JP7032261B2 - Anomaly detection system and anomaly detection method - Google Patents

Description

The present invention relates to an abnormality detection system and an abnormality detection method, and more particularly to an abnormality detection system for detecting an abnormality in an air conditioner and a method thereof.

Anomaly detection, which detects air conditioner failures or signs thereof, is extremely important for improving production efficiency and reducing inspection work and maintenance costs. Recently, there has been a movement to commercialize a service that not only provides air-conditioning equipment to customers, but also collectively undertakes the operation and maintenance of the equipment. In addition, for commercial air conditioners and refrigerating and refrigerating equipment, the Freon Emission Control Law, which came into effect in April 2015, requires users to perform a simple inspection of refrigerant leaks. For this reason, it has become important in the service business to grasp the abnormality of the air conditioner.

Regarding abnormality detection of equipment, for example, in Patent Document 1, an abnormality measure is calculated based on a feature vector created by processing sensor signals obtained in time series from a plurality of sensors mounted on equipment or devices. , It is possible to judge the rising tendency of the abnormal measure, detect the continuous section of the abnormal measure rising tendency, specify the related sensor for each section and judge the presence or absence of the tendency, and detect the abnormal measure rising tendency and its An abnormal tendency detection system that realizes identification of related sensors with a small computer load is disclosed.

Japanese Unexamined Patent Publication No. 2016-58010 (Patent No. 6223936)

The technology described in Patent Document 1 includes not only equipment such as gas turbines and steam turbines, but also railroad vehicles, tracks, escalator, elevators, factory production equipment, and at the equipment / component level, deterioration and life of on-board batteries. Alternatively, in sensing data for humans such as brain waves and electrocardiograms, it is possible to detect not only the presence or absence of abnormalities in the target but also the tendency that the degree of abnormality gradually worsens at an early stage (paragraph 0013). ..

However, Patent Document 1 does not suggest a special problem regarding abnormality detection when applied to an air conditioner. The abnormality detection described in Patent Document 1 aims to identify an abnormality sensor based on a plurality of sensor signals. Even if this abnormality detection technology is applied to an air conditioner, it would be possible to identify the abnormality sensor, but it is difficult to easily grasp the abnormality in the air conditioning function.

An object of the present invention is to provide an abnormality detection system and a method capable of easily grasping an abnormality in air conditioning function. Furthermore, the present invention is to detect an abnormality in the air conditioner based on the air conditioning functional force according to the air conditioner refrigerant.

According to a preferred example, the abnormality detection system according to the present invention has a database that stores a plurality of arithmetic expressions representing the air conditioning functional force according to the air conditioner refrigerant, and a sensor signal acquired from a sensor that detects the state of the air conditioner. Has a processing device that processes information using
The processing device is
For the sensor signal, a first processing unit that selects a plurality of arithmetic expressions representing the air conditioning functional power according to the air conditioner refrigerant from the database, calculates the capacity value, and regards this capacity value as normal data to learn.
For the sensor signal acquired by the first processing unit in a period different from the learning period of the air conditioner, a plurality of arithmetic expressions representing the air conditioning functional force corresponding to the air conditioner refrigerant are selected, the capacity value is calculated, and the capacity value is calculated. It has a first processing unit that detects abnormalities based on the learned normal ability value data.
When the second processing unit detects an abnormality, it is configured as an abnormality detection system characterized by specifying an abnormality-related ability value and outputting information on the abnormal state.
The present invention is also understood as an abnormality detection method in the abnormality detection system.

According to the present invention, a plurality of air-conditioning function power values are calculated using an arithmetic formula representing the air-conditioning function power according to the air conditioner refrigerant, and learning and abnormality detection are performed based on the capacity values to obtain the air-conditioning function power. It is possible to easily grasp the abnormality.

It is a figure which shows the structure of the abnormality detection system of the air conditioner by one Example. It is a figure which shows the calculation formula which calculates the air-conditioning function power value for each type of an air conditioner refrigerant. It is a figure which shows the calculation formula which calculates the air-conditioning function power value for each type of an air conditioner refrigerant. It is a figure which shows the processing flow which learns a normal air-conditioning function power value. It is a figure which shows the example of the 2D distribution density graph of normal data. It is a figure which shows the processing flow which specifies the abnormality-related capacity value with respect to the capacity value of the evaluation air conditioner based on the normal air-conditioning function capacity value. It is a figure which shows the processing flow (details of the processing flow of S513) which creates the 2D distribution density of normal data by the combination of a plurality of air-conditioning function power values. It is a figure which shows the processing flow (detailed processing flow of S514) which specifies an abnormality-related ability value using a two-dimensional distribution density, and determines a state. It is a figure which shows the structure of the air-conditioning function power value state table. It is a figure which shows the structure of each ability value state table. It is a figure which shows the example of the screen which displays the abnormal state of an air conditioner. It is a figure which shows the example of the screen which displays the abnormality detection result of the air conditioner, and the abnormality-related capacity value state. It is a figure which shows the outline of the abnormality detection system of an air conditioner. It is a figure which shows the structure of the abnormality detection system of the air conditioner by Example 2. FIG. It is a figure which shows the flow which estimates the abnormal state with respect to the capacity value of the evaluation air conditioner based on the normal air-conditioning function power value. It is a figure which shows the example which the state of the air-conditioning function power value is made into a database for each abnormal state, and the abnormal state is estimated by the degree of agreement with the abnormality-related capacity value. It is a figure which shows the example of the screen which displays the abnormal air conditioner and the abnormal state.

Hereinafter, preferred embodiments of the present invention will be described with reference to the drawings.

FIG. 1 shows the configuration of an abnormality detection system for an air conditioner. The schematic configuration is shown in FIG.
The abnormality detection system of the air conditioner includes an air conditioner 101 installed at a plurality of customer sites 900 and a data center 100 connected to the air conditioner 101 via a network. The plurality of air conditioners 101 installed at the customer site 900 include various air conditioner refrigerants of different models, manufacturers, and the like. In each air conditioner 101, for example, one or a plurality of various sensors 102 such as a temperature sensor, a current sensor, and a pressure sensor are installed. The data center 110 can detect an abnormality of a plurality of air conditioners installed at a plurality of customer sites by using a plurality of sensor signals 1021 acquired from these sensors 102.

The data center 100 has one or more servers 110 and a database (DB) 111 that stores various information. When the data center format is not used, one or more servers as a data processing device acquire the sensor signal 1021 and process the data.

Although the hardware configuration of the server 110 is not shown, the server 110 stores a processing device (CPU: Central Processing Unit) that executes a program and processes various data, a program, and various data, and stores the database 111. A storage device such as a memory or HDD (Hard Disk Drive) that can be constructed, an input device for inputting data and various instructions, a display device for displaying various data, and a server for customer site 900 or other devices. It is configured to have a communication unit that connects to an external device. The input device and the display device may be configured as an input display function of the terminal, and the terminal may be connected to the communication unit of the server 110 via a network.

The server 110 obtains the air conditioning function force value based on the sensor signal of the air conditioner, and performs learning and abnormality detection. Therefore, when the CPU of the server 110 executes the program, the functions of the learning unit 120, the evaluation unit 130, and the data processing capacity value processing unit 140 are realized. The database 111 stores an arithmetic expression representing the air conditioning functional force according to the refrigerant of the air conditioner. (The details will be described later with reference to FIGS. 2A and 2B.) In the following description, from the viewpoint of the processing process, the learning unit 120, the evaluation unit 130, and the data processing capacity value processing unit 140 are the learning modes, respectively. It may be called an evaluation mode or a data processing ability value processing mode.

The data processing capacity value processing unit 140 selects a plurality of arithmetic expressions representing the air conditioning functional power corresponding to the target air conditioner refrigerant from the database 111, and calculates the capacity value. The learning unit 120 learns the normal ability value data using the ability value created by the data processing ability value processing unit 140, and includes the ability value input unit 121, the feature vector extraction unit 122, and the abnormality measurement calculation. It has a unit 123 and a threshold calculation unit 124. The evaluation unit 130 performs abnormality detection diagnosis based on the capacity value from the data processing capacity value processing unit 140, and includes the capacity value input unit 131, the feature vector extraction unit 132, the abnormality measure calculation unit 133, and the abnormality. It is configured to have a determination unit 134 and an abnormality-related ability value specifying unit 135. In the illustrated example, the learning unit 120 and the evaluation unit 130 are provided with an ability value input unit, a feature vector extraction unit, and an abnormality measure calculation unit, respectively, but these functions are shared and one ability value is used. It may be configured as an input unit, a feature vector extraction unit, and an anomaly measure calculation unit.

Next, the functions of the learning unit 120, the evaluation unit 130, and the data processing ability value processing unit 140 will be described in detail.
The data processing capacity value processing unit 140 was selected by selecting an arithmetic expression representing the air conditioning functional force according to the refrigerant of the air conditioner 101 stored in the database 111 according to the refrigerant detected by the sensor signal 1021. The capacity value 113 is calculated based on the calculation formula and the sensor signal 1021. This calculation will be described later.

In the learning unit 120, the ability value input unit 121 receives the input of the ability value 113 calculated by the data processing ability value processing unit 140, and the feature vector extraction unit 122 extracts the feature vector using the ability value 113. The anomaly measure calculation unit 123 calculates an anomaly measure for each feature vector at predetermined time intervals (sometimes expressed as each time) using a feature vector for a learning period specified in advance. The threshold value calculation unit 124 calculates a threshold value according to the abnormality measure.

In the evaluation unit 130, the capacity value input unit 131 receives the input of the capacity value 113 calculated by the data processing capacity value processing unit 140, and the feature vector extraction unit 132 extracts the feature vector using the capacity value 113. The anomaly measure calculation unit 133 calculates an anomaly measure for each feature vector at predetermined time intervals using a feature vector for a learning period specified in advance. The abnormality determination unit 134 determines the abnormality by comparing the calculated abnormality measure with the threshold value 124 obtained from the abnormality measure of the learning data. The anomaly-related ability value specifying unit 135 identifies the ability value related to the anomaly.

Next, the functions of the feature vector extraction units 122 and 132, the abnormality measure calculation units 123 and 133, the threshold value calculation unit 124, the abnormality determination unit 134, and the abnormality-related ability value identification unit 135 will be described.

The feature vector extraction units 122 and 132 perform normalization with the mean being 0 and the variance being 1 for the sensor signals. Then, for a certain time t (t = 0, 1, 2, ...), One vector x (t) is extracted per time as shown in Equation 1.

[Equation 1] x (t) = (x ₁ (t), x ₂ (t),…, x _M-1 (t), x _M (t)) ^T
Here, x _m (t) (m = 1,2, ..., M) is the m-th sensor signal at time t after normalization.

The anomaly measure calculation units 123 and 133 employ a local subspace classifier (LSC). LSC creates a k-1 dimensional subspace using k neighborhood data xi (i = 1, ..., k) for unknown data. Then, it is a method of determining whether or not it is abnormal based on the unknown data and the projection distance of the k-1 dimensional subspace. Since the anomaly measure is represented by the projection distance between the unknown data and the k-1 dimensional subspace, it is sufficient to find the point on the subspace closest to the unknown data.

The anomaly measure calculation unit 123, 133 performs anomaly measurement by calculating the square root of the sum of squares of the difference between the unknown data and the point on the subspace closest to it. Further, in order to appropriately determine the threshold value for determining whether or not the abnormality is abnormal in the abnormality measurement, the threshold value calculation unit 124 is automatically set by cross-validation of learning data. First, the anomaly measure calculation unit 123 and 133 divides the learning data into N groups equally, and calculates the anomaly measure of each data by LSC using the learning data excluding the group for each group.

The threshold value calculation unit 124 sorts the abnormality measures calculated by the abnormality measure calculation unit 123 in ascending / descending order, and sets the value of the abnormality measure at which the ratio to the total number of data reaches the parameter value of 0 to 1 specified in advance as the threshold value. If it is assumed that the training data does not contain any anomalies, set the parameter value to 1. In this case, the threshold is the maximum value of the anomaly measure.

The abnormality determination unit 134 determines whether or not the abnormality measure calculated by the abnormality measure calculation unit 123 and 133 exceeds the threshold value set by the threshold value calculation unit 124, and if it is determined that the abnormality exceeds the threshold value, it is considered to be abnormal. Judge and judge that everything else is normal.

The anomaly-related ability value specifying unit 135 obtains an abnormality section in which anomalies are continuously detected, and extracts an abnormality-related ability value for each section based on the two-dimensional distribution density. For that purpose, at the time of learning, the two-dimensional feature distribution density of the normal data based on the air conditioning function power value is calculated by the roundabout of the two types of air conditioning function power values and saved in the image format. Furthermore, the degree of isolation of the sensor is calculated according to where the abnormality data of the air conditioning function force value detected at the time of abnormality detection is plotted in the two-dimensional distribution density image, and the degree of isolation is extracted as the abnormality-related ability value in order from the one with the largest degree of isolation. .. This means finding anomalous-related stats whose distribution of anomalous data deviates from the distribution of normal data.

2A and 2B are diagrams showing arithmetic expressions for calculating the air conditioning functional force value for each type of air conditioner refrigerant in the cooling mode and the heating mode.
As shown in FIG. 2A, examples of the air-conditioning function force in the cooling mode include cooling capacity, compressor outlet side refrigerant state, compressor inlet side refrigerant state, outdoor heat exchanger performance, indoor heat exchanger performance, and the like. The cooling capacity is the difference between the suction temperature and the outlet temperature, the compressor outlet side refrigerant state is the difference between the compressor upper temperature and the high pressure pressure temperature conversion, and the compressor inlet side refrigerant state is the difference between the gas pipe temperature and the low pressure pressure temperature conversion. The outdoor heat exchanger performance is the difference between the high-pressure pressure temperature conversion and the outside air temperature divided by the cooling capacity value, and the indoor heat exchanger performance is the difference between the blowout temperature and the low-pressure pressure temperature conversion divided by the cooling capacity value. be. The state of the refrigerant on the outlet side of the compressor requires an arithmetic expression for the number of compressors.

As shown in FIG. 2B, the heating function of the heating mode includes, for example, heating capacity, compressor outlet side refrigerant state, compressor inlet side refrigerant state, outdoor heat exchanger performance, indoor heat exchanger performance, and the like. The heating capacity is the difference between the outlet temperature and the suction temperature, the compressor outlet side refrigerant state is the difference between the compressor upper temperature and the high pressure pressure temperature conversion, and the compressor inlet side refrigerant state is the difference between the gas pipe temperature and the low pressure pressure temperature conversion. The outdoor heat exchanger performance is the difference between the high-pressure pressure temperature conversion and the outside air temperature divided by the heating capacity value, and the indoor heat exchanger performance is the difference between the outlet temperature and the low-pressure pressure temperature conversion divided by the heating capacity value. be. The state of the refrigerant on the outlet side of the compressor requires an arithmetic expression for the number of compressors.
The cooling / heating capacity, refrigerant state, etc. may be referred to as an air conditioning function type. This air conditioning function type can later be displayed on the display screen together with the abnormal state of the air conditioner.
Further, in the database 111 that stores the calculation formula representing the air conditioning function force according to the air conditioner refrigerant, the new type calculation formula can be additionally defined, and can be applied to the construction of learning data and the detection of abnormality. ..

Next, with reference to FIG. 3, a process for learning a normal air conditioning function force value will be described.
First, the sensor signal 1021 of the target air conditioner 101 is input to the data processing capacity value processing unit 140 (S301). Further, the calculation formulas of the cooling mode and the heating mode representing the air conditioning functional force of the air conditioner 101 are selected from the database 111 and input to the data processing capacity value processing unit 140 (S302).

The data processing capacity value processing unit 140 repeats the following processing for each cooling mode or heating mode (S303). If necessary, it is converted into a sensor signal value considering these modes (S304). Then, the capacity value 113 is calculated using the arithmetic expression representing the air conditioning functional force selected in S302 (S305). Only those that match the desired operating operation conditions are extracted from the air conditioning functional force value 113 (S306), and averaged by the exponential moving average (S307).

Then, the capacity value input unit 121 of the learning unit 120 repeats the following processing for the air-conditioning function power value to which the exponential moving average is applied, as many as the number of two-dimensional distribution densities of each capacity value (S308). As shown in FIG. 4 (1), a two-dimensional distribution density 400 of normal data is created by combining a plurality of ability values (S309). Then, as shown in FIG. 4 (2), the maximum value (MAX) for each type of air conditioning functional force value is obtained so that the vertical and horizontal scales of the two-dimensional distribution density are the same by the brute force processing for two air conditioning functional force values. ) And the minimum value (MIN). Further, for the normal air-conditioning function power value 401, the maximum value and the minimum value of each of the columns and rows are also obtained (S310).

Specifically, the detection range of the same type of sensor may differ depending on the environment in which the air conditioner 101 is installed. For example, in the case of a temperature sensor, one air conditioner 101 is set to −10 ° C. to + 40 ° C., while another learning air conditioner 101 is set to −5 ° C. to + 45 ° C. The capacity value of the air conditioner can be considered to be the same. Therefore, first, the maximum value and the minimum value for each type of air conditioning function power value are obtained, and the following processing is executed.

Subsequently, the step size S when the maximum value is divided by the designated number N from the minimum value is calculated. The step size S can be calculated by, for example, S = (MAX-MIN) / N.
Further, in order to provide a preliminary processing range, the processing range for calculating the distribution density is calculated by expanding the range outward from the above minimum and maximum values. The range to be expanded is, for example, MIN is changed to MIN-S × M and MAX is changed to MAX + S × M. Here, M is a predetermined integer of 1 or more. Subsequently, the bin number (BNO) is calculated from the feature value (F) for all the data in the learning period by the formula 2.

[Equation 2] BNO = INT ((F-MIN) / (MAX-MIN))
However, INT (X) represents the integer part of X.

In order to create a two-dimensional distribution density by round-robin processing of two air-conditioning function power values, the following processing is repeated for each air-conditioning function power value in an arbitrary learning period. Brute force includes the case where two air conditioning function power values are of the same type. A two-dimensional array for calculating the two-dimensional distribution density is secured, and 0 is set for all elements. The size of the array is N + 2M. For all air conditioning function force value data during the learning period, 1 is added to the elements of the array corresponding to the bin numbers of the two sensor values. By this processing, a two-dimensional frequency distribution (histogram) with two air conditioning functional force values is calculated. This frequency distribution is converted into an image, and the converted image and an identification number (ID) for identifying the image are stored in association with each other. The conversion method will be described later. Record the maximum and minimum values of the columns and rows for the size of the two-dimensional array, the calculated minimum and maximum values of each air conditioning function, and the normal air conditioning function values. This processing flow will be described with reference to FIG.

An example of the above image conversion method will be described. Find the maximum value of the array elements, that is, the maximum frequency. The image size is the same as the array size, and the pixel value of the corresponding coordinate from the value of each element is, for example, 255 × the element value / maximum frequency of the array. 255 is the maximum value when the pixel value is represented by 8 bits, and if this value is used, it can be saved as it is in the bitmap format.
The image obtained by the above processing is referred to as a distributed density image because the high density part in the two-dimensional feature space is represented by a high pixel value.

In FIG. 4 (1), 0 of the pixel value of the two-dimensional distribution density 400 is represented by white, the maximum is represented by black, and the space between them is represented by shades of gray 401. However, the method of creating an image is not limited to the above method. For example, instead of a simple frequency distribution, a Gaussian distribution or another weighted filter may be assigned to one piece of data and superposed on the data. Alternatively, the image obtained by the above method may be subjected to a maximum value filter of a predetermined size, an average filter, or another weighted filter. Further, it is not always necessary to save in the image format, and the two-dimensional array may be saved in the text format. Further, the pixel value is not a shade of gray, and may be saved in a text format of a binarized two-dimensional array in which the pixel with distribution is 1 and the pixel without distribution is 0.

FIG. 4 (2) shows, as an example, a two-dimensional distribution density image of a capacity value a (cooling capacity) and a capacity value b (compressor outlet side refrigerant state) provided in a certain air conditioner 101. The horizontal axis is the minimum to maximum value of the scaled capacity value a, the vertical axis is the minimum to maximum value of the scaled capacity value b, and the compressor outlet side refrigerant state p1 at the cooling capacity t1 is (X1, It is plotted as Y1), and the compressor outlet side refrigerant state p2 at the cooling capacity t2 is plotted as (X2, Y2). The created two-dimensional distribution density image is provided with identification information S0001A0001 (S0001: customer site identification number, A0001: air conditioner identification number) for identifying the customer site and the air conditioner. In this way, based on all the sensors 102 provided in the air conditioner 101, the air conditioning functional force value is calculated by using the calculation formula expressing the air conditioning functional force according to the refrigerant, and the two air conditioning functional force values are calculated. For the combination, such an imaged two-dimensional distribution density graph is generated and associated with the identification number. It can be said that the two-dimensional distribution density image created by the above processing is a graph showing the correlation between the output values of the two types of air conditioning function power values.

Subsequently, the ability value 113 is input to the feature vector extraction unit 122, and the feature vector extraction unit 122 extracts the feature vector using the ability value 113 (S311). Then, the anomaly measure calculation unit 123 calculates an anomaly measure for each feature vector for each predetermined time (hereinafter, may be expressed as each time) using the feature vector for the learning period designated in advance (S312). Finally, the threshold value calculation unit 124 calculates a threshold value according to the abnormality measure of the calculated learning data (S313).

Next, with reference to FIG. 5, a process for specifying an abnormality-related capacity value with respect to the capacity value of the target air conditioner based on the normal air conditioning function capacity value will be described.
First, the sensor signal 1021 of the target air conditioner 101 is input to the data processing capacity value processing unit 140 (S501). Further, the calculation formulas of the cooling mode and the heating mode representing the air conditioning functional force of the target air conditioner 101 are selected from the database 111 and input to the data processing capacity value processing unit 140 (S502). The data processing capacity value processing unit 140 repeats the following processing for each cooling mode or heating mode (S503). If necessary, it is converted into a sensor signal value considering the mode (S504). Then, the air conditioning functional force value 113 is calculated using the arithmetic expression representing the air conditioning functional force selected in S502 (S505). Only those that match the desired operating operation conditions are extracted from the air conditioning functional force value 113 (S506), and averaged by the exponential moving average (S507).

Subsequently, the feature vector extraction unit 132 extracts a feature vector using the averaged ability value 113 (S508). Then, the anomaly measure calculation unit 133 calculates an anomaly measure for each feature vector at predetermined time intervals (hereinafter, may be expressed as each time) using the feature vector of the learning period designated in advance (S509). Then, the abnormality determination unit 134 determines the abnormality measure of the air conditioner 101 based on the threshold value calculated by the threshold value calculation unit 124 (S510).

Next, in the abnormality-related ability value specifying unit 135, the following processing is repeated for each abnormality data (S511) and the two-dimensional distribution density (S512) of each ability value. As shown in FIG. 4 (1), a two-dimensional distribution density 400 of normal data is created by combining a plurality of ability values (S513). (Details of S513 will be described later with reference to FIG. 6).

Then, the maximum value (MAX) and the minimum value (MIN) for each type of air conditioning function power value are obtained so that the vertical and horizontal scales of the two-dimensional distribution density are the same by brute force processing for each of the two air conditioning function power values. .. Further, the maximum value and the minimum value of each of the columns and rows are also obtained for the normal air conditioning function power value 401 (S514). (Details of S514 will be described later with reference to FIG. 7).
The abnormal state of each ability value can be analyzed (S515), the abnormality-related ability value can be specified, and the ability value state can be obtained (S516).

FIG. 6 is a diagram showing a processing flow for creating a two-dimensional distribution density of normal data by combining a plurality of air conditioning functional force values. This process is a process flow showing the details of S513 in FIG. In this process, the anomaly-related ability value specifying unit 135 performs the ability value a (horizontal axis) and each ability value a (horizontal axis) in each ability value b (vertical axis) of the two-dimensional distribution density shown in FIG. 4 (1). ), The maximum and minimum values of the ability value b (vertical axis) are obtained.

The following processing is repeated for the number of two-dimensional distribution densities (S601) due to the combination of ability values.
First, the two-dimensional distribution density with respect to the capacity value b (vertical axis) (S602) and the capacity value a (horizontal axis) (S603) in which the normal air conditioning function power value 401 data shown in FIG. 4 (1) exists. The process of calculating the maximum value and the minimum value of the ability value a (horizontal axis) in each ability value b (vertical axis) of is repeated (S604). Finally, the maximum and minimum values of the capacity value a (horizontal axis) are also obtained for the normal air conditioning function power value 401 (S605).

Next, for the capacity value a (horizontal axis) (S606) and the capacity value b (vertical axis) (S607) in which normal air conditioning function power value 401 data exists, each capacity value a (2D distribution density) The process of calculating the maximum value and the minimum value of the ability value b (vertical axis) on the horizontal axis) is repeated (S608). Finally, the maximum and minimum values of the capacity value b (vertical axis) are also obtained for the normal air conditioning function power value 401 (S609).

FIG. 7 shows a processing flow diagram for identifying anomalous-related ability values using a two-dimensional distribution density and determining a state. This process is a process flow showing the details of S514 in FIG. In this process, the abnormality-related capacity value specifying unit 135 uses the maximum and minimum values obtained from the two-dimensional distribution density of the normal air-conditioning function power value 401 shown in FIG. 4 (1) to state the air-conditioning function power value. To judge.

For the number of abnormal air-conditioning function power values (S701) that deviate from the normal air-conditioning function power value 401 (learning) data shown in FIG. 4 (1), and the number of two-dimensional distribution densities (S702) due to the combination of capacity values. The following process is repeated.
First, a two-dimensional distribution density is created by combining the air conditioning functional force values of the abnormal data (S703). For the capacity value b (vertical axis) (S704) and the capacity value a (horizontal axis) (S705), the maximum and minimum values obtained from the two-dimensional distribution density of the normal air conditioning function power value 401 are used. The process of determining the state of the air conditioning function power value is repeated (S706). At this time, with respect to the abnormal ability value a (horizontal axis), when it is higher than the maximum value obtained in S604 and S605, when it is lower than the minimum value, and when it is between the maximum value and the minimum value and is normal. (S707).

Next, for the capacity value a (horizontal axis) (S708) and the capacity value b (vertical axis) (S709), the maximum and minimum values obtained from the two-dimensional distribution density of the normal air conditioning function power value 401 are calculated. The process of determining the state of the air conditioning function force value is repeated (S710). At this time, with respect to the abnormal ability value b (vertical axis), when it is higher than the maximum value obtained in S608 and S609, when it is lower than the minimum value, and when it is between the maximum value and the minimum value and is normal. It is classified into three types (S711). The state of the air-conditioning functional force value for each of the two types of the obtained two-dimensional distribution density is stored in the air-conditioning functional force value state table 800.

FIG. 8 is a diagram showing the configuration of two types of air conditioning function force value state tables 800 using the two-dimensional distribution density. The air-conditioning function power value status table 800 is stored in the database 111, and the abnormal states 802, 803, 804 determined by the processing of S704 to S711 are stored in the database 111 for each time-series abnormality data 801 in a two-dimensional distribution density 812,833,814. Register for each. Here, T abnormal data from t = 1 to t = T may be referred to as an ability value pattern.

An abnormal state using the two-dimensional distribution density is determined by the combination of these air conditioning functional force values (S712). At this time, since the two-dimensional distribution density is a combination of two types of air conditioning function power values, it is converted into an abnormal state for each type of air conditioning function power value.

FIG. 9 is a diagram showing the configuration of each capacity value state table 900. Each capacity value state table 900 is stored in the database 111, and the state converted from two types of air conditioning function power values using the two-dimensional distribution density into one type of air conditioning function power value state is registered.

In the process of FIG. 7, the abnormal states 802, 803, 804 of the two-dimensional distribution density 812, 815, 814 are converted into the respective air conditioning functional force value states 902, 903, 904 (S713). As a conversion method, a method of counting with a histogram or the like can be considered. Finally, the abnormality-related ability value can be specified and the ability value state can be obtained (S714). The time series interval may be various, such as 1 minute, 5 minutes, and 1 hour unit. The calculated abnormality-related data is displayed on the display screen.

FIG. 10 is an example of a screen displaying an abnormal state of the air conditioner.
The screen 1001 is displayed on the screen of the display device connected to the server 110. The screen 1001 includes a customer site display 1002, an air conditioner display 1003, an evaluation period display 1004, an abnormality sign detection result graph 1005, and an abnormality state display 1006. This screen is created by a screen creation unit (not shown) arranged after the abnormality-related ability value specifying unit 135, which collects and edits related data from the above functional parts.

Normal data is learned for each air conditioner type 1003 of the customer site 1002 displayed on the screen 1001, and is stored in the database 111. On the screen 1001, the customer site 1002 and the air conditioner type 1003 are selected, and the evaluation period 1004 is set by the input device. When the start button 1007 on the input device is pressed, the server 110 starts the process shown in FIG. After determining the abnormal state, the abnormality sign detection result graph 1005 in the evaluation period, the abnormality-related ability value, and the abnormal state 1006 are displayed. The anomaly-related ability value and its abnormal state are extracted from each ability value state table 900. A plurality of abnormality-related ability values and their abnormal states 1006 can be displayed.

FIG. 11 is an example of a screen displaying the abnormality detection result of the air conditioner and the abnormality-related capacity value state. This screen is also created by the screen creation unit (not shown). When the line of the abnormality-related ability value of the screen 1001 and the abnormality state 1006 shown in FIG. 10 is selected from the input device, the screen 1101 is displayed. On the screen 1101, the abnormality detection result 1102 of the air conditioner, the learning result 1103 of the abnormality-related ability value, and the evaluation result 1104 are displayed.
In the graph of the abnormality detection result, the abnormality measure A is calculated by the abnormality measure calculation unit 133, the threshold value B is calculated by the threshold value calculation unit 124, and the abnormality detection C is obtained from the abnormality determination unit 134. There is one or more anomaly-related ability values and their abnormal state 1006 numbers.

As described above, according to the first embodiment, a plurality of arithmetic expressions representing the air conditioning functional power according to the air conditioner refrigerant are defined, and learning and abnormality detection are performed based on the plurality of air conditioning functional power values to obtain the air conditioning function. It becomes possible to easily grasp the abnormality of the force. In addition, by considering the mode, extracting operation data, and applying the exponential moving average, it is expected that the sensitivity of abnormality sign detection will be increased and the false detection will be reduced.

The abnormality detection system according to the second embodiment will be described with reference to FIGS. 13 to 16. FIG. 13 shows the configuration of the abnormality detection system according to the second embodiment. FIG. 14 shows a processing flow for specifying an abnormality-related capacity value with respect to the capacity value of the evaluation air conditioner based on the normal air conditioning function capacity value.

The difference from Example 1 (FIG. 1) is that the abnormal state database (DB) 114 and the abnormal state estimation unit 136 have been added. That is, when the abnormality-related ability value specifying unit 135 specifies the ability value related to the abnormality, the abnormality state data related thereto is stored in the abnormality state DB 114. Further, the abnormal state estimation unit 136 performs the abnormal state estimation process using the abnormal state data (FIG. 15) stored in the abnormal state DB 114. Further, the difference from the processing step (FIG. 5) is that the processing loop 4 (processing S1317 to S1319 of the abnormal state estimation unit 136) of FIG. 14 is added.
In the following description, the description overlapping with FIGS. 1 and 5 will be omitted.

In the processing flow shown in FIG. 14, after the processing of S516, finally, the following processing is repeated for the number of abnormal states (S1317) stored in the database in the abnormal state DB 114.
As shown in FIG. 15, the abnormal state DB 114 stores an evaluation 1404 representing the capacity value state 1402 of the air conditioner, the abnormality-related capacity value number 1403, the matching capacity value number 14041, and the matching ratio 14042 for each type of abnormal state 1401. .. The capacity value state 1402 of the air conditioner is stored in advance in the abnormal state DB 114 corresponding to the type 1401 of the abnormal state. For example, here, when the number of abnormality-related ability values 1403 is "5", it means that five ability values are abnormal, such as "ability value a is high" and ... "ability value z is low". means. And, when the number of matching capacity values 14041 is "3/5", it means that "3" abnormal values correspond to the air conditioner evaluated this time among the five abnormal capacity values. As a result, it is shown that the matching ratio 14042 is "60"%.

The abnormal ability value state 1402 is data obtained from each ability value state table 900 shown in FIG. Then, as shown in FIG. 9, the abnormal capacity value state of the evaluation air conditioner 1402 is stored in the abnormality-related capacity value specifying unit 135, and the abnormal matching capacity value number 14041 is also obtained. The number of abnormality-related capacity values 1404 of the abnormal state 1403 of the abnormal state DB 114 is compared with the number of abnormal matching capacity values 14041 of the evaluation air conditioner, and the "match ratio" of the evaluation is calculated from the number of matching abnormal capacity values 1406. (S1318). Perform these steps for the number of abnormal state types 1401. Next, the abnormal state estimation unit 136 estimates the abnormal state by ranking in descending order of the degree of matching using the matching ratio 14042 calculated above (S1319).

FIG. 16 is an example of an abnormal air conditioner and a screen for displaying the abnormal state.
Compared to the screen shown in FIG. 10 of the first embodiment, a table 1507 showing a presumed abnormal state is added to the screen 1001.
By operating the input device to select the abnormality-related ability value of the screen 1001 and the number of the abnormality state 1006, the screen 1101 shown in FIG. 11 is displayed. On the screen 1101, the abnormality detection result 1102 of the air conditioner, the learning result 1103 of the abnormality-related ability value, and the evaluation result 1104 are displayed.

As described above, according to the second embodiment, in addition to the operation and effect of the first embodiment, the abnormal state and the failure state (gas leak, compressor) of the air conditioner are used by using the abnormality-related data stored in the abnormal state DB 114. (Failure, etc.) can be estimated. As a result, it becomes easier to take actions such as arranging repair parts and repair plans.

900: Customer site 101: Air conditioner 102: Sensor 100: Data center 110: Server 111: Database 120: Learning unit 130: Evaluation unit 140: Data processing capacity value processing unit 113: Capacity value 121 131: Capacity value input unit 122, 132: Feature vector extraction unit 123: 133: Abnormality measure calculation unit
124: Threshold calculation unit 134: Abnormality determination unit 135: Abnormality-related ability value identification unit

Claims (12)

The air-conditioning machine has a database that stores a plurality of arithmetic expressions representing the air-conditioning functional force according to the refrigerant, and a processing device that processes information using sensor signals acquired from sensors that detect the state of the air-conditioning machine. The processing device is
For the sensor signal, a first processing unit that selects a plurality of arithmetic expressions representing the air conditioning functional power according to the air conditioner refrigerant from the database, calculates the capacity value, and regards this capacity value as normal data to learn.
For the sensor signal acquired by the first processing unit in a period different from the learning period of the air conditioner, a plurality of arithmetic expressions representing the air conditioning functional force corresponding to the air conditioner refrigerant are selected, the capacity value is calculated, and the capacity value is calculated. It has a second processing unit that detects abnormalities based on the learned normal ability value data.
An abnormality detection system characterized in that when the second processing unit detects an abnormality, an abnormality-related ability value is specified and information regarding the abnormal state is output. The processing device is
Data that selects a plurality of arithmetic expressions from the database in relation to the sensor signal acquired from the sensor, extracts them under operating conditions in consideration of the operating state of the air conditioner, and calculates an exponential moving averaged capacity value. It has a processing capacity value processing unit and has a processing capacity value processing unit.
The abnormality detection system according to claim 1, wherein the first processing unit or the second processing unit uses the capacity value calculated by the data processing capacity value processing unit. After the second processing unit identifies the abnormality-related capacity value, the pattern of the air-conditioning function capacity value is stored in the database for each abnormal state of the air conditioner.
The abnormality detection system according to claim 1, further comprising an abnormality state estimation unit that estimates an abnormality state of an air conditioner based on the capacity value stored in the database when the second processing unit detects an abnormality. The abnormality detection system according to claim 1, wherein an arithmetic expression relating to a new type of air conditioner can be added to the database. The first processing unit is
A first capacity value input unit that accepts input of the capacity value calculated by the data processing capacity value processing unit, and
A first feature vector extraction unit that extracts a feature vector using the ability value received from the ability value input unit, and
A first anomaly measure calculation unit that calculates an anomaly measure for each feature vector for each predetermined time extracted by the feature vector extraction unit using a feature vector for a predetermined learning period.
It has a threshold value calculation unit that calculates a threshold value according to the anomaly measure, and has a threshold value calculation unit.
The second processing unit is
A second capacity value input unit that accepts input of the capacity value calculated by the data processing capacity value processing unit, and
A second feature vector extraction unit that extracts a feature vector using the ability value received from the ability value input unit, and a second feature vector extraction unit.
A second anomaly measure calculation unit that calculates an anomaly measure for each feature vector at predetermined time extracted by the second feature vector extraction unit using a feature vector of a predetermined learning period, and a second anomaly measure calculation unit.
An abnormality determination unit that determines an abnormality by comparing an abnormality measure calculated by the second abnormality measure calculation unit with a threshold value calculated by the threshold value calculation unit.
The abnormality detection system according to claim 2, further comprising an abnormality-related ability value specifying unit for specifying an abnormality-related ability value. The first processing unit repeats the following processing for the air-conditioning functional force value to which the exponential moving average is applied for the number of the two-dimensional distribution density of each capacity value;
(A) Create a two-dimensional distribution density of normal data by combining multiple ability values,
(B) By round-robin processing for two air-conditioning function power values, the maximum and minimum values for each type of air-conditioning function power value are obtained so that the vertical and horizontal scales of the two-dimensional distribution density are the same, and the normal air-conditioning function is further performed. For the force value, also find the maximum and minimum values in the columns and rows.
The abnormality detection system according to claim 1. The second processing unit repeats the following processing for each abnormal data and the two-dimensional distribution density of each ability value;
(B) Create a two-dimensional distribution density of normal data by combining multiple ability values,
(B) By round-robin processing for each of the two air-conditioning function power values, the maximum and minimum values for each type of air-conditioning function power value are obtained so that the vertical and horizontal scales of the two-dimensional distribution density are the same, and further, positive. For the normal air-conditioning function power value, the maximum value and the minimum value of each column and row are also obtained.
(C) Analyze the abnormal state of each ability value, identify the abnormal state related ability value, and obtain the ability value state.
The abnormality detection system according to claim 6. The database is
An air-conditioning function force value state table that registers the abnormal state determined by the second processing unit for each two-dimensional distribution density for each abnormal data, and
Each capacity value state table that registers the state converted from two types of air conditioning function power values using the two-dimensional distribution density to one type of air conditioning function power value state, and
The abnormality detection system according to claim 7. An abnormality detection method for detecting an abnormality in an air conditioner by using a sensor signal acquired from a sensor that detects the state of the air conditioner by using a processing device that processes information.
For the sensor signal, the capacity value is calculated by selecting a plurality of calculation formulas representing the air conditioning functional power according to the air conditioner refrigerant from the database storing a plurality of calculation formulas representing the air conditioning functional power corresponding to the air conditioner refrigerant. , The first processing step to learn by regarding this ability value as normal data,
For the sensor signal acquired in a period different from the learning period of the air conditioner in the first processing step, a plurality of arithmetic expressions representing the air conditioning functional force corresponding to the air conditioner refrigerant are selected, the capacity value is calculated, and the capacity value is calculated. The second processing step of detecting an abnormality based on the learned normal ability value data, and
When the second processing step detects an abnormality, a step of specifying an abnormality-related ability value and outputting information on the abnormal state, and a step of outputting the information.
Anomaly detection method characterized by executing. The processing device is
Data that selects a plurality of arithmetic expressions from the database in relation to the sensor signal acquired from the sensor, extracts them under operating conditions in consideration of the operating state of the air conditioner, and calculates an exponentially moving averaged capacity value. Has a processing capacity value processing step,
The abnormality detection method according to claim 9, wherein the first processing step or the second processing step uses the capacity value calculated by the data processing capacity value processing step. After the second processing step identifies the abnormality-related capacity value, the pattern of the air conditioning function capacity value is stored in the database for each abnormal state of the air conditioner.
The abnormality detection method according to claim 9, further comprising an abnormality state estimation step of estimating an abnormality state of the air conditioner based on the capacity value stored in the database when the second processing step detects an abnormality. It is an abnormality detection program that detects an abnormality in the air conditioner using a sensor signal acquired from a sensor that detects the state of the air conditioner, which is executed by the processing device.
For the sensor signal, the capacity value is calculated by selecting a plurality of calculation formulas representing the air conditioning functional power according to the air conditioner refrigerant from the database storing a plurality of calculation formulas representing the air conditioning functional power corresponding to the air conditioner refrigerant. , The first processing step to learn by regarding this ability value as normal data,
For the sensor signal acquired in a period different from the learning period of the air conditioner in the first processing step, a plurality of arithmetic expressions representing the air conditioning functional force corresponding to the air conditioner refrigerant are selected, the capacity value is calculated, and the capacity value is calculated. The second processing step of detecting an abnormality based on the learned normal ability value data, and
When the second processing step detects an abnormality, a step of specifying an abnormality-related ability value and outputting information on the abnormal state, and a step of outputting the information.
An abnormality detection program, characterized in that the processing apparatus is executed.

Images