Method and device for fault detection in an air conditioning system

Abstract

One aspect concerns a method (200) for detection of abnormal conditioning unit operational cycle patterns indicative of fault events in an air conditioning unit (109) for regulating a temperature in a space, comprising - obtaining (A.1) first data representative of the power consumption of the air conditioning unit as a function of time and second data representative of at least one parameter impacting the air conditioning unit's performance; - combining (B.1) a rule-based model and a learning based model; - applying the combined models for the identification (C) of an abnormal air conditioning unit operational cycle pattern as a function of the first and second data; - in case of identification of an abnormal pattern, determining (C1) whether the identified abnormal cycle pattern is statistically significant, and in the affirmative, determining that a fault event took place and performing an action responsive to said determination.

Description

METHOD AND DEVICE FOR FAULT DETECTION IN AN AIR CONDITIONING SYSTEMTECHNICAL FIELD

[0001] A method and a device for fault detection in an air conditioning system are described. The invention can be used in - or in conjunction with - an air conditioning unit, including but not limited to the context of residential or industrial air conditioning.BACKGROUND

[0002] Air conditioning systems typically comprise a number of components subject to damage or failure. Issues may include e.g. refrigerant leaks, fan motor issues, frozen evaporator coils, blocked air filters and vents, faulty fans, dirty condensers, and compressor failure.

[0003] It is desirable to provide a solution allowing failure or damage detection in a non-invasive manner.SUMMARY

[0004] A first aspect concerns a method for detection of abnormal conditioning unit operational cycle patterns indicative of fault events in an air conditioning unit for regulating a temperature in a space, said method being carried out by at least one processor, said method comprising:- obtaining first data representative of the power consumption of the air conditioning unit as a function of time and second data representative of at least one parameter impacting the air conditioning unit’s performance;- combining a rule-based model and a learning based model;- applying the combined models for the identification of an abnormal air conditioning unit operational cycle pattern as a function of the first and second data;- in case of identification of an abnormal pattern, determining whether the identified abnormal cycle pattern is statistically significant, and in the affirmative, determining that a fault event took place and performing an action responsive to said determination.

[0005] Identification and referral of complex failures in AC units, such as refrigerant leaks, FAN motor issues, frozen evaporator coils, blocked air filters and vents, faulty fans, dirty condensers, and compressor cycle issues can be reliably detected, enabling prompt action to address them and ensure good performance of AC units.

[0006] According to one or more embodiments, the data descriptive of the at least one parameter comprises at least one among:- data descriptive of external environmental conditions;- environmental conditions internal to the space;- thermal characteristics of the space; and / or- data descriptive of the AC unit.

[0007] According to one or more embodiments:- data descriptive of external environmental conditions comprises one or more among season, temperature, humidity, air pressure, air quality;- environmental conditions internal to the space comprises one or more among temperature, humidity;- thermal characteristics of the space comprises one or more among level of insulation, size, volume, age, occupancy level;- data descriptive of the AC unit comprises one or more among date of installation, cumulated duration of operation, geographical location.

[0008] According to one or more embodiments, said rule-based model is one among:- an expert rule-based model descriptive of cases where an AC unit behavior is potentially susceptible of causing damage;- a knowledge graph providing semantic reasoning from AC unit concepts and characteristics;- one or more Bayesian models;- one or more belief networks.

[0009] According to one or more embodiments, the learning model is one among:- a machine and / or a deep learning model comprising at least one neural network based on sequence learning, one or more convolution layers and / or a variational auto encoder to capture AC unit cycle patterns;- a link prediction model to learn potential links between patterns of the AC unit cycle on one hand and the possible failure of - or damage caused to - one of the AC unit components, on the other hand.

[0010] According to one or more embodiments the action performed responsive to the determination that a fault event took place is at least one among generation of an alarm and storage of data identifying the fault event on a storage medium.Another aspect concerns a device comprising a processor and memory storing software code, wherein the processor, when executing the software code, causes the device to carry out one of the methods described herein.BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Example embodiments will be more fully understood from the detailed description provided herein and the accompanying drawings, which are given by way of illustration only.Fig. l is a schematic diagram of a system according to one or more embodiments; Fig. 2 is a flowchart of a method according to one or more embodiments.DETAILED DESCRIPTION

[0012] Various exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, specific structural and functional details disclosed herein are merely representative for purposes of describing example embodiments. The exemplary embodiments may be embodied in many alternate forms and should not be construed as limited to only the embodiments set forth herein. It should be understood that there is no intent to limit example embodiments to the particular forms disclosed.

[0013] It should be appreciated by those skilled in the art that any functions, engines, block diagrams, flow diagrams, state transition diagrams, message sequence charts and / or flowcharts herein represent conceptual views of illustrative circuitry embodying the principles of the exemplary embodiments. Functions, actions or steps described herein can be implemented in hardware or software or any combination thereof, and ordering of these functions, actions or steps may be different from that presented. If implemented in software, the functions, blocks of the block diagrams and / or flowchart illustrations can be implemented e.g. using software code executed by a processor or a processing device.

[0014] In the present description, functional blocks denoted as “means configured to perform ...” a certain function are to be understood as functional blocks comprising circuitry that is adapted for performing or configured to perform a certain function. Moreover, any entity described herein as “means”, may be implemented as one or more distinct entities or within an entity providing additional functions. When provided by a processor, the functions may be provided by a single processor or several processors. Moreover, the term “processor” includes one or more of a digital signal processor, remote processor, graphical processing units (‘GPUs’), dedicated or generic circuitry, read only memory for storing software, random access memory, and non-volatile storage.

[0015] According to one or more exemplary embodiments, air conditioning (‘AC’) system failure detection is carried out based on air conditioner cycle pattern detection in the air conditioner power data. The AC unit power data provides valuable insights to the electrical performance history of the system. Cycle patterns which are either shorter or longer than nominal or normal cycle patterns are identified, and detection of failure or damage is performed on the identified patterns as described below.

[0016] A typical conditioner cycle is for example about 20-30 minutes. A normal cycle duration may vary as a function of different factors, such as seasonal changes, climate, indoor and outdoor temperatures, and air quality. A long AC compressor cycle or a frequent short compressor cycle represent a potential hazard to the entire airconditioning system. Such cycles may cause damage to the AC unit and also prevent it from working efficiently and from properly cooling the room.

[0017] According to one or more exemplary embodiments, power consumption data of the AC unit is used as source of information for cycle pattern analysis. A smart breaker connecting a power source to the AC unit can for example be used to obtain power, voltage or current data.

[0018] Fig. 1 is a schematic diagram of a system 100 comprising a device 101 which can carry out the method according to one or more embodiments. In the example illustrated by Fig. 1, device 101 comprises a processor 102, a working memory 103, a non-volatile memory 104, a user interface 105, a mass storage device 106, a series of interfaces 107 and an internal bus 108 linking the various components. Memory 104 holds software code. When the processor executes the software code, it causes the device to carry out the method according to one or more embodiments described herein. System 100 moreover comprises an AC unit 109. Components of AC unit 109 may typically comprise, in a fashion known per se, a compressor 110, an evaporator coil 111, a condenser 112, a fan 113, an air filter 114 and vents 115.

[0019] In the example of Fig. 1, the AC unit also comprises an AC unit controller 116. Controller 116 controls the AC unit 109 and communicates with device 101 through one of the interfaces 107. AC unit 109 is connected to an electrical power source 117.

[0020] System 100 also comprises a sensor for obtaining data representative of the power consumption of the AC unit 109, e.g. a current sensor 118 sensing the current provided by the electrical power source to the AC unit. Resulting time series data may for example be stored in mass storage 106.

[0021] The system of Fig. 1 comprises one or more sources of additional data impacting the functioning cycle of the AC unit. Such sources are descriptive of at least one among: external environmental conditions; internal environmental conditions;- thermal characteristics of the premises (room, house. . .) the temperature of which the AC unit is regulating; data descriptive of the AC unit, e.g. including but not limited to data descriptiveof the AC unit’s operation.

[0022] The additional data may be static or dynamic. Such data may be obtained in various ways, depending on its nature, and can be stored in mass storage 106. Such data may be obtained from one or more additional sensors.

[0023] Fig. 1 shows, as examples of additional sensors, a hygrometer 119 (inhouse and / or external), an external temperature sensor 120 measuring the temperature outside of the premises in which the AC unit 109 is placed and an in-house temperature sensor 121 measuring the temperature of the premises in which the temperature is regulated by AC unit 109.

[0024] Supposing the sensors are those shown in the non-limiting illustrative example of Fig. 1, collected data available for processing comprises: i) Data representative of the current used by the AC unit. ii) Data descriptive of internal environmental: internal temperature, humidity. . . The internal temperature shows the changes in temperature when the AC unit is operating. It is the temperature in the premises where the temperature is regulated by the air conditioning unit. This temperature shows in particular how effectively the AC unit can cool down the premises.When the air is humid, it contains a high amount of water vapor. This means that the AC unit has to work harder to remove that excess moisture and lower the temperature. High humidity can cause the AC unit to work harder, leading to decreased efficiency and increased energy consumption. Research shows that with an increase in humidity by just -10%, the AC unit’s efficiency can decrease by as much as 25-30%. High levels of indoor humidity can make the air feel warmer, causing the AC to work harder to cool the room. For example, having a running washing machine / dryer can increase indoor humidity. iii) Data descriptive of external environmental conditions can include one or more of the following: the season, a weather profile e.g. one or more among external temperature, humidity, air pressure, air quality, wind speed and direction, . . . The external temperature impacts the operation of the AC unit, since at higher temperatures, the room will warm faster and the AC unit will have longerand / or more frequent cycles. An exemplary AC unit might be able to handle a temperature difference of up to 20°C between the outside and the inside (room) at about 60% humidity. This means that when the thermostat is adjusted to keep the room cooler by about 20°C when it is more than 40°C outside, the AC unit is continuously running.A high humidity can negatively impact the operation of the AC unit and the AC unit itself. Humidity, both indoor and outdoor, plays a significant role in the performance of an AC unit and ageing of its components. High outdoor humidity can also affect the AC unit. The AC will have to work harder to remove the heat from the air as well as the heat-absorbing water vapors. If outdoor humidity is high, it can increase indoor humidity as well, especially if the building is not well insulated. This can force the AC unit to work even harder to maintain a comfortable indoor temperature. Air quality (e.g. presence of dust, pollutants, and other particulates) may affect the air filter and overall efficiency of the AC unit. Wind speed and direction may affect the heat exchange process in the outdoor unit of the AC unit.

[0025] Data can also include: iv) Thermal characteristics (or thermal profile) of the premises, which can for example include one or more among: age, level of thermal insulation, size / volume. v) Data descriptive of the AC unit impacting the AC unit’s performance, which can include one or more among: the installation date of AC can be a useful parameter (the age of the AC unit can be a reason for certain faults), the cumulative time of operation of the AC unit (indicative of possible ageing of one or more components). AC cycles may increase in duration, or an AC unit could run continuously because of an air leakage (e.g. through open doors / windows) and / or a low level of thermal insulation.

[0026] Data may be obtained for one or more sources other than a sensor, noting that some data may be obtained either using a sensor or from another type of source, e.g.a server, a non-transitory storage medium. . .

[0027] Furthermore, another parameter can be the geographical location of the AC unit (or of the premises, which would be the same), which enables obtaining local outdoor temperatures and humidity levels as a function of time from an external source, such as a weather station server.

[0028] Data may be static or dynamic. Static data (e.g. AC unit installation date, location. . .) can be obtained through the user interface, or also from an external server, or be programmed in device 101 at the time of installation of the AC unit. Data that is ‘static’ may still be updated when appropriate (e.g. the location, if the AC unit is moved to another location etc. . .). Data defining the current season may be derived from a calendar functionality of device 101.

[0029] The system 100, device 101 and AC unit 109 may comprise components differing from those shown in Fig. 1, depending on the implementation. While device 101 is illustrated as a separate device, it may be part of a larger device with additional functions, e.g., it may be integrated into the air conditioning unit 109. The functions performed by device 101 as described herein may also be split over several distinct devices. E.g. mass storage 106 may be a remote server, the user may interface with the device through a personal computer or a smart phone, etc. . .

[0030] According to one or more embodiments, detection of short and long cycles based on fault-related behaviour is performed, distinguishing these from non-harmful potential peaks.

[0031] Distinguishing short AC unit cycles due to damages or failure from normal peaks in time series data can be a challenging problem, especially when these two types of cycles have similar patterns. E.g. it can be difficult to distinguish between frequent short cycles and other regular peaks in the energy data of the AC unit and find the reasons of failure and the components failing in the AC unit (leakage in the refrigerant, fan motors, etc...). Similarly, a long duration AC unit cycle can lead to damage of the AC unit. It is challenging to distinguish a long duration cycle due to damage from the effectsof e.g. a seasonal increase of temperature or a reduction of the thermal insulation of the room. The AC unit can be working harder than usual in such expected cases and settings (or seasons), while this behaviour of an AC unit could also be indicative of possible damage to the AC unit.

[0032] According to one or more embodiments, use is made of statistical models and advanced data science approaches that can learn patterns in the data and distinguish between different types of signals based on insights such as trends, seasonality, and other features. A ‘trend’ refers to a long-term upward or downward movement in the data, indicating a general increase or decrease over time. The short or long AC unit cycle can eventually affect the patterns of the data. Investigating a trend allows to capture the direction and magnitude of change over a longer period.

[0033] Through running experiments and analyzing the AC unit power signature over time, it becomes possible to distinguish and learn rules which allow to reduce the time and effort needed to predict potential failure of the AC unit due to a damage. According to one or more embodiments, a combination is made of (a) machine and / or deep learning models with (b) use of a priori knowledge from the learned rules. Such rules are in practice based on identifying one or more parameters impacting operation of the AC unit among the additional data mentioned above.

[0034] According to a first embodiment, the machine and / or deep learning system(a) comprises neural networks based on sequence learning, convolution layer(s) and / or a variational auto encoder to capture patterns, and the a priori knowledge (b) is implemented using an expert system rule-based model to capture cases where an AC unit behavior is potentially susceptible of causing damage.

[0035] According to a second embodiment, the identification of patterns and of a failing AC unit or early-stage damaged AC unit uses (a) as a priori knowledge, semantic reasoning from AC unit concepts and characteristics using a knowledge graph and / or(b) as machine and / or deep learning model, a link prediction model to learn potential links between patterns of the AC unit cycle on one hand and the possible failure of- or damage caused to - one of the AC unit components, on the other hand.

[0036] The described approach allows:1) to learn from the AC data for a specific home, e.g. learning automatically from the home’s characteristics (old or new, level of thermal insulation, size of the home);2) taking into consideration parameters such as the season and the weather to learn the impact of weather conditions such as temperature on the AC unit (e.g. these factors may cause the AC unit to work harder than under normal conditions);3) taking into account the years-in-use of the AC unit (e.g. less than 5 years, between 5- 10 years, more than 10 years) as this impacts the performance of an AC unit, to help distinguish potential damage of an AC unit from expected behavior as a function of duration of use.

[0037] Fig. 2 is a diagram illustrating various stages of a method 200 according to one or more embodiments. In the example of Fig. 2, the method comprises an initialization stage A, a stage B at which a normal operational pattern is learned and a stage C at which abnormal operational patterns are identified.

[0038] During the initialization stage A, data is collected at substage A. l. According to one or more embodiments, this comprises the power consumption (e.g. current) time series data and additional data mentioned above. According to the embodiment illustrated by Fig. 2, at substage A.2, the data is stored for further processing.

[0039] During stage B, a normal operational pattern is learned, based on data obtained when the AC unit has normal operational behaviour. Learning can be performed applying various methods known per se. Purely as a non-limiting example, data may be preprocessed (e.g. handling missing data / values and / or normalizing the data), followed by extracting important 'patterns' (features). The data is split data into a training and test set. The ML model is trained on the training set. The trained model is evaluated on the test data set. This model is then used for predictions. In effect, the learning based model can be one among machine / deep learning models (any learning model from historical data, a neural network with pattern learning, sequential characteristics. . .).

[0040] At sub stage B.1, a rule-based model is combined with the learning based model mentioned in the previous paragraph. One non-limiting example of combination isa voting mechanism.

[0041] The rule based model is based on a priori knowledge (e.g.: Bayesian models and belief networks, rule based system, etc.) obtained from rules based on one or more identified main game changing parameters as defined in the additional data, e.g. home profile, weather characteristics and seasonality, useful lifetime of an AC unit or unit component. . .

[0042] In essence, rules are used to distinguish faulty from non-faulty behavior. Rules can be defined to help identify potential issues that might occur in the AC system. According to one embodiment, the rule based model uses a set of if-then rules to make decisions. These rules are predefined and manually set by experts in the field. A very simply rule may concern setting a status as ‘hot’ if the room temperature is higher than 30 degrees Celsius.

[0043] For example:- Rules can relate to the thermal conditions or the environmental conditions internal to the space in which the AC unit operates. E.g. such rules can use a priori knowledge about the house or space size, isolation condition if known and house occupancy level (number of people) that could help distinguish the behaviour of AC unit working harder than usual. This information is used in the rule-based model to enhance the learning and detection of faults by distinguishing faulty vs non-faulty behaviour, e.g. in a case in which the AC unit is just working harder to cool a place larger in size than what the AC unit is intended for.- Rules can be based on the operational behavior of an AC system. For instance, an AC can have normal, short, and long operational cycles. A normal operation cycle lasts about 10-20 minutes. The duration of short cycles depends on the AC and can be less than a few minutes, while long cycles can last more than half an hour to a few hours.- Rules can identify specific issues if certain conditions are met. E.g. issues such as compressor problems or refrigerant leakage can cause short cycle issues. If any abnormal operational patterns associated with electrical spikes are identified, then there could be issues with capacitors or other serious problems in the AC system.

[0044] During stage C, the combination of the rule based model with the learningbased model is used to detect abnormal AC unit cycle patterns. The model knows what a normal operational pattern of AC system is. Therefore, it can identify abnormal patterns, which are deviations from normal patterns. If the detected abnormal cycle patterns (‘events’) are judged statistically significant (substage C. l), the device 101 stores the corresponding events and / or generates an alarm (substage C.2). If the event is not statistically significant, then the method returns to substage B. l for improving the machine learning model. Storage of statistically significant events can for example be performed on a remote server, the cloud... According to one non-limiting example, statistical significance can be established as follows:The first step is to formulate a null hypothesis (HO) and an alternative hypothesis (Hl). The null hypothesis typically represents a theory that has been put forward. The alternative hypothesis is what might be believed to be true or hoped to proven true.For example, if a new model is tested, the null hypothesis might be “The AC has normal pattern” and the alternative hypothesis might be “The AC has abnormal pattern”. Initially, the null hypothesis is assumed true, and then data is collected to test whether there is enough evidence to accept the alternative hypothesis.A next step includes choosing a significance level, which is a threshold for the p- value below which the null hypothesis will be rejected. Commonly used values are 5% (0.05), 1% (0.01), and 0.1% (0.001). The p-value is the probability of obtaining the observed data (or more extreme data) if the null hypothesis is true.

[0045] If the p-value is less than the chosen significance level, then the null hypothesis is rejected in favor of the alternative hypothesis. If the p-value is larger than the significance level, then there is insufficient evidence to reject the null hypothesis. According to one or more embodiments, model improvement uses active learning techniques or similar data science techniques.

[0046] Modelling of fault detection in AC units may face a lack of data, i.e. the lack of sufficient fault data. According to a variant embodiment, generative artificial intelligence is used to augment the quantity of data.

Claims

CLAIMS1. A method (200) for detection of abnormal conditioning unit operational cycle patterns indicative of fault events in an air conditioning unit (109) for regulating a temperature in a space, said method being carried out by at least one processor (102), said method comprising:- obtaining (A.l) first data representative of the power consumption of the air conditioning unit as a function of time and second data representative of at least one parameter impacting the air conditioning unit’s performance;- combining (B.l) a rule-based model and a learning based model;- applying the combined models for the identification (C) of an abnormal air conditioning unit operational cycle pattern as a function of the first and second data;- in case of identification of an abnormal pattern, determining (Cl) whether the identified abnormal cycle pattern is statistically significant, and in the affirmative, determining that a fault event took place and performing an action responsive to said determination.

2. The method of claim 1, wherein said data descriptive of the at least one parameter comprises at least one among:- data descriptive of external environmental conditions;- environmental conditions internal to the space;- thermal characteristics of the space; and / or- data descriptive of the AC unit.

3. The method according to claim 2, wherein- data descriptive of external environmental conditions comprises one or more among season, temperature, humidity, air pressure, air quality;- environmental conditions internal to the space comprises one or more among temperature, humidity;- thermal characteristics of the space comprises one or more among level of insulation, size, volume, age, occupancy level;- data descriptive of the AC unit comprises one or more among date of installation, cumulated duration of operation, geographical location.

4. The method according to one of the claims 1 to 3, wherein said rule-based model is one among:- an expert rule-based model descriptive of cases where an AC unit behavior is potentially susceptible of causing damage;- a knowledge graph providing semantic reasoning from AC unit concepts and characteristics;- one or more Bayesian models;- one or more belief networks.

5. The method according to one of the claims 1 to 4, wherein the learning model is one among:- a machine and / or a deep learning model comprising at least one neural network based on sequence learning, one or more convolution layers and / or a variational auto encoder to capture AC unit cycle patterns;- a link prediction model to learn potential links between patterns of the AC unit cycle on one hand and the possible failure of - or damage caused to - one of the AC unit components, on the other hand.

6. The method according to one of the claims 1 to 5, wherein the action performed responsive to the determination that a fault event took place is at least one among (C.2) generation of an alarm and storage of data identifying the fault event on a storage medium.

7. A device (101) comprising a processor (102) and memory (104), said memory comprising software code for execution by said processor, wherein the processor, when executing the software code, causes the device to carry out a method according to one of the claims 1 to 6.

Images