JP2021189964A - Anomaly diagnostic device and anomaly diagnostic method - Google Patents

Abstract

To improve accuracy in diagnosing an anomaly in an apparatus.SOLUTION: An anomaly diagnostic device has: a data type storage unit 124 that stores data type information for determining whether operation data 101 acquired from an apparatus is performance information or operating condition information; a learned data storage unit 118 that stores learned data, which is a result of learning the operation data 101 in advance, for determining whether the operation data 101 is abnormal or normal; and a learned data adding unit 126 by which, for operation data 101 newly acquired from the apparatus, new operation data in which new operation data 101 is suitable as learned data is additionally stored into the learned data storage unit 118 as new learned data on the basis of a comparison between the learned data stored in the learned data storage unit 118 and the new operation data 101 in a case where the operation data 101 determined to be abnormal is the operating condition information.SELECTED DRAWING: Figure 2

Description

The present invention relates to an abnormality diagnosing device for diagnosing an abnormality in an apparatus and a technique for diagnosing an abnormality.

A system for diagnosing an abnormality from operation data is provided for a device that is expected to operate constantly, such as an air conditioner. Such an abnormality diagnosis determines (diagnoses) whether the state of the device is a normal operating state or an abnormal state, and if it is determined that the device is in an abnormal state as a result, the abnormality is notified. be. Here, the abnormal state means a state in which some kind of failure or deterioration has occurred in the device. It is expected that such abnormality diagnosis will reduce the downtime of equipment by early detection of deterioration and failure.

These abnormality diagnoses are performed by detecting the state of the air conditioner with a sensor or the like and using a predetermined diagnostic method based on the detected value of the sensor. Among the various devices to be diagnosed, there are a wide variety of operating conditions depending on the environment in which they are installed, the model of the device, and the conditions in which they are used. In that case, it may be very difficult to determine whether the change in the measured value in a specific sensor item is due to an abnormality in the device itself or a change in operating conditions. That is, even though the device itself has no abnormality, it may be diagnosed as "abnormal" due to a change in operating conditions.

For such a problem, for example, Patent Document 1 is disclosed. Patent Document 1 states that "the abnormality diagnosis system is an abnormality diagnosis system for performing an abnormality diagnosis of an air conditioner, and includes a first out-licensing unit, a second out-licensing unit, and an abnormality diagnosis unit. Derives a derived value based on the state value related to the state of the air conditioner at the time of diagnosis. The second derivation unit derives a normal value based on the state value of the air conditioner during normal operation. An abnormality diagnosis system is disclosed in which the derived value and the normal value are tested by statistical processing, and when the derived value and the normal value are considered to have a significant difference, the air conditioner is diagnosed as being in an abnormal state. " ..

Japanese Unexamined Patent Publication No. 2009-002650

However, in the technique described in Patent Document 1, the classification of the operating state is initially created. Then, in the technique described in Patent Document 1, only the data of the operating state conforming to this classification is targeted for diagnosis. Therefore, depending on the sensor data, it may not be possible to make a diagnosis target because it does not conform to the classification. Further, in the technique described in Patent Document 1, data in which the state is stable is extracted by the stable state extraction unit. Therefore, if the device fails and is not in a stable state, it may not even be possible to extract data.

The present invention has been made in view of such a background, and it is an object of the present invention to improve the accuracy in diagnosing abnormalities of equipment.

In order to solve the above-mentioned problems, in the present invention, the operation data acquired from the device is either performance information which is information derived from the performance of the device itself or operation condition information which is information related to the operation of the device. This is the result of learning the operation data in advance and the data type storage unit that stores the data type information for determining whether the operation data is abnormal or normal. With respect to the learning data storage unit that stores the learning data of the above and the new operation data that is the operation data newly acquired from the device, the new operation data determined to be abnormal based on the learning data is the said. In the case of operating condition information, based on the comparison between the learning data stored in the learning data storage unit and the new operation data, the new operation data appropriate as the training data is obtained as new learning data. It is characterized by having a learning data addition unit that is additionally stored in the training data storage unit.
Other solutions are described in the embodiments.

According to the present invention, it is possible to improve the accuracy in diagnosing an abnormality of a device.

It is a figure which shows the configuration example of the management system in this embodiment. It is a functional block diagram which shows the structural example of the abnormality diagnosis apparatus which concerns on this embodiment. It is a figure which shows the example of the type data stored in the data type storage part. It is a flowchart which shows the processing procedure of the abnormality diagnosis method performed in this embodiment. It is a figure which shows an example of the method of the addition possibility determination. It is a figure which shows the hardware configuration of the abnormality diagnosis apparatus.

Next, an embodiment (referred to as “embodiment”) for carrying out the present invention will be described in detail with reference to the drawings as appropriate.

[Abnormality diagnostic device 1]
FIG. 1 is a diagram showing a configuration example of the management system Z in the present embodiment.
The management system Z has at least one diagnosis target system D, a data terminal 5 installed in the service center C2, and an abnormality diagnosis device 1 installed in the remote monitoring center C1.
Among the diagnosis target systems D, the diagnosis target system Da includes two air conditioners 2 and a control device 3 connected to these air conditioners 2.
The air conditioner 2 is the target of the diagnosis performed by the abnormality diagnosis device 1, and controls the air in the room to an arbitrary state. The air conditioner 2 is a multi air conditioner, an industrial air conditioner for cooling a room containing a computer, or the like.
The air conditioner 2 is provided with a plurality of sensors 201 for detecting an operating state and an environmental state. These sensors 201 detect state detection values such as the outside air temperature, the room temperature, the frequency of the compressor of the air conditioner 2, the temperature and pressure of the refrigerant circulating inside, the suction pipe temperature of the refrigerant, and the like.

In the diagnosis target system Da, the control device 3 is connected to a plurality of (two in the example of FIG. 1) air conditioners 2 provided in the diagnosis target system Da by a communication line. Then, the control device 3 controls a plurality of air conditioners 2 by setting setting values for control for each air conditioner 2. Further, the control device 3 is connected to the abnormality diagnosis device 1 via the communication network N. Then, the control device 3 transmits the operation data 101 (see FIG. 3) of the air conditioner 2 to the abnormality diagnosis device 1 installed in the remote monitoring center C1. The operation data 101 includes a state detection value detected by the sensor 201 of each air conditioner 2, a setting value of control for the air conditioner 2, control contents executed by the air conditioner 2, and an air conditioner. 2 power consumption and the like are included. These operation data 101 are detected every predetermined time, for example, every minute. Then, the operation data 101 is continuously or intermittently collected and accumulated by the control device 3. Then, as described above, the control device 3 transmits the accumulated operation data 101 to the abnormality diagnosis device 1. In the example of the diagnosis target system Da, the operation data 101 of the two air conditioners 2 are distinguished and transmitted to the abnormality diagnosis device 1.

The diagnosis target system Da is provided with two air conditioners 2, but may be one or three or more. Further, in the diagnosis target system Da, two air conditioners 2 are connected to one control device 3, but the present invention is not limited to this. For example, the control device 3 may be connected to each air conditioner 2.

Each of the diagnostic target systems Db and Dc is provided with a device 4 and a control device 3. The device 4 may be an air conditioner 2 or another device 4. The other device 4 is a device such as a refrigerator, an elevator, a plant device, etc., which is expected to operate steadily. Alternatively, the device 4 may be configured by connecting one or more devices 4 such as all air conditioning units in the building. Further, the device 4 may include other equipment such as a hot water supply device and lighting.
Further, the air conditioner 2 of the system to be diagnosed Da may be replaced with the device 4.
In the diagnosis target systems Db and Dc, the device 4 is connected to the control device 3, and the control device 3 transmits the operation data 101 of the device 4 to the abnormality diagnosis device 1 in the same manner as the diagnosis target system Da.

The abnormality diagnosis device 1 is a device for performing abnormality diagnosis of the air conditioner 2 and the device 4 installed in each of the plurality of diagnosis target systems D. Further, the abnormality diagnosis device 1 is arranged in the remote monitoring center C1 away from the air conditioner 2 and the diagnosis target system D in which the device 4 is arranged. However, the abnormality diagnosis device 1 may be arranged in any of the diagnosis target systems D. As a result, the abnormality diagnosis device 1 remotely monitors the diagnosis target system D.
The abnormality diagnosis device 1 determines whether the air conditioner 2 is in a normal operating state or an abnormal state from the air conditioner 2 sent by the control device 3 and the operation data 101 of the device 4. If it is determined that the condition is abnormal, the abnormality diagnosis device 1 notifies the administrator of the diagnosis target system D and the like of the abnormality, and also notifies the data terminal 5 of the service center C2 of the abnormality.

Upon receiving the notification of the abnormality, the data terminal 5 notifies the serviceman P belonging to the service center C2 of the instruction of the maintenance work of the air conditioner 2. Upon receiving the notification, the serviceman P is dispatched to the maintenance work for the system D to be diagnosed in which the abnormality is detected.
The abnormal state here means a state in which the air conditioner 2 or the device 4 has some kind of failure or deterioration.

Further, here, it is assumed that the operation data 101 of the air conditioner 2 and the device 4 is transmitted to the abnormality diagnosis device 1 via the communication network N. That is, in the example of FIG. 1, the control device 3 is connected to the abnormality diagnosis device 1 via the communication network N. However, the control device 3 may be directly connected to the abnormality diagnosis device 1 without going through the communication network N, that is, in the form of peer-to-peer (Peer to Peer) or VPN (Virtual Private Network).

[Abnormality diagnostic device 1]
FIG. 2 is a functional block diagram showing a configuration example of the abnormality diagnosis device 1 according to the present embodiment.
In the following, it is assumed that the device 4 to be diagnosed is an air conditioner 2.
The abnormality diagnosis device 1 has an input / output vector extraction unit 111, a learning data selection unit 112, a regression model creation unit 113, an abnormality degree calculation unit 114, a threshold value calculation unit 115, and an abnormality determination unit 116. Further, the abnormality diagnosis device 1 has a data type identification unit 121, a valid / invalidity determination unit 122, a notification unit 123, and a learning data addition unit 126. Further, the abnormality diagnosis device 1 has an evaluation data storage unit 102, a learning data storage unit 118, a data type storage unit 124, an invalid data storage unit 125, a characteristic creation unit 119, a characteristic data storage unit 119A, and an additional determination unit 127.

The input / output vector extraction unit 111, the learning data selection unit 112, the regression model creation unit 113, the abnormality degree calculation unit 114, the threshold value calculation unit 115, the abnormality determination unit 116, and the learning data storage unit 118 are described in JP2013-25367. Since it is the technique described in the publication, the explanation here is simple. In the following, the input / output vector extraction unit 111, the learning data selection unit 112, the regression model creation unit 113, the abnormality degree calculation unit 114, the threshold value calculation unit 115, the abnormality determination unit 116, the learning data storage unit 118, and the characteristic creation unit 119. The characteristic data storage unit 119A may be referred to as a learning / abnormality determination unit 110.

Then, the data type specifying unit 121, the valid / invalid determination unit 122, the notification unit 123, the learning data addition unit 126, the data type storage unit 124, the characteristic creation unit 119, the characteristic data storage unit 119A, and the additional determination unit 127 are carried out. It is a characteristic part of the form. The data type specifying unit 121, the valid / invalid determination unit 122, the notification unit 123, the learning data addition unit 126, the data type storage unit 124, the invalid data storage unit 125, and the additional learning unit 127 may be referred to as an additional learning unit 120.

The input / output vector extraction unit 111 extracts the input vector and the output vector used for the regression model from the operation data 101 output from the air conditioner 2. The input vector and the output vector are those described in Japanese Patent Application Laid-Open No. 2013-25367, but the input vector is the actual operation data 101 which is the basis of the prediction, and the output vector is the predicted operation data 101. As described in Japanese Patent Application Laid-Open No. 2013-25367, the output vector may also be the actual operation data 101.
The extracted input vector and output vector are stored as learning data in the learning data storage unit 118.
Further, the characteristic creation unit 119 extracts the tendency related to the performance from the operation data 101 output from the air conditioner 2 and creates the characteristic data of the air conditioner 2. Then, the characteristic creation unit 119 stores the created characteristic data in the characteristic data storage unit 119A. This characteristic data becomes data for determining whether or not the added data (invalid data described later) is appropriate as learning data when new learning data is added to the learning data storage unit 118. The characteristic data created by the characteristic creation unit 119 will be described later.

The learning data selection unit 112 selects learning data (here, an input vector) from the learning data stored in the learning data storage unit 118 based on the degree of similarity with the newly extracted input vector. The newly extracted input vector is stored in the evaluation data storage unit 102 as described later.
The regression model creation unit 113 creates a regression model using the selected training data. As the regression model here, a Gaussian process or the like is used as described in Japanese Patent Application Laid-Open No. 2013-25367.

The abnormality degree calculation unit 114 calculates the abnormality degree based on the regression model created by the regression model creation unit 113. Specifically, the abnormality degree calculation unit 114 calculates the abnormality degree by the method described in JP2013-25367A.
The threshold value calculation unit 115 calculates the threshold value described in Japanese Patent Application Laid-Open No. 2013-25367.
The abnormality determination unit 116 describes the operation data 101 according to the method described in Japanese Patent Application Laid-Open No. 2013-25367 based on the abnormality degree calculated by the abnormality degree calculation unit 114 and the threshold value calculated by the threshold value calculation unit 115. Make an abnormality judgment. That is, the abnormality determination unit 116 determines whether or not the newly acquired operation data 101 is abnormal based on the learning data stored in the learning data storage unit 118.

The evaluation data stored in the evaluation data storage unit 102 will be described later.

As described above, the learning / abnormality determination unit 110 in the present embodiment determines the abnormality of the operation data 101 by using the method described in JP2013-25367A. However, this method does not correspond to the device 4 in which the operation data 101 is often changed, such as the air conditioner 2 and the like. That is, in the method described in Japanese Patent Application Laid-Open No. 2013-25367, although there is no abnormality in the air conditioner 2 itself, there is a possibility that it is determined to be abnormal simply because the operation data 101 is changed. Therefore, further improvement is needed. For such improvements, the present embodiment includes a characteristic creation unit 119, a characteristic data storage unit 119A, and an additional learning unit 120 in the learning / abnormality determination unit 110. Each configuration of the additional learning unit 120 will be described later.

The type data is stored in the data type storage unit 124. Here, the type data will be described.
FIG. 3 is a diagram showing an example of type data stored in the data type storage unit 124.
As shown in FIG. 3, the type data has each field of "No", "data name", "data collection method", and "type".
"No" is a serial number related to each data type.
The "data name" is a name of data such as "outside air temperature" and "suction air temperature".
The "data collection method" indicates how the data was acquired. In the example of FIG. 3, there are "sensor data" and "physical quantity" as "data collection method". “Sensor data” is raw data acquired from sensor 201. The "physical quantity" is calculated by a predetermined mathematical formula, a map, or the like based on the "sensor data".

The "type" stores information on whether the data is data related to "operating conditions" or data related to "performance". The "operating conditions" are data that change according to the operating state of the air conditioner 2 and the environment. "Performance" is data derived from the air conditioner 2 itself.
Information regarding the type of the operation data 101 is created in advance by an administrator or the like and stored in the data type storage unit 124.

As described above, in the present embodiment, the operation data 101 is provided with type data in which data indicating "operating conditions" or data indicating "performance" is classified in advance. The classification as shown in FIG. 3 is a value that is almost uniquely determined by the device 4 (here, the air conditioner 2) to be diagnosed. Therefore, there is no difficulty in setting as long as there is knowledge of general equipment 4.

Returning to the description of FIG.
The data type specifying unit 121 specifies the type of the operation data 101 whose abnormality is determined by the abnormality determination unit 116.
The valid / invalid determination unit 122 determines whether the abnormality determined by the abnormality determination unit 116 is valid or invalid based on the type specified by the data type identification unit 121. The fact that the anomaly is valid or invalid will be described later.
The notification unit 123 notifies the user that an abnormality has occurred when the validity / invalidity determination unit 122 determines that the abnormality is valid.
The invalid data storage unit 125 stores the operation data 101 whose abnormality is determined to be "invalid" by the valid / invalid determination unit 122.
In the additional determination unit 127, the operation is created by the characteristic creation unit 119 and stored in the invalid data storage unit 125 based on the characteristic data of the air conditioner 2 and the device 4 stored in the characteristic data storage unit 119A. It is determined whether or not the data 101 can be added to the training data.
The learning data addition unit 126 adds, among the operation data 101 stored in the invalid data storage unit 125, the data whose determination result in the addition determination unit 127 is “addable” to the learning data storage unit 118 as learning data. .. Further, at this time, the notification unit 123 notifies the user that the learning data is to be added to the learning data storage unit 118.

The operation data 101 includes a state detection value detected by the sensor 201 of each air conditioner 2, a setting value of control for the air conditioner 2, control contents executed by the air conditioner 2, and the air conditioner 2. Power consumption etc. are included. Further, as shown in FIG. 3, the operation data 101 includes not only the sensor data of the value itself detected by the sensor 201 but also a physical quantity calculated based on the value detected by the sensor 201.

The abnormality diagnosis device 1 collects training data for a certain period of time in a state where there is no abnormality in the device 4 in order to store the input vector and the output vector to be the learning data in the learning data storage unit 118. The state in which there is no abnormality in the device 4 is a period in which there is no abnormality in the device 4, for example, a certain period after being newly installed. Further, the fixed period here is preferably a period in which the operating conditions assumed by the device 4 to be diagnosed (air conditioner 2 in the example of the present embodiment) appear, but it is not necessary to cover all the operating conditions. The period in which the operating conditions assumed in the device 4 to be diagnosed appears is, for example, the period from the summer setting to the switching to the winter setting when the device 4 to be diagnosed is the air conditioner 2. Then, the input / output vector extraction unit 111 of the abnormality diagnosis device 1 extracts from the input / output vector extraction unit 111. The extracted operation data 101 is stored as learning data in the learning database 6.

Further, after the above-mentioned fixed period has passed, the abnormality diagnosis device 1 stores the acquired operation data 101 in the evaluation data storage unit 102. The operation data 101 stored in the evaluation data storage unit 102 is referred to as evaluation data. Then, when a predetermined amount of evaluation data is accumulated in the evaluation data storage unit 102, the input / output vector extraction unit 111 extracts an input vector and an output vector from the operation data 101 stored in the evaluation data storage unit 102. Then, the learning data selection unit 112 selects learning data from the learning data storage unit 118 based on the degree of similarity between the learning data stored in the learning data storage unit 118 and the learning data extracted from the evaluation data. .. The learning data here is an output vector.

[flowchart]
FIG. 4 is a flowchart showing a processing procedure of the abnormality diagnosis method performed in the present embodiment.
Note that FIG. 4 shows the process after the processing of the abnormality determination unit 116, which is a characteristic portion of the present embodiment.
First, the abnormality determination process is performed by the abnormality determination unit 116 (S1). The abnormality determination unit 116 determines whether or not the newly acquired operation data 101 is abnormal based on the learning data stored in the learning data storage unit 118. Actually, the regression model creation unit 113 creates a regression model based on the training data selected by the training data selection unit 112. Then, the abnormality degree calculation unit 114 calculates the abnormality degree based on the regression model created by the regression model creation unit 113. Then, the abnormality determination unit 116 determines whether or not the newly acquired operation data 101 is abnormal based on the abnormality degree calculated by the abnormality degree calculation unit 114 and the threshold value calculated by the threshold value calculation unit 115. Is determined.

If it is not determined to be abnormal as a result of the abnormality determination process in step S1, that is, if it is determined to be normal (S1 → normal), the abnormality diagnosis device 1 returns the process to step S1.
When an abnormality is determined as a result of the abnormality determination in step S1 (S1 → abnormality), the data type specifying unit 121 is operated data determined to be abnormal based on the type data stored in the data type storage unit 124. The type of 101 is determined (S2). The type of the operation data 101 is the "type"("operationcondition","performance") shown in the example of FIG. Information related to the "data name" and the like is attached as attribute information to the operation data 101 acquired from the control device 3.
The fact that the operation data 101 is determined to be abnormal also means that the target operation data 101 is similar to the operation data 101 that has a great influence on the calculation of the degree of abnormality.

Then, the valid / invalid determination unit 122 determines whether the operation data 101 determined to be abnormal is valid or invalid based on the result of step S2 (S3). Specifically, when the type of the operation data 101 determined to be abnormal is "performance", the valid / invalid determination unit 122 determines that the abnormality determination is "valid". Further, when the type of the operation data 101 determined to be abnormal is "operation condition", the valid / invalid determination unit 122 determines that the abnormality determination is "invalid". This is because when the type is "operation condition", it may be determined as abnormal because it was not accumulated as learning data based on the corresponding operation data 101. Further, when the type of the operation data 101 is "performance", it is determined to be "effective" because the abnormality may have been determined due to the device 4 (air conditioner 2) itself.

If it is determined to be "valid" as a result of step S3 (S3 → "valid"), the notification unit 123 notifies that an abnormality has occurred (S4), and the abnormality diagnosis device 1 returns the process to step S1. For example, the notification informs the display unit (not shown) of the data terminal 5 provided in the service center C2 that an abnormality has occurred in the device 4 (air conditioner 2) to the device 4 (air conditioner 2). It is displayed together with the ID of 2). Upon receiving the notification, the serviceman P of the service center C2 is dispatched to the maintenance work of the device 4 (air conditioner 2) in which the abnormality has occurred.

When it is determined as "invalid" as a result of step S3 (S3 → "invalid"), the valid / invalid determination unit 122 stores the evaluation data (= operation data 101) determined as "invalid" in the invalid data storage unit 125. It is stored in (S11). The evaluation data (= operation data 101) stored in the invalid data storage unit 125 is appropriately referred to as invalid data. Further, the input vector and the output vector extracted from the operation data 101 may be stored in the invalid data storage unit 125. The process of step S11 can be omitted. When the process of step S11 is omitted, the invalid data storage unit 125 can also be omitted.
Then, the additional determination unit 127 determines whether or not the amount of invalid data has reached a certain amount, for example, at predetermined time intervals (S12). The predetermined time is one week, one month, or the like. In addition, step S12 may determine whether a certain time (for example, one week or one month) has passed since the addition of the previous learning data.
As a result of step S12, when the amount of invalid data has not reached a certain amount (S12 → No), the abnormality diagnosis device 1 returns the process to step S1.

As a result of step S12, when the amount of invalid data has reached a certain amount (S12 → Yes), the addition determination unit determines whether or not addition is possible (S13). In the addition availability determination, the addition determination unit 127 compares the characteristic data stored in the characteristic data storage unit 119A with the invalid data. Then, the additional determination unit 127 determines whether or not each invalid data can be added to the learning data storage unit 118 based on the result of the comparison. That is, the additional determination unit 127 determines whether the invalid data satisfies the predetermined characteristics of the learning data stored in the air conditioner 2 and the learning data storage unit 118 of the device 4. Of the invalid data accumulated in a certain amount, those that satisfy the characteristics can be added to the learning data storage unit 118, and if they do not satisfy the characteristics, they cannot be added to the learning data storage unit 118.

(Determining whether or not to add)
Here, with reference to FIG. 5, an example of the method for determining whether or not addition is possible in step S13 of FIG. 4 will be described.
FIG. 5 is a diagram showing an example of a method for determining whether or not addition is possible in step S13 of FIG.
The characteristic creation unit 119 extracts two or more predetermined learning data from the training data, and creates characteristic data from the correlation. This characteristic data is created by learning data related to operating conditions.
In FIG. 5, as an example of the characteristic data to be created, the learning data of the operating condition equivalent index and the output equivalent index are extracted and their correlations are displayed in a two-dimensional graph. This two-dimensional graph is called a characteristic graph. The operating condition equivalent index is, for example, the set temperature of the air conditioner 2, and the output equivalent index is, for example, the room temperature cooled by the air conditioner 2. The user decides what to select for the coordinate axes of the characteristic graph.

Further, in FIG. 5, the black circle M1 indicates the learning data stored in the learning data storage unit 118. The x mark M2 is invalid data stored in the invalid data storage unit 125, and is data scheduled to be added to the learning data storage unit 118 as learning data from now on according to the determination of the additional determination unit 127.
Generally, the operation data 101 in the actual device 4 (air conditioner 2) tends to be dispersed with a certain width from the approximate line obtained from the least squares method or the like due to an error or a difference in operating conditions other than the extracted sensor data. have. Here, in consideration of this range, the tendency of the upper limit of the output equivalent index with respect to the increase of the operating condition equivalent index is characterized.

This will be specifically described below.
Here, the line L1 indicates an approximate line obtained by the least squares method with respect to the invalid data indicated by the black circle M1. Further, the line L21 shows a straight line having a value of + σ with respect to the line L1, and the line L22 shows a straight line having a value of −σ with respect to the line L1. Where σ is the standard deviation. Here, the lines L21 and L22 are straight lines having a value of ± σ with respect to the approximate line (line L1) obtained by the least squares method, but the present invention is not limited to this. For example, the lines L21 and L22 may be determined by setting ± 2σ with respect to the line L1, ± (σ + C) (C is a predetermined constant), or based on a value obtained from the experience so far. ..

The two coordinate axes in FIG. 5, the invalid data indicated by the black circle M1, and the lines L1, L21, and L22 are the characteristic data.

The additional determination unit 127 extracts invalid data of the operating condition equivalent index and the output equivalent index from the invalid data accumulated in a certain amount (x mark M2), and the correlation of the invalid data satisfies the characteristics of the existing learning data. If so, it is determined that "addition is possible". Further, if the correlation of the invalid data does not satisfy the characteristics of the existing learning data, the addition determination unit 202 determines that "addition is not possible". Then, the learning data addition unit 126 additionally stores the invalid data determined to be “addable” in the learning data storage unit 118 as new learning data.

Specifically, the area above the line 21 in FIG. 5 (the area indicated by the dots) is set as the non-addable area A1. Further, the area below the line 21 in FIG. 5 (the area other than the area indicated by the dots) is set as the addable area A2. Then, among the invalid data indicated by the x mark M2, the invalid data (x mark M21) belonging to the non-addable area A1 deviates from the characteristics of the existing learning data, so that the additional determination unit 202 is unsuitable as the learning data. judge. That is, the addition determination unit 202 determines that the invalid data indicated by the x mark M21 is “cannot be added”. This is because it is considered that the invalid data (x mark M21) belonging to the non-addable area A1 has an abnormal value due to deterioration or the like.

On the other hand, among the invalid data indicated by the x mark M2, the invalid data (x mark M22) belonging to the non-addable area A1 (addable area A2) exists within the range of the characteristics of the existing learning data. Judged as appropriate as training data. That is, the additional determination unit 127 determines that the invalid data indicated by the x mark M22 can be added to the learning data storage unit 118.

In this embodiment, the region above the line L1 is defined as the non-addable region A1 in FIG. 5, but the region is not limited to this. In FIG. 5, the region below the line L2 may be designated as the non-addable region A1, or both the region above the line L1 and the region below the line L2 may be designated as the non-addable region A1.

Further, in the present embodiment, the non-addable area A1 and the addable area A2 are based on the standard deviation of the learning data (black circle M1) stored in the learning data storage unit 118 with respect to the approximate line (line L1) obtained by the least squares method. Is set. However, the present invention is not limited to this, and for example, an addable area A2 may be set by cluster analysis, and an area other than the addable area A2 may be set as a non-addable area A1.

The determination performed in step S13 may be performed based on the equation and the coordinates, or may be determined based on the image. Judgment based on the image means that the invalid data indicated by x mark M21 is determined in the non-addable area A1 based on the coordinates, the plot status, the non-addable area A1 and the image of the addable area A2 as shown in FIG. The additional determination unit 127 determines whether it is inside or inside the addable area A2. That is, the format of the determination does not matter as long as it is easy to compare with the invalid data indicated by the mark M21 by the additional determination unit 127. Further, when it is desired to make a more detailed determination, two or more characteristic data may be used. That is, the addition determination unit 127 can perform the addition possibility determination a plurality of times based on the plurality of characteristic data. In this case, it is possible to add invalid data determined to be "addable" to the learning data storage unit 118 by all the addition possibility determinations, and invalid data determined to be "addable" a predetermined number of times is learned data. It can also be added to the storage unit 118.

Return to the description of FIG.
As a result of step S13, the learning data addition unit 126 adds invalid data determined to be addable as learning data of the learning data storage unit 118 (S14). The invalid data determined to be addable is the data indicated by the x mark M22 plotted in the addable area A2 in FIG.

Then, the notification unit 123 notifies the user that the learning data is to be added to the learning data storage unit 118 (S21).
Next, the characteristic creation unit 119 determines whether or not a predetermined time has elapsed since the characteristic data was previously created (S22).
As a result of step S22, if the predetermined time has not elapsed (S22 → No), the abnormality diagnosis device 1 returns the process to step S1.
As a result of step S22, when a predetermined time has elapsed (S22 → Yes), the characteristic creation unit 119 creates characteristic data based on the learning data stored in the learning data storage unit 118 (S23). The characteristic creation unit 119 stores the created characteristic data in the characteristic data storage unit 119A. After that, the abnormality diagnosis device 1 returns the process to step S1.
The process of step S22 can be omitted.

[Hardware configuration]
FIG. 6 is a diagram showing a hardware configuration of the abnormality diagnosis device 1.
The abnormality diagnosis device 1 includes a memory 301, a CPU (Central Processing Unit) 302, a storage device 303 such as an HD (Hard Disk), and a communication device 304.
The communication device 312 communicates with the air conditioner 2, the device 4, and the evaluation data storage unit 102.
Further, the program stored in the storage device 303 is loaded into the memory 301 and executed by the CPU 302. As a result, the learning / abnormality determination unit 110 and the additional learning unit 120 shown in FIG. 2, and the units 111 to 119 and 121-127 constituting the learning / abnormality determination unit 110 and the additional learning unit 120 are embodied.
The storage device 303 includes a learning data storage unit 118, a data type storage unit 124, and an invalid data storage unit 125 in FIG. 2.

In the present embodiment, in the process (S13) by the learning data addition unit 126, the operation data 101 is stored in accordance with the already accumulated learning data, but the present invention is not limited to this. For example, when the added learning data changes significantly with respect to the already stored learning data, the learning data addition unit 126 erases the already accumulated learning data and then applies the additional learning data. It may be stored in the learning database 6. Alternatively, the learning data storage unit 118 may be updated by the method described in JP2013-25367A.

When an abnormality is determined, the abnormality diagnosis device 1 of the present embodiment stores the operation data 101 in the invalid data storage unit 125 when the type of the operation data 101 is an operation condition. As described above, this indicates that there is a possibility that an abnormality has been erroneously determined because the operating conditions are different from the conventional ones. In other words, it can be said that the abnormality diagnosis device 1 detects changes in operating conditions. That is, the abnormality diagnosis device 1 detects that the operating conditions of the air conditioner 2 to be diagnosed have changed, and stores the operating data 101 for adding learning data when the operating conditions change. As a result, the operation data 101 required for additional learning can be accumulated.

Further, as described above, the abnormality diagnosis device 1 of the present embodiment can add learning data that serves as a reference for abnormality determination as the operating conditions change. That is, the abnormality diagnosis device 1 collects learning data for a certain period of time before starting the diagnosis operation, and then starts the actual diagnosis operation.
As a result, it is not always necessary to have learning data for all the operating conditions assumed in the device 4 to be diagnosed at the start of the diagnosis operation. That is, since training data may be added later, it is not necessary to collect training data first over all operating conditions (for example, over spring, summer, autumn, and winter). That is, the learning data for determining the abnormality of the operation data 101 can be created under some operating conditions, for example, only during the summer of the four seasons, and then the diagnosis can be started. Then, when the evaluation data (operation data 101) deviates from the operation conditions, learning data is added. Therefore, the abnormality diagnosis device 1 of the present embodiment can accelerate the time when the abnormality diagnosis can be started. That is, it is possible to shorten the period during which abnormality diagnosis cannot be performed. Further, according to the present embodiment, the amount of learning data at the start of diagnostic operation can be reduced.

Further, as described above, the abnormality diagnosis device 1 of the present embodiment is configured to be added after it is determined that the invalid data to be additionally learned matches the characteristics of the learning data. As a result, it is possible to suppress the mixing of inappropriate data due to deterioration or the like into the learning data and add the learning data, so that it is possible to accurately diagnose an abnormality.

Then, in the present embodiment, after a certain amount of invalid data is accumulated in the invalid data storage unit 125, it is determined whether or not the invalid data is added as learning data by the additional determination unit 127 (step S13 in FIG. 4). ) Is performed. As a result, it is not possible to constantly determine whether or not to add invalid data as learning data, so that the processing load can be reduced.

Further, in the present embodiment, as the characteristic data used for determining whether or not invalid data is added as training data, a part of the training data stored in the training data storage unit 118 is used for the characteristic data. Will be. By doing so, it is possible to reduce the processing load of the addition availability determination in step S13 of FIG.

Then, in the present embodiment, as shown in step S23 of FIG. 4, characteristic data is newly created using the learning data stored in the learning data storage unit 118 to which invalid data is newly added. .. As a result, the accuracy of the addition availability determination in step S13 of FIG. 4 can be improved.

Further, in the present embodiment, when the learning data is added to the learning data storage unit 118, the user is notified to that effect. As a result, the user can recognize that the learning data stored in the learning data storage unit 118 has changed.

In this embodiment, learning data is added as needed (when the operating conditions change). However, not limited to this, the type of the operation data 101 is almost uniquely determined by the method of the device 4. Therefore, it is also possible to apply the learning data from the device 4 of the same type that has already started the diagnosis to the device 4 that newly starts the diagnosis. As a result, the time when the abnormality diagnosis can be started can be further accelerated.
Further, according to the present embodiment, since the learning data (input vector / output vector) can be added, the accuracy of abnormality determination can be improved. That is, the accuracy of the predictive diagnosis can be improved.

As described above, the abnormality diagnosis device 1 of the present embodiment adds learning data that serves as a reference for abnormality diagnosis based on changes in operating conditions. Therefore, at the start of the diagnostic operation, even an unknown abnormal state can be diagnosed as an abnormal state as time passes. As a result, even when the user has little knowledge about the device 4 to be diagnosed, the abnormality diagnosis can be performed accurately and easily.

Further, in the abnormality diagnosis device 1 of the present embodiment, the types of the operation data 101 are classified in advance, and the notification is given only when the type of the acquired operation data 101 is performance. This makes it possible to start maintenance work quickly.

Further, according to the present embodiment, it is possible to prevent erroneous detection of an abnormality due to a change in the operating conditions of the device 4. That is, in the operation data 101 whose type is the operation condition, it is possible to prevent the device 4 itself from being determined to be abnormal even though the device 4 itself does not originally have an abnormality due to the change in the operation condition. By doing so, it is possible to make a more detailed determination as compared with the case where the abnormality is determined by simply setting a threshold value for the operation data 101. As a result, the accuracy of the maintenance work can be improved and the efficiency of the maintenance work can be improved.

Further, in the present embodiment, after a certain amount of the operation data 101 determined to be invalid is stored in the invalid data storage unit 125, it is determined whether or not the learning data can be added. By doing so, it is not always necessary to perform additional processing of training data, and efficient processing can be performed.

Further, in the present embodiment, the configuration of the learning / abnormality determination unit 110 has the same configuration as that described in JP2013-25367A. However, as long as the operation data 101 is learned and the abnormality determination of the operation data 101 is performed based on this learning, the configuration may not be the same as that described in JP2013-25367A.

Further, in this embodiment, the method described in JP2013-25367A is used as the learning method, but other machine learning such as clustering may be applied.

The present invention is not limited to the above-described embodiment, and includes various modifications. For example, the above-described embodiment has been described in detail in order to explain the present invention in an easy-to-understand manner, and is not necessarily limited to those having all the described configurations.

Further, each of the above-mentioned configurations and functions, each unit 110-116, 121-123, 126, each storage unit 102, 118, 124-125, etc. may be partially or all of them designed by, for example, an integrated circuit. It may be realized by hardware. Further, each configuration, function, etc. described above may be realized by software by interpreting and executing a program in which a processor such as a CPU (not shown) realizes each function. In addition to storing information such as programs, tables, and files that realize each function in HD, memory, recording devices such as SSD (Solid State Drive), IC (Integrated Circuit) cards, and SD (Secure) It can be stored in a recording medium such as a Digital) card or DVD (Digital Versatile Disc).

Further, in the present embodiment, the control lines and information lines are shown as necessary for explanation, and not all the control lines and information lines are necessarily shown in the product. In practice, you can think of almost all configurations as interconnected.

1 Abnormality diagnosis device 2 Air conditioner 3 Control device (equipment)
4 Equipment 101 Operation data (new operation data)
116 Abnormality determination unit 118 Learning data storage unit (learning data)
119 Characteristic creation unit (addable area creation unit)
121 Data type identification unit 122 Valid / invalid judgment unit 123 Notification unit 124 Data type storage unit 125 Invalid data storage unit (data storage unit)
126 Learning data addition unit 127 Addition judgment unit A2 Addable area S1 Abnormality judgment processing (abnormality judgment step)
S2 type judgment (type judgment step)
S14 Add invalid data that can be added (additional storage step)

Claims (9)

Data type information for determining whether the operation data acquired from the device is either performance information, which is information derived from the performance of the device itself, or operating condition information, which is information related to the operation of the device. The stored data type storage unit and
A learning data storage unit that is a result of learning the operation data in advance and stores learning data for determining whether the operation data is abnormal or normal.
Regarding the new operation data which is the operation data newly acquired from the device, when the new operation data determined to be abnormal based on the learning data is the operation condition information, it is stored in the learning data storage unit. A learning data addition unit that additionally stores the new operation data appropriate as the training data in the training data storage unit as the new learning data based on the comparison between the learning data and the new operation data. When,
An abnormality diagnostic device characterized by having. The additional determination unit that determines that the new operation data contained in the addable area set based on the error range of the learning data stored in the learning data storage unit is appropriate as the learning data. The abnormality diagnosis device according to claim 1, wherein the abnormality diagnostic apparatus is characterized by having. It is characterized by having an addable area creating unit that creates an addable area based on the learning data stored in the learning data storage unit in which the new operation data is additionally stored as new learning data. The abnormality diagnostic device according to claim 2. It has a data storage unit for storing the new operation data determined to be the operation condition information.
The learning data addition unit
After the new operation data determined to be the operation condition information is accumulated in the data storage unit in a predetermined amount, the comparison and the processing related to the addition of the learning data are performed. The abnormality diagnostic apparatus according to claim 1. The abnormality diagnosis device according to claim 1, further comprising a notification unit that outputs notification to the outside when the new operation data is not determined to be the operation condition information. The abnormality diagnosis device according to claim 1, wherein the learning data used in the comparison is a part of the learning data stored in the learning data storage unit. The abnormality diagnosis device according to claim 1, further comprising a notification unit that notifies the administrator of the device that the learning data will be added when the learning data is added. The abnormality diagnosis device according to claim 1, wherein the device is an air conditioner. Data type information for determining whether the operation data acquired from the device is either performance information, which is information derived from the performance of the device itself, or operating condition information, which is information related to the operation of the device. The stored data type storage unit and
A learning data storage unit that is a result of learning the operation data in advance and stores learning data for determining whether the operation data is abnormal or normal.
Abnormality diagnostic device with
An abnormality determination step for determining whether or not the new operation data, which is the operation data newly acquired from the device, is abnormal based on the learning data stored in the learning data storage unit, and
Based on the data type information stored in the data type storage unit, whether the new operation data determined to be abnormal as a result of the abnormality determination step is the performance information or the operation condition information. Judgment type Judgment step and
As a result of the type determination step, when the new operation data is the operation condition information, as the learning data based on the comparison between the learning data stored in the learning data storage unit and the new operation data. An additional storage step for additionally storing appropriate new operation data as new learning data in the training data storage unit, and
An abnormality diagnosis method characterized by executing.

Images