evaluation value of the above-mentioned characteristic quantity judged to be consistent is added, and when the above-mentioned abnormal cause is not suitable for the above-mentioned characteristic quantity, the above-mentioned evaluation value of the above-mentioned characteristic quantity judged to be not suitable is subtracted, so as to obtain the abnormal cause inference evaluation value used to infer the above-mentioned abnormal cause.

[0019] Effects of the Invention

[0020] According to the present invention, by performing a subtraction process on the evaluation value of the feature amount that is not suitable for the cause of the abnormality, it is possible to reduce false detections, and thus it is possible to estimate the cause of the abnormality with high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a structural diagram showing the basic structure of the abnormality detection system of the air conditioning system to which the present invention is applied.

[0022] Figure 2It is a control block diagram showing a processing block of an abnormality detection system in an air conditioning system according to an embodiment of the present invention.

[0023] Figure 3 It is an explanatory diagram for explaining the two-dimensional distribution density for determining the feature quantity related to abnormality.

[0024] Figure 4 This is an explanatory diagram for explaining abnormality contribution values ​​of feature quantities related to abnormality causes.

[0025] Figure 5A It is an explanatory diagram for explaining the state of the first abnormal feature quantity related to the abnormal cause.

[0026] Figure 5B It is an explanatory diagram for explaining the state of the second abnormal feature quantity related to the abnormal cause.

[0027] Figure 6 This is an explanatory diagram of an abnormality cause estimation map for explaining the relationship between the abnormality feature quantity state indicating the abnormality cause and the abnormality cause.

[0028] Figure 7 : is a flowchart explaining the calculation process of the abnormality cause index.

[0029] Fig. 8A It is an explanatory diagram for explaining an example of analysis results in the abnormality sign detection unit.

[0030] Figure 8B Yes Description Figure 2 An explanatory diagram of an example of analysis results in the abnormality cause index calculation unit in FIG.

[0031] Description of Reference Numerals

[0032] 100…Air conditioning system

[0033] Ssig…sensor signals from multiple devices

[0034] 101… Air conditioning system

[0035] 102…Sensor signal input part

[0036] 200…Monitoring Computer

[0037] 300…Sensor signal processing department

[0038] 400…Abnormal sign detection unit

[0039] 500…Abnormal cause inference system

[0040] 501 ... abnormality related feature quantity determination unit

[0041] 502…Exception cause speculation mapping

[0042] 503…Abnormal Cause Index Calculation Department

[0043] 504 ...Abnormality cause determination unit. DETAILED DESCRIPTION

[0044] Embodiments of the present invention will be described in detail below using the drawings, but the present invention is not limited to the following embodiments, and various modifications and application examples are also included in the technical concept of the present invention.

[0045] Figure 1 The basic structure of the abnormality detection system in the air conditioning system is shown. The abnormality detection system in the air conditioning system is composed of an air conditioning system 100 having a plurality of devices for air conditioning and sensors for detecting operating states, etc., and a monitoring computer 200 for monitoring the operation of the air conditioning system 100.

[0046] The air conditioning system in the air conditioning equipment system 100 is basically composed of an outdoor unit and an indoor unit, which are connected by a refrigerant pipe. The outdoor unit is equipped with a compressor, a heat exchanger, an expansion valve, a blower fan and other equipment, and the indoor unit is equipped with a heat exchanger, an expansion valve, a blower fan and other equipment. In addition, each device is provided with a sensor for detecting the operating state quantity (for example, temperature, pressure, current, etc.) of the device.

[0047] The air conditioning system is composed of one or more outdoor units and one or more indoor units, so the multiple devices that constitute each outdoor unit and indoor unit are grasped as device types. Here, the device type refers to the type of device that performs a certain function. For example, when focusing on the compressor as the device type, the compressors of different outdoor units are "the same type of device" when they are considered as compressors. Similarly, when focusing on the expansion valve and the air supply fan as the device type, the expansion valve and the air supply fan of different outdoor units are respectively "the same type of device". Of course, the same is true for indoor units.

[0048] The action state quantity includes a sensor signal that can be directly measured from a sensor and a feature quantity that cannot be measured by a sensor but is obtained by calculation based on the sensor signal, but basically, both are described as feature quantities below. In addition, as a sensor signal or feature quantity, in the case of special processing, the meaning is recorded.

[0049] In addition, the monitoring computer 200 includes an interface with input and output functions and a processor (computation unit) with a computing function, and the processor can perform the operations of the present embodiment described below according to the control program. In addition, as an example of a processor, a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit) are considered, but other semiconductor devices can also be used as long as they are the main body for performing the prescribed computing processing.

[0050] The monitoring computer 200 may be integrated with the air conditioning system, or may be a cloud system connected by wire or wireless connection. In addition, since the control program has a control function, it can be understood as a control function block.

[0051] Figure 2 The structure of the abnormality prediction and abnormality cause estimation system in the air conditioning system is shown. The air conditioning equipment system 100 is installed in multiple customer sites, and the sensor information and the like are transferred to the monitoring computer 200 via the network. The customer site corresponds to a building (such as a building), for example, a PAC air conditioning system and a VRF air conditioning system etc. possessed by the building.

[0052] The air conditioning system 100 is composed of one or more air conditioning systems 101, and a sensor signal input unit 102 that inputs sensor signals Ssig of a plurality of devices installed in the air conditioning system 101 and transmits them to the monitoring computer 200. In the air conditioning system 101, sensors are installed in one or more of the various devices. In addition, the sensor signals Ssig of the plurality of devices are, for example, signals related to temperature, current, pressure, etc.

[0053] The monitoring computer 200 detects abnormalities in the air conditioning system 101 and infers the causes of the abnormalities (refrigerant shortage, expansion valve abnormality, compressor abnormality, indoor heat exchanger abnormality, outdoor heat exchanger abnormality, etc.). These are functions executed by the control program of the processor. For details of the processing flow executed by these control programs, see Figure 3 to Figure 6 is described in.

[0054] The monitoring computer 200 inputs the sensor signals Ssig of the plurality of devices transmitted from the air conditioning system 100 to the sensor signal processing unit 300 , and uses the characteristic amount Ssta of the device generated by the sensor signal processing unit 300 to detect abnormality signs and infer the cause of the abnormality.

[0055] Next, the control function of the monitoring computer 200 is described. In the monitoring computer 200, the sensor signals Ssig of the plurality of devices transmitted from the air conditioning system 100 are processed into the characteristic quantities Ssta of the devices corresponding to the sensor signals Ssig by the sensor signal processing unit 300. Here, the characteristic quantity Ssta of the device refers to a sensor signal that can be directly measured from a sensor, and a characteristic quantity that cannot be measured by a sensor but is obtained by calculation based on the sensor signal.

[0056] These processed feature quantities Ssta are input to the subsequent abnormality sign detection unit 400. The abnormality sign detection unit 400 learns and stores the feature quantities Ssta of the equipment obtained during the normal operation of the air conditioning system 100. This is treated as normal learning data.

[0057] Then, when evaluating the operation of the air conditioning system 100, the feature quantity extracted by sampling is compared with the learned feature quantity Ssta, and when it deviates from the learned normal range, it is detected as an abnormality sign. The abnormality sign information from the abnormality sign detection unit 400 is input to the subsequent abnormality cause estimation system 500.

[0058] The characteristic quantity Ssta (learning data) of the equipment obtained during normal operation of the air conditioning equipment system 100 and the characteristic quantity Ssta (abnormal data) of the equipment detected to be abnormal by the abnormality sign detection unit 400 are input to the abnormality-related characteristic quantity determination unit 501 of the subsequent stage constituting the abnormality cause estimation system 500.

[0059] In the abnormality-related characteristic quantity determination unit 501, the characteristic quantity Ssta (learning data) and the characteristic quantity Ssta (abnormal data) are compared to obtain at least (1) the "abnormality contribution value (ACN)" of the characteristic quantity (which can be expressed as abnormality frequency), and (2) the "abnormality characteristic quantity state (DFS)" of the characteristic quantity (a parameter indicating a state that is larger or smaller than the normal state), and input them into the abnormality cause index calculation unit 503.

[0060] The abnormal contribution value (ACN) is based on Figure 3 The abnormal feature state (DFS) is obtained based on the grayscale image information based on the feature quantity shown in FIG. 5 , and these will be described later. Figure 3 The same image information as that of FIG. 5 can be used.

[0061] In addition to the above-mentioned abnormal contribution value (ACN) and abnormal characteristic state (DFS), the abnormal cause inference map 502 is also inputted in the abnormal cause inference map 502. The abnormal cause inference map 502 is obtained by forming a database for each abnormal cause based on the changes in the refrigeration cycle assumed when the abnormality occurs, and the abnormal characteristic state (MFS) and evaluation value predetermined according to the operating conditions, the performance of the components of the refrigeration cycle, etc. Figure 6 This is also described later.

[0062] In the abnormal cause index calculation unit 503, the abnormal feature quantity state (DFS) of the abnormal data of the abnormal associated feature determination unit 501 and the abnormal contribution value (ACN) of each feature quantity, as well as the abnormal feature quantity state (MFS) and evaluation value of the abnormal cause inference map 502 are used to calculate the abnormal cause index (CI) for each abnormal cause.

[0063] The abnormality cause index (CI) is used as an indicator of the occurrence of abnormality. The abnormality cause index (CI) is input to the abnormality cause determination unit 504. In addition, the specific processing of the abnormality cause index calculation unit 503 and the abnormality cause determination unit 504 is performed by using Figure 7 In the abnormality cause identification unit 504, an abnormality cause index (CI) is input for each abnormality cause, and the abnormality cause having the largest abnormality cause index (CI) is estimated as the most likely abnormality to identify the abnormality cause.

[0064] The abnormality sign detection unit 400 can handle not only machine learning but also AI (Artificial Intelligence) and deep learning. However, it is necessary to be able to extract the date and time when the abnormality is detected and data (feature amount).

[0065] Next, the abnormality contribution value (ACN) and the abnormality feature state (DFS) in the abnormality-related feature determination unit 501 will be described.

[0066] Figure 3 A two-dimensional distribution density image is used to determine the feature quantity associated with the abnormality. In addition, the image can be obtained by a known method. When learning the feature quantity, the two-dimensional distribution density image is calculated by a cyclic system of two feature quantities, and saved in the form of a grayscale image. The two-dimensional distribution density refers to a two-dimensional histogram with two feature quantities as the horizontal axis and the vertical axis.

[0067] Regarding these images, the areas where the feature quantity Ssta (learning data) which is normal learning data does not exist are represented by "white" pixels, that is, the areas where abnormalities have occurred, and the areas where the feature quantity Ssta (learning data) exists are represented by "black" to "gray" pixels. The darker the gray, the higher the density, that is, no abnormalities have occurred. Therefore, they are called distribution density images.

[0068] And, in Figure 3 , which shows the state of sampling abnormal data in a certain operating state of the air conditioning system. The portion indicated by the "○" mark in the figure is the plotting position of the abnormal data, and the abnormal state can be determined by the grayscale color at the plotting position.

[0069] Therefore, when evaluating the operation of the air conditioning system, the feature quantity that deviates from the distribution of the feature quantity Ssta (learning data) as normal learning data is regarded as the feature quantity that contributes to the abnormality. That is, based on the feature quantity that deviates from the distribution of the normal feature quantity Ssta (learning data), the abnormal contribution value (ACN) and the abnormal feature quantity state (DFS) are derived.

[0070] Figure 4 Indicates the abnormality contribution value (ACN) of the feature quantity related to the abnormality. Figure 3 The abnormal contribution value (ACN) is calculated based on the depiction position of each distribution density image of the detected abnormal data. The abnormal contribution value (ACN) is calculated by summing the number of "white" pixels (meaning that there is no feature value Ssta (learning data)) in the two-dimensional distribution density image corresponding to the depiction position of the abnormal data for each feature value.

[0071] exist Figure 4 In the figure, the number of "white" pixels at the plotted position in multiple feature quantities (A) to feature quantities (J) in any period in which abnormality detection has been performed is shown. Therefore, the more "white" pixels there are, the higher the possibility of being considered abnormal. Figure 4 In the figure, feature (I) is the first abnormal contribution, followed by feature (C) as the second abnormal contribution, and feature (D) as the third abnormal contribution. Thus, the larger the abnormal contribution value (ACN), the more it can be regarded as a feature related to the abnormality. In this embodiment, the use of the abnormal contribution value (ACN) for abnormality inference is also a large feature.

[0072] Next, the abnormal feature state (DFS) will be described. Figure 5A , Figure 5B Indicates the feature state related to the abnormality. The abnormal feature state (DFS) can be used Figure 3The abnormal feature quantity state (DFS) is determined by whether the sampled feature quantity is larger or smaller than the feature quantity Ssta (learning data).

[0073] Figure 5A In the figure, both the horizontal and vertical axes represent the two-dimensional distribution density image of the feature quantity (I). When the sampled feature quantity exists at the position shown in area A, the feature quantity (I) deviates from the feature quantity Ssta (learning data) in a larger direction, so the feature quantity (I) is judged to be an abnormal feature quantity state (DFS) that is larger than the normal state.

[0074] Similarly, in Figure 5B In the figure, the horizontal axis represents the feature quantity (C) and the vertical axis represents the two-dimensional distribution density image of the feature quantity (D). When the sampled feature quantity exists at the position shown in the area B, the feature quantity (D) deviates from the feature quantity Ssta (learning data) to the larger side, so the feature quantity (D) is judged to be in a state where the abnormal feature quantity state (DFS) is larger than the normal state. On the other hand, since the feature quantity (C) does not deviate much from the feature quantity Ssta (learning data), it is judged that the abnormal feature quantity state (DFS) of the feature quantity (C) is in a normal state.

[0075] Next, the abnormality cause estimation map 502 input to the abnormality cause index calculation unit 503 will be described. Figure 6 It is a table showing the abnormality cause estimation map 502 showing the mutual relationship (suitable / unsuitable) between the abnormality feature quantity state (MFS: a parameter indicating a state larger or smaller than that in a normal state) and the abnormality cause for each abnormality cause.

[0076] Figure 6 The abnormality cause estimation map 502 (see Figure 2 ) is a mapping made by pre-calculating the abnormal characteristic state (MFS) of the equipment such as the operating conditions and the performance of the constituent equipment and evaluating the value according to each abnormal cause based on the changes in the refrigeration cycle assumed when an abnormality occurs, and storing them in a database.

[0077] The abnormal feature state (MFS) is used to match the sampled abnormal feature state (DFS) and is identified as an abnormal state when matched. In addition, the abnormal cause estimation map 502 sets the evaluation value of the feature used in the calculation of the abnormal cause index (CI) described later, that is, the weight coefficient (MFW) and makes it into a database.

[0078] Here, the abnormal feature state (MFS) and the weight coefficient (MFW) of the feature are obtained through experience, experiments or design, prepared in advance and stored in the storage element in the form of mapping or table. That is, the abnormal feature state (MFS) can be grasped in advance for each abnormal cause and each feature. In addition, the relationship between the abnormal cause and the appearance of the feature (the relationship between the suitability / unsuitability of the abnormal cause and the feature) can also be grasped in advance. Therefore, when the abnormal cause and the appearance of the feature are not suitable, the information of ignoring the weight coefficient (MFW) is given. Figure 7 The processing of these data is described in the flowchart.

[0079] return Figure 6 On the vertical axis, abnormal causes (i), namely, (1) insufficient refrigerant, (2) excess refrigerant, (3) abnormal expansion valve 1, (4) abnormal expansion valve 2, (5) abnormal compressor, (6) abnormal indoor heat exchanger, (7) abnormal outdoor heat exchanger 1, and (8) abnormal outdoor heat exchanger 2, are set as the database. In addition, other abnormal causes (i) different from these abnormal causes can of course be set.

[0080] In addition, on the horizontal axis, the feature quantities (j) related to each of the above-mentioned abnormal causes (i), that is, feature quantities (1) to (10) are set as a database. These feature quantities (j) are multiple feature quantities related to a certain abnormal cause (i), but not all feature quantities (j) are related to the same abnormal cause. In other words, a typical feature quantity (j) with high sensitivity is assigned to one abnormal cause (i).

[0081] like Figure 6 As shown, the abnormal feature state (MFS) for an abnormal cause (i) is set in the upper part of the table. The symbol in the upper part of the table represents the abnormal feature state (MFS) suitable for the abnormal cause. Here, suitable means that when the feature state is observed in the detected abnormal cause (i), it is likely to be the corresponding abnormal cause.

[0082] "H" indicates that the sampled feature value (j) takes a larger value than the normal feature value (j) in response to the abnormal cause (i) (high sensitivity), "L" indicates that the sampled feature value (j) takes a smaller value than the normal feature value (j) in response to the abnormal cause (i) (high sensitivity). "0" indicates a feature value (j) within the normal feature value range or not in response to the abnormal cause (i).

[0083] For example, in Figure 6In the example, when the abnormal cause (i) is (1) insufficient refrigerant, the characteristic value (9) responds to (1) insufficient refrigerant and takes a value larger than the normal characteristic value, so an evaluation of "H" is assigned. As described above, the symbols representing these abnormal characteristic value states (MFS) are obtained based on experience, experiments or design.

[0084] Next, the numbers in the lower part of the table are the weight coefficients (MFW) of the feature quantities used in the calculation of the abnormal cause index (CI) described later. The weight coefficient (MFW) of the feature quantity (j) suitable for the abnormal cause (i) is given a "+" (positive) attribute. Here, the value of the weight coefficient (MFW) is set so that the total value becomes "1" for one abnormal cause (i).

[0085] For example, for the abnormal cause of (1) insufficient refrigerant, since only the characteristic quantity (9) is set to "H" as the abnormal characteristic quantity state (MFS), the weight coefficient (MFW) is set to "1.0". In addition, the weight coefficient (MFW) is also "1.0" in (2) excess refrigerant and (5) compressor abnormality to (8) outdoor heat exchanger 2 abnormality.

[0086] On the other hand, for the abnormal cause of (3) expansion valve 1, the characteristic quantity (4) and the characteristic quantity (8) are both set to "H" as the abnormal characteristic quantity state (MFS), so the weight coefficient (MFW) is proportionally allocated and set to "0.5" respectively. The same is true for (4) expansion valve 2.

[0087] The weight coefficient (MFW) may be allocated in proportion to the number of feature quantities suitable for the cause of the abnormality so that the total sum becomes "1", or may be set to different coefficients so that the total sum becomes "1".

[0088] On the other hand, the weight coefficient (MFW) of the feature quantity (j) that is not suitable for the abnormality cause (i) is given a "-" (negative) attribute. In this case, the weight coefficient (MFW) is always "-1".

[0089] Here, the feature quantity that is not suitable for the abnormal cause is assumed as follows. That is, the same feature quantity may appear repeatedly even if it is a different abnormal cause. Therefore, if it is a feature quantity based on the original abnormal cause, there is no problem even if a weight coefficient (evaluation value) is added, but if it is a feature quantity based on a different abnormal cause, adding a weight coefficient (evaluation value) will lead to an error in the estimation of the abnormal cause.

[0090] Therefore, since the consistency between the abnormal cause and the occurrence of the feature quantity can be grasped in advance, the consistency is reflected as "+" or "-" in the abnormal cause estimation map 502. In the case of "+", it is appropriate to indicate that the abnormal cause and the feature quantity are in the original relationship, and in the case of "-", it is inappropriate to indicate that the abnormal cause and the feature quantity are not in the original relationship.

[0091] For example, the characteristic quantity (6) in (5) compressor abnormality appears as a characteristic quantity even though it is originally a characteristic quantity unrelated to the compressor abnormality. Therefore, this characteristic quantity is not a characteristic quantity of the compressor abnormality, and the weight coefficient (evaluation value) is set to "-1" and subtraction processing is performed. In other words, the evaluation value of the characteristic quantity is ignored.

[0092] In this way, for feature quantities that are not suitable for the cause of abnormality, by assigning a "-" attribute to a weight coefficient (evaluation value) and performing subtraction processing, it is possible to reduce the estimation error of the cause of abnormality.

[0093] Here, although the weight coefficient (MFW) is used for explanation in the embodiment, it is also possible to express it as an evaluation value without using the weight coefficient (MFW). In the present embodiment, although the weight coefficient (MFW) is used for explanation, it can be understood as an evaluation value including the weight coefficient (MFW) as a method of thinking.

[0094] In addition, the feature value (j) is set in a manner that the feature values ​​related to the abnormal cause are not repeated. As a result, the number of feature values ​​of the abnormal cause to be extracted can be reduced, and the estimation of the abnormal cause becomes easier. For example, if only a plurality of feature values ​​suitable for each abnormal cause are set, the feature values ​​that are repeated with other abnormal causes increase, and there is a problem that any abnormal cause becomes the same estimation level and it is difficult to estimate the abnormal cause.

[0095] Next, based on Figure 7 The following describes a process for obtaining an abnormality cause index (CI) using a detected abnormality feature state (DFS), an abnormality contribution value (ACN), a preset abnormality feature state (MFS), and a preset feature weight coefficient (MFW).

[0096] Figure 7 The calculation process flow of the abnormality cause index shown is executed by the abnormality cause estimation system 500 of the processor of the monitoring computer 200 and is started at a predetermined time interval. The time interval is set to any interval such as every 1 minute, every 5 minutes, every 10 minutes, every hour, every day, etc.

[0097] <Step S701>

[0098] In step S701, for each of the plurality of abnormal causes (I), loop processing 1 is started, and the following processing is repeatedly performed until the number of abnormal causes (i) is completed. Thus, the abnormal cause inference operation in all abnormal causes (i) can be performed in sequence, and the abnormal cause can be inferred. When loop processing 1 is started, the processing of step S702 is performed.

[0099] <Step S702>

[0100] In step S702, the abnormality cause index (CIi) corresponding to the number of abnormality causes (I) set in step S701 is set and initialized (value 0). When this processing is completed, the processing of step S703 is executed.

[0101] <Step S703>

[0102] In step S703, loop processing 2 is started for each of the plurality of feature quantities (J), and the following processing is repeatedly performed until the number of feature quantities (j) is reached. When loop processing 2 is started, the processing of step S704 is performed.

[0103] <Step S704>

[0104] In step S704, the abnormal contribution value (ACN) of each feature quantity is extracted using the two-dimensional distribution density image. The abnormal contribution value (ACN) is obtained by summing up the abnormal contribution value for each feature quantity. Figure 3 The number of white pixels (meaning that there is no learning data) in the two-dimensional distribution density image corresponding to the plotted position of the abnormal data shown in FIG. Figure 4 The abnormality contribution value (ACN) is as shown. The larger the abnormality contribution value (ACN), the more it becomes a feature quantity related to abnormality. When this process is completed, the process of step S705 is executed.

[0105] <Step S705>

[0106] In step S705, the feature quantity state (DFS) is extracted from the state of the feature quantity based on the abnormal cause using the two-dimensional distribution density by the method shown in FIG5. The feature quantity state based on the feature quantity of the abnormal cause is judged by whether the feature quantity of the sampled abnormal cause is larger or smaller relative to the learned feature quantity Ssta (learning data). When this process is completed, the process of step S706 is executed.

[0107] <Step S706>

[0108] In step S706, the abnormal feature quantity state (MFS) of each abnormality cause obtained in advance is extracted from the abnormality cause estimation map 502. When this processing is completed, the processing of step S707 is performed.

[0109] <Step S707>

[0110] In step S707, a weight coefficient (MFW) of the feature amount of each abnormality cause is extracted from the abnormality cause estimation map 502. When this process is completed, the process of step S708 is performed.

[0111] <Step S708>

[0112] In step S708, the characteristic state (DFS) of each abnormality cause obtained based on FIG5 is confirmed. Figure 6 The consistency of the abnormal feature state (MFS) of each abnormal cause in the abnormal cause estimation map 502.

[0113] If the state of the feature quantity of the predetermined abnormal cause matches the state of the feature quantity of the abnormal cause at the current time, the abnormal cause can be regarded as probable. On the contrary, if they do not match, the abnormal cause cannot be regarded as probable.

[0114] When the feature quantity state (DFS) of each abnormal cause is consistent with the abnormal feature quantity state (MFS) of each abnormal cause in the abnormal cause inference map, the processing of step S709 is performed. On the other hand, when the abnormal feature quantity state (MFS) is inconsistent with the abnormal feature quantity state (DFS), the processing of step S710 is performed.

[0115] <Step S709>

[0116] When it is determined in step S708 that the abnormal feature state (MFS) is consistent with the abnormal feature state (DFS), in step S709, "1" is set as the abnormal feature state fitness value (FSF). In the case of "FSF = 1", the possibility of abnormality is high, so the multiplication form of the next step to obtain the abnormal cause index (CI) formula (1) is valid. When this process is completed, the process of step S711 is executed.

[0117] <Step S710>

[0118] On the other hand, when it is determined in step S708 that the abnormal feature state (MFS) is inconsistent with the abnormal feature state (DFS), "0" is set as the abnormal feature state fit value (FSF) in step S710. In the case of "FSF = 0", the possibility of abnormality is low, so the multiplication form of the next step to obtain the abnormal cause index (CI) formula (1) is invalid. When this process is completed, the process of step S711 is executed.

[0119] <Step S711>

[0120] In step S711, the following formula (1) is used to add the product of the abnormal contribution value (ACN) of the characteristic quantity of each abnormal cause, the characteristic quantity state suitability value (FSF), and the weight coefficient (MFW) of the characteristic quantity of each abnormal cause to the abnormal cause index (CI) of each abnormal cause.

[0121] CIi+=(ACNj×FSFij×MFWij)…Formula (1)

[0122] The formula (1) represents that, according to a relative abnormal cause (for example, Figure 6 Each of the multiple characteristic quantities (for example, Figure 6 For each of the characteristic quantities (1) to (10) of the abnormality, the abnormality contribution value (ACN) and the weight coefficient (MFW) are multiplied, and these product values ​​are added together to obtain the abnormality cause index (CIi) of an abnormality cause.

[0123] For example, when looking at Figure 6 (1) When the refrigerant is insufficient, Figure 4 The contribution values ​​(ACN) of the characteristic quantities (A) to (J) shown correspond to Figure 6 The weight coefficient (MFW) of the feature quantities (1) to (8) and (10) is "0", so the accumulated abnormality cause index (CIi) is "0". However, Figure 4 The abnormal contribution value (ACN) of the characteristic quantity (I) shown is "74", and the corresponding Figure 6 The weight coefficient (MFW) of the feature quantity (9) is "1.0", so the accumulated abnormality cause index (CIi) is "74".

[0124] In formula (1), the abnormality contribution value (ACN) is multiplied by the weight coefficient (MFW), but the abnormality cause index (CIi) can also be obtained using only the weight coefficient (MFW). In this case, as described above, the weight coefficient (MFW) serves as the positioning of the evaluation value.

[0125] By executing loop processing 1, the operation shown in formula (1) is performed for each abnormal cause (i) to obtain the abnormal cause index (CIi). In this way, the magnitude of the abnormal cause index (CIi) represents the magnitude of the possibility of the abnormal cause. When this processing is completed, the processing of step S712 is executed.

[0126] <Step S712>

[0127] In step S712, the loop processing 2 of the arbitrary number of feature quantities (j) is completed. When this processing is completed, the processing of step S713 is executed.

[0128] <Step S713>

[0129] In step S713, the loop processing 1 of the arbitrary number of abnormal causes (i) is completed. When this processing is completed, the processing of step S714 is executed.

[0130] <Step S714>

[0131] In step S714, the abnormal cause corresponding to the largest abnormal cause index (CIi) among the abnormal cause indexes (CIi) for each abnormal cause (i) obtained through the completion of loop processing 1 and loop processing 2 of steps S712 and S713 is inferred as the most likely abnormal cause, completing the inference processing flow.

[0132] In addition, the abnormal cause index (CIi) of each abnormal cause is set to the cumulative value obtained by accumulating the product value obtained by multiplying the weight coefficient (MFW) of the characteristic value and the abnormal contribution (ACN), but it can also be set to the maximum value of the product value obtained by multiplying the weight coefficient (MFW) of the characteristic value and the abnormal contribution (ACN) without accumulation.

[0133] Here, the above-mentioned estimation process flow is executed at predetermined time intervals, such as 1 minute, 5 minutes, 10 minutes, 1 hour, 1 day, etc. As described above, it is also possible to execute the estimation process at any period, such as 1 day, 5 days, 1 week, 1 month, 6 months, 1 year, or more than 1 year, as the evaluation period.

[0134] In addition, in order to improve the stability of the estimation of the abnormal cause, the temporal variation of each abnormal cause index can also be smoothed by moving average. For example, a daily average, a weekly average, a monthly average, etc. can be obtained. In addition, for example, a daily moving average, a weekly moving average, a monthly moving average, etc. can be obtained.

[0135] Next, use Fig. 8A , Figure 8B An example of analysis results by the abnormality sign detection unit 400 and the abnormality cause index calculation unit 503 of the monitoring computer 200 will be described respectively.

[0136] Fig. 8A An example of the analysis result of abnormal sign detection based on the abnormal sign detection unit 400 is shown. In the abnormal sign detection curve graph (Grp), the horizontal axis represents the passage of time (date), "AM" represents the abnormal measurement degree based on the abnormal sign detection unit 400, that is, the abnormality, "SL" represents the threshold value (SL) set by the abnormal sign detection unit 400, and "AD" represents the presence or absence of abnormality. And, when the abnormal measurement degree, that is, the abnormality (AM) exceeds the threshold value (SL), it is judged that there is an abnormality (AD=1). On the other hand, when the abnormal measurement degree, that is, the abnormality (AM) does not exceed the threshold value (SL), it is judged that there is no abnormality (AD=0).

[0137] Figure 8B An example of analysis results of abnormal cause inference based on the abnormal cause index calculation unit 503 is shown. In the abnormal cause inference curve graph (Gmp), the horizontal axis represents the passage of time (date), and "CI1" and "CI2" represent abnormal cause indices based on the abnormal cause index calculation unit 503. In addition, "CI1" is an abnormal cause index related to insufficient refrigerant, and "CI2" is an abnormal cause index related to abnormal expansion valve 1. In addition, the abnormal cause index (CI) can be set to multiple according to the number of abnormal causes.

[0138] Furthermore, the display device connected to the monitoring computer 200 can display Fig. 8A Abnormal sign detection curve diagram (Grp), Figure 8B The abnormal cause estimation curve (Gmp) is thus able to read the time evolution of abnormal sign detection and abnormal cause index, and thus the sign of abnormal occurrence or the abnormal cause can be read.

[0139] As described above, the present invention is a system for estimating the cause of abnormality that monitors the operation of air-conditioning equipment composed of multiple types of equipment.

[0140] The abnormal cause inference system includes:

[0141] a characteristic quantity (learning data) calculation unit that detects a plurality of characteristic quantities representing the operating state of the equipment type when the air conditioning equipment is operating normally, and obtains the characteristic quantity (learning data);

[0142] a characteristic quantity (abnormal data) calculation unit that detects a plurality of characteristic quantities indicating the operating state of the equipment type when the air conditioner is abnormally operating, and obtains the characteristic quantity (abnormal data); and

[0143] an abnormality cause inferring unit, which infers the abnormality cause in the abnormal action based on the characteristic amount (normal data) and the characteristic amount (abnormal data), and

[0144] The abnormal cause inference unit includes:

[0145] an abnormality cause estimation map in which a state of a feature quantity related to an abnormality cause (hereinafter referred to as an abnormality feature quantity state (map)) is pre-set for each feature quantity, and an evaluation value including a correspondence between the abnormality cause and the feature quantity is pre-set;

[0146] An abnormal cause index calculation unit determines, for each feature quantity, whether an abnormal feature quantity state (abnormal data) obtained based on the feature quantity (abnormal data) during abnormal action is consistent with an abnormal feature quantity state (mapping) of an abnormal cause inference map. If they are consistent, for each feature quantity, when the abnormal cause is consistent with the feature quantity, an evaluation value of the feature quantity judged to be consistent is added; when the abnormal cause is not consistent with the feature quantity, the evaluation value of the feature quantity judged to be inconsistent is subtracted to obtain an abnormal cause inference evaluation value used to infer the abnormal cause.

[0147] Thus, by subtracting the evaluation value from the feature amount that is not suitable for the cause of the abnormality, it is possible to reduce false detections, and thus it is possible to estimate the cause of the abnormality with high accuracy.

[0148] In addition, the present invention is not limited to the above-mentioned embodiments, and includes various modified examples. The above-mentioned embodiments are examples described in detail in order to easily explain the present invention, and are not necessarily limited to having all the structures described. In addition, a part of the structure of a certain embodiment can be replaced with the structure of other embodiments, and the structure of other embodiments can be added to the structure of a certain embodiment. Regarding the structure of each embodiment, other structures can also be added, deleted, or replaced.

Claims

1. An abnormality cause inference system that monitors the operation of air conditioning equipment consisting of multiple types of equipment, wherein the abnormality cause inference system is characterized by: The abnormal cause inference system comprises: a learning data feature quantity calculation unit that detects a plurality of feature quantities indicating the operating state of the equipment type when the air conditioner is operating normally, and obtains a learning data feature quantity as a feature quantity of the learning data; an abnormal data feature quantity calculation unit that detects a plurality of feature quantities indicating the operating state of the equipment type when the air conditioner is abnormally operating, and obtains an abnormal data feature quantity as a feature quantity of the abnormal data; and an abnormality cause inferring unit for inferring the abnormality cause of the abnormal action based on the learning data feature quantity and the abnormal data feature quantity, The abnormal cause inference unit comprises: an abnormality cause inference map, in which a state of the feature quantity related to the abnormality cause, i.e., an abnormality feature quantity state map, is pre-set for each feature quantity, and an evaluation value including the consistency between the abnormality cause and the feature quantity is pre-set; An abnormal cause index calculation unit determines, for each of the feature quantities, whether the abnormal data abnormal feature quantity state obtained based on the abnormal data feature quantity during the abnormal action is consistent with the abnormal feature quantity state mapping of the abnormal cause inference mapping. If they are consistent, the following processing is performed for each of the feature quantities: when the abnormal cause is consistent with the feature quantity, the evaluation value of the feature quantity judged to be consistent is added; when the abnormal cause is not suitable for the feature quantity, the evaluation value of the feature quantity judged to be not suitable is subtracted, so as to obtain an abnormal cause inference evaluation value used to infer the cause of the abnormality.

2. The abnormality cause estimation system according to claim 1, characterized in that: The abnormal reason includes multiple abnormal reasons. The abnormality cause estimation unit detects a plurality of the feature quantities for each of the abnormality causes, and obtains a new evaluation value based on the plurality of feature quantities for each of the abnormality causes.

3. The abnormality cause estimation system according to claim 2, characterized in that: The abnormality cause estimation unit includes an abnormality cause determination unit that determines the abnormality cause having the largest evaluation value among the plurality of abnormality causes as the final abnormality cause.

4. The abnormality cause estimation system according to claim 1, characterized in that: The evaluation value is a weight coefficient, The abnormality cause inference unit includes an abnormality-related feature quantity determination unit, which generates a two-dimensional distribution density based on the two feature quantities and calculates the frequency of deviation of the abnormal data feature quantity from the learning data feature quantity as the abnormality contribution value of the feature quantity. The abnormality cause index calculation unit multiplies the weight coefficient by the abnormality contribution value to obtain the new evaluation value.

5. The abnormality cause estimation system according to claim 4, characterized in that: The two-dimensional distribution density is represented by a grayscale two-dimensional distribution density image, the learning data feature is "black" to "gray" pixels, and the abnormal data feature is "white" pixels. The abnormal contribution value is the number of "white" pixels.

6. The abnormality cause estimation system according to claim 1, characterized in that: The plurality of feature quantities in the abnormality cause estimation map are set so that the feature quantities corresponding to the abnormality cause do not overlap.

7. An abnormality cause inference method of an abnormality cause inference system, wherein the abnormality cause inference system monitors the operation of air conditioning equipment composed of multiple equipment types, wherein the abnormality cause inference method is characterized by: The abnormality cause inference method performs the following steps: a learning data feature quantity calculation step of detecting a plurality of feature quantities representing the operating state of the equipment type when the air conditioning equipment is operating normally, and obtaining a learning data feature quantity as a feature quantity of the learning data; an abnormal data feature quantity calculation step of detecting a plurality of feature quantities indicating the operating state of the equipment type when the air conditioner is operating abnormally, and obtaining an abnormal data feature quantity as a feature quantity of the abnormal data; and an abnormality cause inferring step of inferring the abnormality cause of the abnormal action based on the learning data feature quantity and the abnormal data feature quantity, In the abnormality cause estimation step, The following steps are performed using an abnormality cause inference map, wherein the abnormality cause inference map pre-sets a state of the feature quantity related to the abnormality cause for each feature quantity, and pre-sets an evaluation value including the consistency between the abnormality cause and the feature quantity, and hereinafter, the state of the feature quantity related to the abnormality cause is referred to as an abnormality feature quantity state map: A step of judging, for each of the feature quantities, whether the abnormal feature quantity state of abnormal data obtained based on the abnormal data feature quantity at the time of the abnormal action is consistent with the abnormal feature quantity state map of the abnormal cause inference map, and if they are consistent, for each of the feature quantities, when the abnormal cause is consistent with the feature quantity, adding the evaluation value of the feature quantity judged to be consistent, and when the abnormal cause is not consistent with the feature quantity, subtracting the evaluation value of the feature quantity judged to be not consistent; and By performing the addition process and the subtraction process, an abnormality cause estimation evaluation value for estimating the abnormality cause is obtained.

8. The abnormality cause inference method according to claim 7, characterized in that: The abnormal reason includes multiple abnormal reasons. The abnormality cause estimation step is a step of detecting a plurality of the feature quantities for each of the abnormality causes and obtaining a new evaluation value based on the plurality of feature quantities for each of the abnormality causes.

9. The abnormality cause inference method according to claim 8, characterized in that: The abnormality cause estimating step is a step of determining the abnormality cause having the largest evaluation value among the plurality of abnormality causes as the final abnormality cause.

10. The abnormality cause inference method according to claim 7, characterized in that: The evaluation value is a weight coefficient, The abnormality cause inference process performs the following process: generating a two-dimensional distribution density based on the two feature quantities, calculating the frequency of the deviation of the abnormal data feature quantity from the learning data feature quantity as the abnormality contribution value of the feature quantity, multiplying the weight coefficient by the abnormality contribution value, and obtaining the evaluation value.

11. The abnormality cause inference method according to claim 7, characterized in that: The plurality of feature quantities in the abnormality cause estimation map are set so that the feature quantities corresponding to the abnormality cause do not overlap.

Images