Abnormality detection device, abnormality detection method, and abnormality detection program

Abstract

An abnormality detection device (100) detects an abnormality of a device. A model generation unit (110) extracts a feature amount of the control signal time-series data (143) as training data, and generates, on the basis of the training data, a control signal model (144) that outputs the degree of abnormality of each control signal in accordance with the time-series data of the control signals. The reliability calculation unit (120) calculates the reliability of the control signal in a specific time range by inputting, as verification data, time-series data of the control signal in the specific time range of an abnormality detection result detected in the sensor signal to the control signal model (144). The abnormality diagnosis unit (130) outputs, as an abnormality diagnosis result (146), information in which the reliability of the control signal in the specific time range is added to the abnormality detection result in the specific time range.

Description

Technical Field

[0001] The present disclosure relates to an abnormality detection device, an abnormality detection method, and an abnormality detection program. Background Art

[0002] In equipment such as power generators, downtime due to failures and other reasons will incur huge costs. Therefore, predictive maintenance is being promoted by introducing abnormality detection devices. Here, in abnormality detection for equipment such as power generators, there is very little data when abnormalities occur, so techniques such as outlier detection are used.

[0003] In outlier detection, a model consisting of a calculation formula for the degree of abnormality and a threshold value for determining abnormality is generated based on training data, which is data that contains no abnormalities or almost no abnormalities. Then, the degree of abnormality is calculated by applying validation data, which is data that is unclear whether it is abnormal or normal, to the model, and when the degree of abnormality is above the threshold, it is determined to be abnormal.

[0004] Equipment such as generator sets have multiple modes. For example, in the case of a generator set, multiple modes refer to control states such as starting, stopping, constant power generation, and switching power generation from 100% to 80%, which have nothing to do with normal or abnormal.

[0005] In a device with multiple modes, the amount of data in each mode is uneven, and sometimes the amount of data in some modes is very small. As a result, when the data in a mode is small, sometimes the verification data of a normal device is mistakenly detected as abnormal due to the nature of outlier detection.

[0006] In the technology described in Patent Document 1, based on data of a group of events such as start / stop or alarm output by a device and their occurrence time, training data and verification data are divided in the time series direction for each mode. In addition, in the technology described in Patent Document 1, a model is generated using the divided data, and the threshold is adjusted using the sufficiency of the divided data.

[0007] Prior art literature

[0008] Patent Literature

[0009] Patent Document 1: Japanese Patent Application Publication No. 2015-172945 Summary of the invention

[0010] Problem that the invention aims to solve

[0011] In the following description, data related to sensors installed on the device will be recorded as sensor signals, data related to indication values ​​used to control the device will be recorded as control signals, and data related to groups of events such as start / stop or alarm occurrence output by the device and their occurrence times will be recorded as event signals.

[0012] When an abnormality occurs in a device such as a generator set and the device is inspected, the device needs to be stopped. When the device is stopped, the cost associated with the stop or the opportunity loss caused by the stop of the device is incurred. Therefore, it is necessary for a person to carefully check the results of the abnormality detection device and make the final decision to stop the device.

[0013] In the method described in Patent Document 1, only the sufficiency of the amount of data in the training data of each pattern is used when generating the model. Therefore, when the abnormality detection result of the abnormality detection device is abnormal, it is impossible to explain whether it is judged as abnormal when the training data of the same pattern as the pattern of the verification data is sufficient or when the training data is very small. Therefore, the person who makes the stop judgment needs to perform data analysis to judge the above content, which has the problem of high judgment cost.

[0014] In the present disclosure, the same outlier detection method as that used in abnormality detection of sensor signals is applied to time series data of control signals representing patterns to generate a control signal model for calculating the degree of abnormality of the control signal. Thus, the purpose of the present disclosure is to calculate the reliability of an abnormality detection result determined based on time series data of sensor signals according to the degree of abnormality of the control signal.

[0015] Means used to solve problems

[0016] The abnormality detection device disclosed in the present invention is provided with a sensor for detecting abnormalities in a device controlled by a control signal, wherein the abnormality detection device is provided with: a reliability calculation unit, which inputs the time series data of the control signal within a specific time range of the abnormality detection result detected by the sensor signal obtained from the sensor into the control signal model as verification data, thereby obtaining the abnormality degree of the control signal within the specific time range, and calculating the reliability of the control signal within the specific time range based on the abnormality degree, the control signal model is a model that outputs the abnormality degree of each control signal according to the time series data of the control signal based on training data, and the training data is a feature value of the time series data of the control signal in the past; and an abnormality diagnosis unit, which outputs the information obtained by adding the reliability of the control signal within the specific time range to the abnormality detection result within the specific time range as an abnormality diagnosis result.

[0017] Effects of the Invention

[0018] In the abnormality detection device of the present invention, a control signal model for calculating the degree of abnormality of the control signal is used for the time series data of the control signal representing the pattern. The control signal model is a model generated based on training data of feature quantities of the time series data of the control signal in the past, and outputs the degree of abnormality of each control signal according to the time series data of the control signal. Therefore, according to the abnormality detection device of the present invention, the reliability of the abnormality detection result determined based on the time series data of the sensor signal can be calculated according to the degree of abnormality of the control signal. Thus, according to the abnormality detection device of the present invention, it is possible to immediately determine whether the time series data of the sensor signal diagnosed by the abnormality detection device is determined to be abnormal in a mode with sufficient data volume, or in a mode with very little data volume. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a diagram showing a hardware configuration example of the abnormality detection device according to the first embodiment.

[0020] Figure 2 This is a diagram showing a functional configuration example of the abnormality detection device according to the first embodiment.

[0021] Figure 3 This is a diagram showing a structural example of the correspondence relationship information according to the first embodiment.

[0022] Figure 4 This is a diagram showing a configuration example of abnormality detection technique information according to the first embodiment.

[0023] Figure 5 This is a diagram showing a structural example of control signal time-series data according to the first embodiment.

[0024] Figure 6 This is a diagram showing a structural example of abnormality detection result information according to the first embodiment.

[0025] Figure 7 This is a flowchart showing the operation of the model generation unit according to the first embodiment.

[0026] Figure 8 This is a diagram showing the first embodiment. Figure 3 The corresponding relationship information and Figure 4 A diagram showing the state in which the abnormality detection technique information is combined.

[0027] Fig. 9 This is a flowchart showing the operation of the reliability calculation unit according to this embodiment.

[0028] Fig.10 This is a diagram showing an example of a method of obtaining an abnormality diagnosis result by the abnormality diagnosing unit according to the first embodiment.

[0029] Fig.11 This is a flowchart showing the operation of the abnormality diagnosis unit according to the first embodiment.

[0030] Fig.12 This is a flowchart showing a learning process for generating a control signal model in the abnormality detection process according to the first embodiment.

[0031] Fig.13 This is a flowchart showing an estimation process of obtaining the degree of abnormality from a control signal model in the abnormality detection process of the first embodiment.

[0032] Fig.14 This is a diagram showing a configuration example of an abnormality detection device according to a second modification of the first embodiment.

[0033] Fig.15 This is a diagram showing a structural example of correspondence information according to the second embodiment.

[0034] Fig.16 This is a diagram showing a configuration example of abnormality detection technique information according to the second embodiment.

[0035] Fig.17 This is a diagram showing an example of the result of combining the correspondence information and the abnormality detection technique information according to the second embodiment.

[0036] Fig.18 This is a diagram showing an example of a method of obtaining an abnormality diagnosis result by the abnormality diagnosing unit according to the second embodiment.

[0037] Fig.19 This is a flowchart showing the operation of the abnormality diagnosis unit according to the second embodiment.

[0038] Fig. 20 This is a diagram showing a functional configuration example of an abnormality detection device according to the third embodiment.

[0039] Fig.21 This is a diagram showing a structural example of event signal information according to the third embodiment.

[0040] Fig. 22 This shows that the event signal information of implementation mode 3 is set to Fig.21 FIG. 1 is an example of pattern segmentation in the case of .

[0041] Fig.23 This is a flowchart showing the operation of the mode division unit according to the third embodiment.

[0042] Fig.24 This is a flowchart showing the process during learning for generating a control signal model according to the third embodiment.

[0043] Fig.25This is a flowchart showing the process of obtaining an estimate of the degree of abnormality based on the control signal model according to the third embodiment. DETAILED DESCRIPTION

[0044] The present embodiment is described below using the figures. In each figure, the same or corresponding parts are marked with the same reference numerals. In the description of the embodiment, the same or corresponding parts are appropriately omitted or simplified. The arrows in the figures mainly show the flow of data or the process.

[0045] Implementation method 1.

[0046] ***Structure description***

[0047] Figure 1 It is a diagram showing a hardware configuration example of the abnormality detection device 100 according to the present embodiment.

[0048] Figure 2 1 is a diagram showing a functional configuration example of the abnormality detection device 100 according to the present embodiment.

[0049] The abnormality detection device 100 detects abnormality in a device. The device is equipped with a sensor and is controlled by a control signal. The abnormality detection device 100 detects abnormality in the device using a sensor signal obtained from the sensor, and determines the reliability of the detection result when the abnormality is detected.

[0050] The abnormality detection device 100 is a computer and includes a processor 910 and other hardware such as a memory 921, an auxiliary storage device 922, an input interface 930, an output interface 940, and a communication device 950. The processor 910 is connected to other hardware via a signal line and controls the other hardware.

[0051] The abnormality detection device 100 includes a model generation unit 110 , a reliability calculation unit 120 , an abnormality diagnosis unit 130 , and a storage unit 140 as functional elements.

[0052] The storage unit 140 stores correspondence information 141 , abnormality detection method information 142 , control signal time series data 143 , a control signal model 144 , abnormality detection result information 145 , and abnormality diagnosis result 146 .

[0053] The abnormality detection device 100 is a device that calculates the reliability of an abnormality detection result related to a device based on data stored in the storage unit 140 .

[0054] The functions of the model generation unit 110, the reliability calculation unit 120, and the abnormality diagnosis unit 130 are implemented by software. The storage unit 140 is provided in the memory 921. In addition, the storage unit 140 may be provided in the auxiliary storage device 922, or may be provided in the memory 921 and the auxiliary storage device 922 in a distributed manner.

[0055] The processor 910 is a device that executes an abnormality detection program. The abnormality detection program is a program that realizes the functions of the model generation unit 110, the reliability calculation unit 120, and the abnormality diagnosis unit 130.

[0056] Processor 910 is an IC that performs calculation processing. Specific examples of processor 910 are CPU, DSP, and GPU. IC is the abbreviation of Integrated Circuit. CPU is the abbreviation of Central Processing Unit. DSP is the abbreviation of Digital Signal Processor. GPU is the abbreviation of Graphics Processing Unit.

[0057] The memory 921 is a storage device for temporarily storing data. A specific example of the memory 921 is SRAM or DRAM. SRAM is the abbreviation of Static Random Access Memory. DRAM is the abbreviation of Dynamic Random Access Memory.

[0058] The auxiliary storage device 922 is a storage device for storing data. A specific example of the auxiliary storage device 922 is an HDD. In addition, the auxiliary storage device 922 may also be a removable storage medium such as an SD (registered trademark) memory card, CF, NAND flash memory, a floppy disk, an optical disk, a high-density disk, a Blu-ray (registered trademark) optical disk, or a DVD. In addition, HDD is the abbreviation of Hard Disk Drive. SD (registered trademark) is the abbreviation of Secure Digital. CF is the abbreviation of CompactFlash (registered trademark; flash memory). DVD is the abbreviation of Digital Versatile Disk.

[0059] The input interface 930 is a port connected to an input device such as a mouse, a keyboard, or a touch panel. Specifically, the input interface 930 is a USB terminal. In addition, the input interface 930 may also be a port connected to a LAN. USB is the abbreviation of Universal Serial Bus. LAN is the abbreviation of Local Area Network.

[0060] The output interface 940 is a port to which a cable of an output device such as a display is connected. Specifically, the output interface 940 is a USB terminal or an HDMI (registered trademark) terminal. Specifically, the display is an LCD. The output interface 940 is also called a display interface. HDMI (registered trademark) is an abbreviation for High Definition Multimedia Interface. LCD is an abbreviation for Liquid Crystal Display.

[0061] The communication device 950 has a receiver and a transmitter. The communication device 950 is connected to a communication network such as a LAN, the Internet, or a telephone line. Specifically, the communication device 950 is a communication chip or a NIC. NIC is an abbreviation of Network Interface Card.

[0062] The abnormality detection program is executed in the abnormality detection device 100. The abnormality detection program is read by the processor 910 and executed by the processor 910. In the memory 921, not only the abnormality detection program but also the OS is stored. OS is the abbreviation of Operating System. The processor 910 executes the abnormality detection program while executing the OS. The abnormality detection program and the OS can also be stored in the auxiliary storage device 922. The abnormality detection program and the OS stored in the auxiliary storage device 922 are loaded into the memory 921 and executed by the processor 910. In addition, part or all of the abnormality detection program can also be incorporated into the OS.

[0063] The abnormality detection device 100 may include a plurality of processors instead of the processor 910. The plurality of processors share the execution of the abnormality detection program. Each processor is a device that executes the abnormality detection program similarly to the processor 910.

[0064] Data, information, signal values, and variable values ​​used, processed, or output by the abnormality detection program are stored in the memory 921 , the auxiliary storage device 922 , or a register or cache memory within the processor 910 .

[0065] The "unit" of each unit of the model generation unit 110, the reliability calculation unit 120, and the abnormality diagnosis unit 130 may be rewritten as "circuit", "process", "procedure", "processing" or "line". The abnormality detection program enables the computer to perform model generation processing, reliability calculation processing and abnormality diagnosis processing. The "processing" of the model generation processing, reliability calculation processing and abnormality diagnosis processing may also be rewritten as "program", "program product", "computer-readable storage medium storing the program", or "computer-readable recording medium recording the program". In addition, the abnormality detection method is a method performed by the abnormality detection device 100 executing the abnormality detection program.

[0066] The abnormality detection program may be provided by being stored in a computer-readable recording medium. In addition, the abnormality detection program may be provided as a program product.

[0067] ***Function description***

[0068] Next, an overview of each functional element of the abnormality detection device 100 according to the present embodiment will be described.

[0069] A sensor signal is data obtained from a sensor installed on a device. A control signal is data indicating an instruction value for controlling a device. In addition, an event signal is data output by a device indicating a set of events such as start / stop and alarm occurrence and their occurrence time.

[0070] The abnormality detection device 100 of this embodiment uses multi-dimensional time series data based on sensor signals obtained from a device having multiple modes as input to determine the abnormality of the device. In addition, the abnormality detection device 100 calculates the reliability of the abnormality detection result for the device based on the time series data of the control signal related to the device control for realizing the multiple modes.

[0071] The model generation unit 110 extracts a feature amount from the data of the control signal, and generates a control signal model 144 which is a model for calculating the degree of abnormality.

[0072] The reliability calculation unit 120 obtains the reliability of the abnormality detection result within a specific period from the time series data of the control signal based on the control signal model 144 .

[0073] The abnormality diagnosis unit 130 generates output content for a user based on the reliability of the abnormality detection result for a specific period of the sensor signal.

[0074] Figure 3 It is a diagram showing a configuration example of the correspondence relationship information 141 according to the present embodiment.

[0075] The correspondence information 141 is data showing the correspondence between the sensor signal and the control signal. In the correspondence information 141, the sensor signal and the control signal affecting the sensor signal are associated with each other. The correspondence information 141 is used, for example, for the abnormality diagnosis unit 130 to select a control signal corresponding to the sensor signal.

[0076] The correspondence information 141 may be any data as long as it includes a set of one or more sensor signal IDs and one or more control signal IDs.

[0077] exist Figure 3 In the example of FIG. 1 , it is shown that the sensor signal S0001 , the sensor signal S0002 , the sensor signal S0003 , and the sensor signal S0004 have a correspondence relationship with the control signal C0001 , the control signal C0002 , and the control signal C0003 .

[0078] Figure 4 It is a diagram showing a configuration example of the abnormality detection method information 142 according to the present embodiment.

[0079] The abnormality detection method information 142 is data indicating a method for detecting abnormality in a sensor signal.

[0080] The abnormality detection method information 142 may be any data as long as it includes one or more sensor signals, an abnormality detection method, input data, dimensions of the input data, a start time, an end time, and parameters of the abnormality detection method.

[0081] exist Figure 4 In the example, isolate-forest-based anomaly detection is performed on sensor signals S0001, S0002, S0003, and S0004. In addition, the time series value at a specific time is used in the input data for the anomaly detection method. In addition, the dimension of the input data is set to the number of sensor signal IDs. Figure 4 In the example, the dimension of the input data becomes 4. In addition, the number of decision trees, which is one of the parameters of isolate-forest, is 100.

[0082] In addition, the start time is the start time of the time series value, and the end time is the end time of the time series value.

[0083] Figure 5 1 is a diagram showing a configuration example of the control signal time series data 143 according to the present embodiment.

[0084] The control signal time series data 143 is time series data of the control signal.

[0085] The control signal time series data 143 includes data for uniquely identifying the control signal, the time when the data is measured from the device, and the command value measured at each time. In addition, the measured time must be at equal intervals except for defects caused by failures, etc.

[0086] exist Figure 5 In the example of , the boiler in the power generation unit is taken as the object, and the data of the indicated value of the temperature of the boiler is recorded in units of 1 minute. In the example, the data for uniquely identifying the control signal is recorded as one time series data of C0001, but multiple control signals can also be stored.

[0087] The control signal model 144 is a model for calculating the degree of abnormality of the control signal. The control signal model 144 is generated by the model generation unit 110 by applying the same abnormality detection method as that used for abnormality detection of sensor signals to the control signal time series data 143 .

[0088] The control signal model 144 may be any data as long as it includes information and a model that can uniquely identify a model stored for each control signal.

[0089] Figure 6 It is a diagram showing a configuration example of the abnormality detection result information 145 according to the present embodiment.

[0090] The abnormality detection result information 145 stores the abnormality detection result of the sensor signal in a specific period.

[0091] The data stored in the abnormality detection result information 145 may be any data as long as it includes data that uniquely identifies the sensor signal, the start time of determination as abnormality, and the end time of determination as abnormality.

[0092] exist Figure 6 In the example of , it is shown that an abnormality occurred in the sensor signal S0001 during the period from “2021-06-01 00:00:00” to “2021-06-01 00:15:00”.

[0093] The abnormality diagnosis result 146 is generated by the abnormality diagnosis unit 130 and presented to the user. The abnormality diagnosis result 146 will be described below.

[0094] ***Action description***

[0095] Next, the operation of the abnormality detection device 100 of this embodiment will be described. The operation process of the abnormality detection device 100 corresponds to the abnormality detection method. In addition, the program that realizes the operation of the abnormality detection device 100 corresponds to the abnormality detection program.

[0096] <Model generation process>

[0097] Figure 7 This is a flowchart showing the operation of the model generation unit 110 according to the present embodiment.

[0098] The model generation unit 110 extracts a feature amount of the control signal time series data 143 which is time series data of the control signal, and generates a control signal model 144 which is a model for calculating the degree of abnormality of each control signal based on the feature amount.

[0099] The following describes an overview of the model generation process.

[0100] The model generation unit 110 uses the time series data of past control signals as training data, and generates a control signal model 144 that outputs the degree of abnormality of the control signal from the time series data of the control signal based on the training data.

[0101] The model generation unit 110 determines the abnormality detection method to be applied to the control signal set in the corresponding relationship information 141 based on the corresponding relationship information 141, the abnormality detection method information 142, and the control signal time series data 143. Then, the model generation unit 110 generates training data input in the abnormality detection method by replacing the data format of the sensor signal input in the abnormality detection method with the data format of the control signal included in the time series data of the past control signal. The model generation unit 110 generates the control signal model 144 by inputting the generated training data in the abnormality detection method.

[0102] Here, the correspondence information 141 is information that corresponds the sensor signal to the control signal that affects the sensor signal. The abnormality detection method information 142 is information that sets the sensor signal, the abnormality detection method applied to the sensor signal, and the data format of the sensor signal input to the abnormality detection method. The control signal time series data 143 is time series data of past control signals.

[0103] Specifically, it is as follows.

[0104] In step S001, the model generation unit 110 determines the abnormality detection method used to generate the control signal model 144 based on the control signal time series data 143. When determining the abnormality detection method, the data of the sensor signal ID and the control signal ID recorded in the correspondence information 141 and the data of the sensor signal ID, input data, and the dimension of the input data recorded in the abnormality detection method information 142 are used.

[0105] Specifically, the model generation unit 110 combines, ie associates, the data of the correspondence information 141 with the data of the abnormality detection method information 142 by using the sensor signal ID. Thus, the model generation unit 110 determines the abnormality detection method associated with the control signal ID of the control signal time series data 143 .

[0106] Furthermore, the model generation unit 110 determines the abnormality detection method for all control signal groups included in the correspondence relationship information 141. There is a case where a control signal group includes only one control signal.

[0107] Figure 8 This is to show that the present embodiment Figure 3 The corresponding relationship information 141 and Figure 4 A diagram showing a state in which the abnormality detection method information 142 is combined.

[0108] according to Figure 8 , Isolate-forest is applied as an abnormality detection method to control signal C0001, control signal C0002 and control signal C0003.

[0109] exist Figure 3 In the correspondence information 141, the group of sensor signals S0001, S0002, S0003, and S0004 corresponds to the group of control signals C0001, C0002, and C0003. Figure 4 In the abnormality detection method information 142, the sensor signals S0001, S0002, S0003, and S0004 correspond to the same abnormality detection method. Figure 4 In the abnormality detection method information 142 , the sensor signals S0001 , S0002 , S0003 , and S0004 correspond to the same abnormality detection method, and therefore one abnormality detection method is determined for the group of control signals C0001 , C0002 , and C0003 .

[0110] On the other hand, for example, when the sensor signals S0001 and S0002 correspond to different abnormality detection methods, the processing is performed as follows.

[0111] In order to simplify the description, the sensor signals are reduced to only S0001 and S0002.

[0112] For example, there is a case where sensor signals S0001 and S0002 correspond to different abnormality detection methods. Specifically, if the abnormality detection method information 142 contains information of "sensor signals S0001, S0002", "sensor signal S0001", and "sensor signal S0002", different abnormality detection methods are set for them, that is, three abnormality detection methods.

[0113] In such a case, all the techniques are applied according to the correspondence information 141. For example, when there are three groups of "sensor signals S0001, S0002 and control signal C0001", "sensor signal S0001 and control signal C0001", and "sensor signal S0002 and control signal C0001" in the correspondence information 141, the following is performed. Three abnormality detection techniques selected according to the groups of the sensor signals are applied to the control signal C0001.

[0114] In step S002 , the model generation unit 110 generates input data based on the data generated in step S001 from the control signal time series data 143 in order to generate input for the abnormality detection method.

[0115] exist Figure 8 In the example of , the time series value at a specific time is input, and thus a three-dimensional vector is generated by combining the data of the control signal C0001, the control signal C0002, and the control signal C0003 at each time.

[0116] exist Figure 8 In the example, the dimension of the input data is the number of sensor signal IDs, so the dimension of the input data is rewritten as the number of control signal IDs. Therefore, the input data becomes a vector set of time series values ​​at the same time in the control signal group. The number of vectors contained in the set of input data is defined as follows based on the start time and the end time.

[0117] The number of vectors included in the input data set = (end time - start time) / acquisition unit of data in the time series value

[0118] Specifically, the acquisition unit of the data in the time series value is 1 minute, 2 minutes, or 3 minutes.

[0119] exist Figure 8 In the input data, when the control signal i is Ci, the time series value of the control signal i at moment t is Ci[t], the start time is start, the end time is end, and the vector is <>, the input data is as follows.

[0120] <C C0001 [start], C C0002 [start], C C0003 [start]>、<C C0001 [start+1], C C0002 [start+1], C C0003 [start+1]>、···、<C C0001[end]、C C0002 [end]、C C0003 [end]>

[0121] In step S003 , the model generation unit 110 generates the control signal model 144 by inputting the input data generated in step S002 to the abnormality detection method determined in step S001 .

[0122] For example, the data generated in step S001 is Figure 8 In the case of information, the model generation unit 110 uses Isolate-forest as an anomaly detection method. The model generation unit 110 sets the number of decision trees, which is one of the parameters of Isolate-forest, to 100, and generates 100 decision trees and anomaly thresholds. The input data used when generating the decision tree is the data generated in step S002.

[0123] In this way, a control signal model 144 for each control signal is generated, which is composed of 100 decision trees and abnormal threshold values.

[0124] In step S004 , the model generation unit 110 stores the control signal model 144 in the storage unit 140 .

[0125] <Reliability calculation processing>

[0126] Fig. 9 : is a flowchart showing the operation of the reliability calculation unit 120 according to the present embodiment.

[0127] The reliability calculation unit 120 calculates reliability indicating whether the training data for each pattern is sufficient when obtaining the abnormality detection result, with respect to the abnormality detection result output by the abnormality detection method in a specific period.

[0128] The following describes an overview of the reliability calculation process.

[0129] The reliability calculation unit 120 inputs the time series data of the control signal within a specific time range of the abnormality detection result detected by the sensor signal as verification data to the control signal model 144. The reliability calculation unit 120 obtains the abnormality degree of the control signal within the specific time range by inputting the verification data to the control signal model 144, and calculates the reliability of the control signal within the specific time range based on the abnormality degree.

[0130] The reliability calculation unit 120 generates verification data based on the correspondence information 141, the abnormality detection technique information 142, the control signal model 144 and the control signal time series data 143 within the specific time range of the abnormality detection result.

[0131] Specifically, it is as follows.

[0132] In step S101, the reliability calculation unit 120 combines the data of the sensor signal ID and control signal ID recorded in the correspondence information 141, the sensor signal ID recorded in the abnormality detection technique information 142, the input data, the dimension of the input data, the start time and the end time through the sensor signal ID.

[0133] In step S102 , the reliability calculation unit 120 generates input data for each control signal based on the data generated in step S101 from the control signal time series data 143 in order to generate input for the abnormality detection method. This process is the same as step S002 .

[0134] In the reliability calculation process, the start time and end time are rewritten as Figure 6 The abnormality detection result information 145 of the abnormality detection result information 145 is set as the "abnormality start date and time" to the "abnormality end date and time". In the reliability calculation process, the "abnormality start date and time" to the "abnormality end date and time" are set as the specific time range. Figure 6 In the example, the number of vectors included in the set of input data is defined as follows based on the abnormality start time and the abnormality end time.

[0135] Number of vectors included in the input data set = (abnormal end time - abnormal start time) / acquisition unit of data in time series value

[0136] The input data generated in the model generation process is the training data in machine learning. On the other hand, the input data for the reliability calculation process is the estimated data, that is, the verification data. The method of creating data is the same except that the period of the time series data is different.

[0137] In step S103, the reliability calculation unit 120 calculates the degree of abnormality within a specific time range from the start time to the end time of each control signal using the control signal model 144 and the input data generated in step S102. In addition, each control signal includes the meaning of each control signal group.

[0138] The control within a specific time range of each control signal refers to the fine control of each device for realizing a mode. That is, the abnormality level within a specific time range of each control signal refers to the abnormality level of each mode.

[0139] For example, when the abnormality detection method is Isolate-forest, the reliability calculation unit 120 inputs the input data generated in step S102 to the decision tree stored in the control signal model 144, and takes the average of the results of all decision trees as the degree of abnormality within a specific time range of each control signal.

[0140] Here, a specific example of the relationship between the control signal and the mode will be described in the device of the power generation group.

[0141] Mode refers to the state of stopping, generating 80% of power, and generating 100% of power in the generator set. The control signal is the time series data of the instruction value that indicates how each component of the unit should act when stopping. For example, when stopping the generator set, various valves are closed to stop the supply of fuel. At this time, the instruction value "close" is sent to the valve of the valve. "Stop" in this example is the mode, and the numerical value indicating the action of "closing" the valve of the valve becomes the control signal.

[0142] The specific time range in the reliability calculation process becomes the range in which the sensor signal recorded in the abnormality detection result information 145 is judged to be abnormal. The abnormality detection device of this embodiment has the following function: using the time series data of the control signal, it has the following function: to obtain the reliability of the judgment of "whether it is abnormal" obtained from the time series data of the sensor signal. Therefore, the specific time range basically becomes the period of the training data of the abnormality detection method for the sensor signal and the period in which it is judged to be "abnormal" by the verification data. The specific time range in the reliability calculation process becomes the period in which the abnormality detection method for the sensor signal is judged to be "abnormal" by the verification data.

[0143] In step S104, the reliability calculation unit 120 calculates the reliability of each specific time range in each control signal based on the abnormality degree obtained in step S103. As an example of a method for calculating the reliability, when the abnormality degree is set as amn and the threshold value for judging the abnormality degree as abnormal is set as thr, the calculation is performed by the following formula. In the following formula, if the abnormality degree is smaller than the abnormal threshold value, the reliability becomes greater than 1, and if the abnormality degree is greater than the abnormal threshold value, the reliability becomes less than 1.

[0144] Reliability = 1 / (amn / thr)

[0145] In addition, as another example of a calculation method of reliability accuracy, it can also be defined as the following formula. In this case, if the abnormality is greater than the abnormality threshold, the reliability becomes greater than 1, and if the abnormality is less than the abnormality threshold, the reliability becomes less than 1.

[0146] Reliability = anm / thr

[0147] <Abnormal diagnosis and processing>

[0148] The abnormality diagnosis unit 130 generates an abnormality diagnosis result based on the correspondence relationship information 141 , from the abnormality detection result of the sensor signal stored in the abnormality detection result information 145 of the sensor signal and the reliability of the abnormality diagnosis calculated by the reliability calculation unit 120 .

[0149] The abnormality diagnosis unit 130 outputs information obtained by adding the reliability of the control signal within the specific time range to the abnormality detection result within the specific time range as the abnormality diagnosis result.

[0150] Specifically, it is as follows.

[0151] Fig.10 1 is a diagram showing an example of a method of obtaining an abnormality diagnosis result by the abnormality diagnosis unit 130 of the present embodiment.

[0152] Fig.11 1 is a flowchart showing the operation of the abnormality diagnosis unit 130 according to the present embodiment.

[0153] exist Fig.10 In , the anomaly detection results and reliability represent the results for the same period.

[0154] In step S201, the abnormality diagnosis unit 130 obtains the correspondence relationship between the sensor signal and the control signal from the correspondence relationship information 141. Fig.10 In the example of , it is a group of sensor signals S0001, S0002, S0003, and S0004 and a group of control signals C0001, C0002, and C0003.

[0155] In step S202, the abnormality diagnosis unit 130 obtains the abnormality detection result of the sensor signal from the abnormality detection result information 145 of the sensor signal based on the data obtained in step S201. In addition, the abnormality diagnosis unit 130 obtains the reliability of the control signal from the reliability output by the reliability calculation unit 120. Fig.10 In the example, the sensor signal is abnormal and the reliability of the control signal becomes 0.3.

[0156] In step S203, the reliability of the abnormal detection result of a certain sensor signal is determined based on the correspondence between the sensor signal and the control signal, as well as the abnormal detection result of the sensor signal and the reliability of the control signal. If the reliability is less than 1, the words "control state with very little training data" are added to the prefix of the abnormal detection result. Fig.10 In the example, the reliability is 0.3, so it is determined that it is necessary to add "control state with very little training data" to the prefix of the abnormality detection result.

[0157] In step S204, an abnormality diagnosis result consisting of a combination of each sensor signal, an abnormality detection result, and reliability using the result of step S203 is generated and output. Fig.10 In the example shown in FIG. 1 , the group of sensor signals S0001 , S0002 , S0003 , and S0004 is detected as abnormal in a mode with very little training data, and the reliability of abnormality detection is 0.3.

[0158] **Description of the Effects of the Embodiment**

[0159] In the abnormality detection device of the present embodiment, the model generation unit extracts the time series data of the control signal, i.e., the feature quantity of the control signal time series data, and generates a model for calculating the degree of abnormality of each control signal, i.e., the control signal model, based on the feature quantity. In addition, the reliability calculation unit inputs the time series data of the control signal to the control signal model, thereby calculating the reliability indicating whether the past data is sufficient for the period of the detection result of an abnormality detection method applied to the sensor signal obtained from the sensor. In addition, the abnormality diagnosis unit combines the reliability with the abnormality detection result of the abnormality detection method applied to the sensor signal based on the abnormality detection result information storing the abnormality detection result for the sensor signal, and uses the reliability to correct the abnormality detection result of the abnormality detection method applied to the sensor signal. Then, the abnormality diagnosis unit adds the reliability to the abnormality detection result and outputs it as the abnormality diagnosis result.

[0160] In addition, in the abnormality detection device of the present embodiment, the model generation unit determines the group of control signals input for calculating reliability according to each abnormality detection method based on the correspondence information, the abnormality detection method information and the control signal time series data. Then, the model generation unit generates the control signal model based on the abnormality detection method for the group of control signals. The correspondence information is information obtained by corresponding the control signal and the sensor signal affected by the control signal. The abnormality detection method information is information that stores the combination of the abnormality detection method of the device and the sensor signal input in the abnormality detection method.

[0161] In addition, in the abnormality detection device of the present embodiment, the reliability calculation unit generates input data to be input into the control signal model based on the correspondence information, the abnormality detection technique information, the control signal model and the control signal time series data, and calculates the reliability by inputting the input data into the control signal model.

[0162] Thus, in the abnormality detection device of the present invention, the same method as the outlier detection used in the abnormality detection of the sensor signal is applied to the time series data of the control signal representing the mode to generate a control signal model for calculating the degree of abnormality of the control signal. Therefore, according to the abnormality detection device of the present invention, the reliability of the abnormality detection result determined based on the time series data of the sensor signal can be calculated according to the degree of abnormality of the control signal. Thus, according to the abnormality detection device of the present invention, it is possible to immediately determine whether the time series data of the sensor signal diagnosed by the abnormality detection device is determined to be abnormal in a mode with sufficient data volume or in a mode with very little data volume.

[0163] In the abnormality detection device of this embodiment, the reliability calculation unit determines whether the pattern is rare based on the abnormality of the control signal, and calculates the reliability of the abnormality detection result of the sensor signal affected by the control. Determining whether the pattern is rare means determining whether the training data of "a certain pattern" is very small. That is, for the entire training data including multiple patterns, it is determined whether the training data of only a certain pattern is very small or there is no training data.

[0164] In addition, in the abnormality detection device of the present embodiment, the abnormality diagnosis result of the sensor signal output to the person is corrected based on reliability by the abnormality diagnosis unit 130. As a result, in the abnormality detection of the device where the control occurs, it is possible to detect whether it is a control that occurred in the past, and reflect it in the result of the abnormality detection. Therefore, it is possible to immediately determine whether the time series data of the sensor signal diagnosed by the abnormality detection device is judged to be abnormal in a mode with sufficient data volume or in a mode with very small data volume.

[0165] **Other structures**

[0166] <Variation 1>

[0167] In the first modification of the present embodiment, Figure 7 , 9 , 11, a modified example of the processing flow of the abnormality detection device 100 described.

[0168] The processing of the abnormality detection device 100 can be represented as two processing flows: a learning process for generating the control signal model 144 and an estimation process for obtaining the degree of abnormality based on the control signal model 144 .

[0169] The learning process for generating the control signal model 144 corresponds to the above-mentioned model generation process.

[0170] Obtaining an estimate of the degree of abnormality based on the control signal model 144 corresponds to the above-mentioned reliability calculation process.

[0171] Fig.12 1 is a flowchart showing a learning process for generating the control signal model 144 in the abnormality detection process of the present embodiment.

[0172] Fig.13 FIG. 1 is a flowchart showing an estimation process of finding the degree of abnormality based on the control signal model 144 in the abnormality detection process of the present embodiment. Fig.13 This also includes abnormal diagnosis and processing.

[0173] exist Fig.12 In the illustrated learning flow, the model generation unit 110 , correspondence relationship information 141 , abnormality detection technique information 142 , control signal time series data 143 , and a control signal model 144 are used.

[0174] In step S301 , the model generation unit 110 obtains the relationship between the group of sensor signals and the group of control signals from the correspondence relationship information 141 .

[0175] In step S302 , the model generation unit 110 acquires the abnormality detection method applied to a specific sensor signal, information related to the input, and parameters of the abnormality detection method from the abnormality detection method information 142 .

[0176] In step S303 , the model generation unit 110 acquires the time series data of the control signal from the control signal time series data 143 .

[0177] In step S304, the model generation unit 110 generates a model for each control signal group based on the data obtained in steps S301, S302, and S303. Figure 7 It's the same.

[0178] In step S305 , the model generation unit 110 stores all the models obtained in step S304 in the storage unit 140 as the control signal model 144 .

[0179] Fig.12 The processing is equivalent to Figure 7 processing.

[0180] exist Fig.13 In the illustrated estimation flow, the reliability calculation unit 120 , correspondence information 141 , abnormality detection technique information 142 , control signal time series data 143 , a control signal model 144 , abnormality detection result information 145 , and abnormality diagnosis result 146 are used.

[0181] In step S401 , the reliability calculation unit 120 obtains the relationship between the group of sensor signals and the group of control signals from the correspondence relationship information 141 .

[0182] In step S402 , the reliability calculation unit 120 obtains the abnormality detection method applied to a specific sensor signal, information related to the input, and parameters of the abnormality detection method from the abnormality detection method information 142 .

[0183] In step S403 , the reliability calculation unit 120 acquires the time series data of the control signal from the control signal time series data 143 .

[0184] In step S404 , the reliability calculation unit 120 obtains a model corresponding to the control signal from the control signal model 144 .

[0185] In step S405, the reliability calculation unit 120 calculates the reliability of the abnormality detection results of the sensor signal group corresponding to the control signal group recorded in the correspondence information 141 based on the data obtained in steps S401, S402, S403 and S404. Fig. 9 The flowchart described in the reliability calculation process is the same.

[0186] In step S406 , the abnormality diagnosis unit 130 acquires the abnormality detection result of each sensor signal from the sensor signal abnormality detection result information 145 .

[0187] In step S407, the abnormality diagnosis unit 130 converts the abnormality detection results of all sensor signals based on the information obtained from steps S401, S405, and S406, and generates an abnormality diagnosis result 146 consisting of a set of sensor signals, abnormality diagnosis results, and reliability. Fig.11 It's the same.

[0188] Fig.13 The processing is equivalent to Fig. 9 and Fig.11 processing.

[0189] <Variation 2>

[0190] In this embodiment, the functions of the model generation unit 110, the reliability calculation unit 120, and the abnormality diagnosis unit 130 are implemented by software. As a modification, the functions of the model generation unit 110, the reliability calculation unit 120, and the abnormality diagnosis unit 130 may also be implemented by hardware.

[0191] Specifically, the abnormality detection device 100 includes an electronic circuit 909 instead of the processor 910 .

[0192] Fig.14 This is a diagram showing a configuration example of an abnormality detection device 100 according to a second modification of the present embodiment.

[0193] The electronic circuit 909 is a dedicated electronic circuit that realizes the functions of the model generation unit 110, the reliability calculation unit 120, and the abnormality diagnosis unit 130. Specifically, the electronic circuit 909 is a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, a logic IC, a GA, an ASIC, or an FPGA. GA is the abbreviation of Gate Array. ASIC is the abbreviation of Application Specific Integrated Circuit. FPGA is the abbreviation of Field-Programmable Gate Array.

[0194] The functions of the model generation unit 110 , the reliability calculation unit 120 , and the abnormality diagnosis unit 130 may be implemented by one electronic circuit or may be implemented by distributing them among a plurality of electronic circuits.

[0195] As another variation, part of the functions of the model generation unit 110, the reliability calculation unit 120, and the abnormality diagnosis unit 130 may be implemented by electronic circuits, and the remaining functions may be implemented by software. In addition, part or all of the functions of the model generation unit 110, the reliability calculation unit 120, and the abnormality diagnosis unit 130 may be implemented by firmware.

[0196] The processor and the electronic circuit are also referred to as processing circuits, respectively. That is, the functions of the model generation unit 110, the reliability calculation unit 120, and the abnormality diagnosis unit 130 are implemented by the processing circuits.

[0197] Implementation method 2.

[0198] In this embodiment, points different from the first embodiment and points added to the first embodiment are mainly described.

[0199] In this embodiment, the same reference numerals are given to components having the same functions as those in Embodiment 1, and description thereof will be omitted.

[0200] In this embodiment, a case where the abnormality detection method is a method for a single signal and one sensor signal is affected by a plurality of control signals will be described.

[0201] ***Structure description***

[0202] In the present embodiment, the data formats of the correspondence information 141 and the abnormality detection method information 142 are different from those in the first embodiment.

[0203] Other structures and Figure 1 and Figure 2 The structure described in is the same.

[0204] Fig.15 It is a diagram showing a configuration example of the correspondence relationship information 141 according to the present embodiment.

[0205] The data stored in the correspondence information 141 may be any data as long as it includes a set of one sensor signal ID, one or more control signal IDs, and conditions used in reliability calculation.

[0206] exist Fig.15 In the example, it is shown that the sensor signal S0001 and the control signal C0001, and the sensor signal S0002 and the control signals C0001 and C0002 have a corresponding relationship. In addition, for the sensor signal S0002 and the control signals C0001 and C0002, the reliability combination method and the reliability output content are saved as the reliability determination of the abnormality detection result. The combination method is logical sum, and the reliability output content is the minimum.

[0207] Fig.16 It is a diagram showing a configuration example of the abnormality detection method information 142 according to the present embodiment.

[0208] The data stored in the abnormality detection method information 142 includes a sensor signal, an abnormality detection method, input data, the dimension of the input data, and the parameters of the abnormality detection method. Fig.16 In the example, anomaly detection based on the k-nearest neighbor method is performed on sensor signal S0001 and sensor signal S0002. In addition, in the input of anomaly detection, sensor signal S0001 is time series data, while sensor signal S0002 is a high-frequency component. The number of dimensions of the input data of both sensor signals is 10. Regarding the number of neighbors as a parameter, sensor signal S0001 is 2, and sensor signal S0002 is 3.

[0209] In addition, Fig.16 The start time and end time are omitted in the abnormality detection method information 142. Fig.16 The abnormality detection method information 142 also includes information on the start time and the end time as in the first embodiment.

[0210] ***Action description***

[0211] Next, the operation of the abnormality detection device 100 of this embodiment will be described. The operation process of the abnormality detection device 100 corresponds to the abnormality detection method. In addition, the program that realizes the operation of the abnormality detection device 100 corresponds to the abnormality detection program.

[0212] In this embodiment, a sensor signal and a control signal group are associated with each other, and a reliable combination method in the control signal group is set in the correspondence information 141. The model generation unit 110 applies an abnormality detection method corresponding to the sensor signal to each control signal in the control signal group, and generates a control signal model 144.

[0213] The reliability calculation unit 120 calculates the reliability of the control signal group using a combination method of reliability.

[0214] Specifically, it is as follows.

[0215] <Model generation process>

[0216] The model generation unit 110 generates a control signal model 144 for calculating the degree of abnormality indicating the degree of abnormality based on the time series data of the control signal stored in the control signal time series data 143. The basic processing flow of the model generation unit 110 is similar to that described in the first embodiment. Figure 7 The present invention is the same as that of the first embodiment, but the following describes the points different from the first embodiment.

[0217] In step S001, the model generation unit 110 determines the abnormality detection method used to generate the control signal model. In the determination, the sensor signal ID and the control signal ID recorded in the correspondence information 141, and the sensor signal ID, input data, and the data of the dimension of the input data recorded in the abnormality detection method information 142 are used. Specifically, the model generation unit 110 combines the data of each storage area by the sensor signal ID, and makes a determination based on the control signal ID and the abnormality detection method of the combined data.

[0218] Fig.17 This is a diagram showing an example of the result of combining the correspondence relationship information 141 and the abnormality detection technique information 142 according to the present embodiment.

[0219] according to Fig.17 , the abnormality detection method of the sensor signal S0001 corresponds to the control signal C0001, and the abnormality detection method of the sensor signal S0002 corresponds to the control signals C0001 and C0002. However, the control signals C0001 and C0002 are abnormality detections for two signals, and therefore are separated in a manner of becoming one control signal. Therefore, for the control signal C0001, abnormality detection is performed by both the abnormality detection method implemented on the sensor signal S0001 and the abnormality detection method implemented on the sensor signal S0002.

[0220] In addition, as mentioned above, Fig.17 The start and end times are also omitted.

[0221] In step S002, the model generation unit 110 extracts input data based on the information of the data combined in the previous stage from the data stored in the control signal time series data 143 in order to generate the input of the abnormality detection method. Fig.17 In the example of , the model generation unit 110 extracts 10-dimensional partial time series data for the control signal C0001, and extracts 10-dimensional high-frequency components for the control signal C0001 and the control signal C0002.

[0222] In step S003, the model generation unit 110 generates a model for the anomaly detection method by inputting the data generated in step S002. The parameters of the anomaly detection method use the parameters recorded in the data generated in step S001. Fig.17 The model generated for the signal whose control signal ID is C0001 is obtained when the information is . The k-nearest neighbor method is used as an abnormality detection method. In addition, for the signal whose control signal ID is C0001, 10-dimensional time series data and 10-dimensional input data of high-frequency components are used. In the k-nearest neighbor method where the input data is time series data, the number of neighbors as one of the parameters is set to 2, and in the k-nearest neighbor method where the input data is a high-frequency component, the number of neighbors as one of the parameters is set to 3. Thus, a model that maps the input data to a 10-dimensional space and an abnormal threshold are generated.

[0223] In step S004 , the model generation unit 110 stores the model generated in step S003 in the storage unit 140 as the control signal model 144 .

[0224] <Reliability calculation processing>

[0225] The operation flow of the reliability calculation unit 120 is the same as that of Embodiment 1. However, in step S101 , similarly to step S001 of the model generation unit 110 , when there are two or more control signals for one sensor signal, the control signal is separated into one.

[0226] <Abnormal diagnosis and processing>

[0227] The abnormality diagnosis unit 130 generates an abnormality diagnosis result based on the correspondence relationship information 141 , from the abnormality detection result information 145 of the sensor signal and the reliability of the abnormality diagnosis calculated by the reliability calculation unit 120 .

[0228] Fig.18 1 is a diagram showing an example of a method of obtaining an abnormality diagnosis result by the abnormality diagnosis unit 130 of the present embodiment.

[0229] Fig.191 is a flowchart showing the operation of the abnormality diagnosis unit 130 according to the present embodiment.

[0230] exist Fig.18 In , the anomaly detection results and reliability represent the results for the same period.

[0231] In step S201, the abnormality diagnosis unit 130 obtains the correspondence relationship between the sensor signal and the control signal from the data stored in the correspondence relationship information 141. Fig.18 In the example of FIG. 1 , the sensor signal S0001 corresponds to the control signal C0001 , and the sensor signal S0002 corresponds to the control signals C0001 and C0002 .

[0232] The processing of step S201 is Fig.11 The step S201 described in is the same.

[0233] In step S202 , based on the data obtained in step S201 , the abnormality diagnosis unit 130 obtains the abnormality detection result of the sensor signal from the abnormality detection result information 145 of the sensor signal, and obtains the reliability of the control signal of each abnormality detection result of the sensor signal from the reliability output by the reliability calculation unit 120 .

[0234] exist Fig.18 In the example, sensor signal S0001 is normal, sensor signal S0002 is abnormal, the reliability of control signal C0001 using the abnormality detection method of sensor signal S0001 is 1.2, the reliability of control signal C0001 using the abnormality detection method of sensor signal S0002 is 1.4, and the reliability of control signal C0002 using the abnormality detection method of sensor signal S0002 is 0.3.

[0235] The processing of step S202 is Fig.11 The step S202 described in is the same.

[0236] In step S203a, the abnormality diagnosis unit 130 converts the reliability of the control signal into a logical value. Here, the case where the reliability is sufficient and the value is 1 or above is converted to 0, and the case where the reliability is insufficient and the value is 1 or below is converted to 1. Fig.18 In the example, the logic value of the control signal C0001 using the abnormality detection method of the sensor signal S0001 is 0, the logic value of the control signal C0001 using the abnormality detection method of the sensor signal S0002 is 0, and the logic value of the control signal C0002 using the abnormality detection method of the sensor signal S0002 is 1.

[0237] In step S203b, the abnormality diagnosis unit 130 determines whether the reliability of the abnormality detection result of a certain sensor signal is sufficient based on the correspondence between the sensor signal and the control signal, the abnormality detection result of the sensor signal, and the result of step S203a. Specifically, for the logical value of the reliability of the control signal corresponding to each sensor signal, if the result determined by the combination method recorded by the correspondence between the signals is 1, the words "control state with very little training data" are added to the prefix of the abnormality detection result. Fig.18 In the example, the result of the sufficiency of the training data of the sensor signal S0001 is 0, and the result of the sufficiency of the training data of the sensor signal S0002 is 1. Therefore, it is determined that "control state with very little training data" needs to be added to the prefix of the abnormality detection result of the sensor signal S0002.

[0238] The processing of step S203a and step S203b corresponds to Fig.11 This embodiment is different from the first embodiment in that the reliability of the control signal is converted into a logical value.

[0239] In step S204, the abnormality diagnosis unit 130 generates and outputs an abnormality diagnosis result composed of a combination of each sensor signal, an abnormality detection result, and reliability using the result of step S204. Fig.18 In the example, sensor signal S0001 is judged as normal in a mode with sufficient training data, and sensor signal S0002 is judged as abnormal in a mode with very little training data. Regarding reliability, for sensor signal S0001, the reliability of control signal C0001 is directly output. For sensor signal S0002, control signal C0001 is compared with control signal C0002, and the reliability of control signal C0002 with the smallest value is output.

[0240] The processing of step S204 is Fig.11 The step S204 described in is the same.

[0241] There are two processes in the abnormality detection device 100: a learning process for generating a model and a process for estimating the degree of abnormality from the model. The learning process and the estimation process are the same as those in the first embodiment.

[0242] **Description of the Effects of the Embodiment**

[0243] In the abnormality detection device of this embodiment, in the correspondence information, a combination method for control signals and sensor signals is added to the existing data. In addition, in the abnormality detection device of this embodiment, the constraint related to the input of the abnormality detection method is only one signal. In the case where the sensor signal and the control signal are one to many, a control signal model is generated based on the abnormality detection method according to each control signal.

[0244] In addition, in the abnormality detection device of the present embodiment, the reliability calculation unit bundles the reliability obtained based on the control signal model based on the combination method to generate a reliability corresponding to the abnormality detection result. In this way, in the abnormality detection device of the present embodiment, even if the control signal corresponding to the specific sensor signal does not satisfy the constraints related to the input of the abnormality detection method, the sufficiency of the past data can be determined by bundling the model of the control signal of the existing abnormality detection method.

[0245] According to the abnormality detection device of the present embodiment, even when the abnormality detection method is a method for a single signal and one sensor signal is affected by a plurality of control signals, the same effects as those of the first embodiment can be obtained.

[0246] Implementation method 3.

[0247] In this embodiment, points different from Embodiments 1 and 2 and points added to Embodiments 1 and 2 are mainly described.

[0248] In this embodiment, the same reference numerals are given to components having the same functions as those in Embodiments 1 and 2, and description thereof will be omitted.

[0249] In this embodiment, a solution will be described when an imbalance occurs between modes.

[0250] ***Structure description***

[0251] The hardware structure of this embodiment is similar to that described in Embodiment 1. Figure 1 It's the same.

[0252] Fig. 20 1 is a diagram showing a functional configuration example of the abnormality detection device 100 according to the present embodiment.

[0253] The abnormality detection device 100 of the present embodiment includes a pattern division unit 150 in addition to the functional elements of the first embodiment. The pattern division unit 150 divides data according to control content.

[0254] Furthermore, the abnormality detection device 100 of the present embodiment stores event signal information 147 in addition to the configuration of the first embodiment in the storage unit 140. The event signal information 147 stores log data of event signals.

[0255] In the storage unit 140, the control signal model generated by the model generation unit 110 is stored in the control signal model 144. On the other hand, the corresponding relationship information 141, the abnormality detection method information 142, the control signal time series data 143, the abnormality detection result information 145 of the sensor signal, and the event signal information 147 need to store the corresponding data before the execution of the abnormality detection device 100. In addition, the corresponding relationship information 141, the abnormality detection method information 142, the control signal time series data 143, and the abnormality detection result information 145 of the sensor signal are equivalent to those in the first embodiment or the second embodiment.

[0256] The control signal model 144 storing the control signal model generated by the model generation unit 110 may be any data as long as it includes information and a model that can uniquely identify the model stored by the control signal and the pattern.

[0257] Fig.21 It is a diagram showing a configuration example of the event signal information 147 according to the present embodiment.

[0258] The data stored in the event signal information 147 may be any data as long as it includes a set of date and time when the event occurred in addition to event information such as control content and warning. Fig.21 This is an example of data saved for a generator set. Fig.21 In the example, it is shown that the adjustment of the power generation starts at 1:00 on January 1, 2020, the adjustment of the power generation ends at 1:30 on January 1, 2020, the boiler temperature warning occurs at 12:00 on January 1, 2020, and the adjustment of the power generation starts again at 12:00 on January 2, 2020.

[0259] The event signal indicates an indication value of a larger unit than the control signal. For example, in the case of stopping a generator set, a log such as "Stop the generator set" corresponds. Therefore, the event signal is not time series data recorded at equal intervals (for example, 1 minute intervals) like the control signal, but becomes log data recorded irregularly. In this way, the control signal indicates "the fine control of each device when the generator set is stopped", and the pattern indicates the result of a larger control such as "stopping the generator set".

[0260] ***Action description***

[0261] Next, the operation of the abnormality detection device 100 of this embodiment will be described. The operation process of the abnormality detection device 100 corresponds to the abnormality detection method. In addition, the program that realizes the operation of the abnormality detection device 100 corresponds to the abnormality detection program.

[0262] In this embodiment, the abnormality detection device includes a pattern segmentation unit 150 that segments time series data of control signals for each control state based on event signal information storing log data of event signals indicating control states or alarms of equipment.

[0263] The model generation unit 110 generates a control signal model 144 based on the time series data of the control signal for each mode.

[0264] Specifically, it is as follows.

[0265] The model generation process performed by the model generation unit 110 is the same as that in the first embodiment or the second embodiment.

[0266] The reliability calculation process performed by the reliability calculation unit 120 is the same as that in the first embodiment or the second embodiment.

[0267] The abnormality diagnosis process performed by the abnormality diagnosis unit 130 is the same as that in the first embodiment or the second embodiment.

[0268] <Mode division processing>

[0269] Fig. 22 This is to show that the event signal information 147 of this embodiment is set to Fig.21 FIG. 1 is an example of pattern segmentation in the case of .

[0270] Fig.23 This is a flowchart showing the operation performed by the mode division unit 150 according to the present embodiment.

[0271] The pattern dividing unit 150 divides the data into each specific pattern based on the data stored in the event signal information 147 .

[0272] In step S501, the mode segmentation unit 150 extracts only event information for determining a mode from the data stored in the event signal information 147. Fig.21 In the example of , the warning is different from the information of the confirmation mode, so it is excluded. As a result, the remaining logs are only those related to the power generation amount adjustment.

[0273] In step S502 , the pattern segmentation unit 150 obtains a period during which a specific pattern exists, using the event information for identifying the pattern extracted in step S501 and the time.

[0274] exist Fig. 22 In the example, the period from 1:00 on January 1, 2020, when the power generation adjustment starts, to 1:30 on January 1, 2020, when the power generation adjustment ends, and the period from 1:30 on January 1, 2020, when the power generation adjustment ends, to 12:00 on January 2, 2020, when the power generation adjustment starts, are extracted as the period.

[0275] In step S503, the pattern of each period is determined based on the event information at the beginning of each period. Fig.21 In the period from 1:00 on January 1, 2020 to 1:30 on January 1, 2020, the event information of the start date and time is the start of power generation adjustment, so it corresponds to the transition period of power generation adjustment. Similarly, the period from 1:30 on January 1, 2020 to 12:00 on January 2, 2020 corresponds to the period when power generation adjustment is completed and the operation is performed at 100% power generation.

[0276] The abnormality detection device 100 includes two processes: a learning process for generating the control signal model 144 and a process for obtaining an estimate of the degree of abnormality based on the control signal model 144 .

[0277] The learning for generating the control signal model 144 corresponds to the above-mentioned model generation process.

[0278] Obtaining an estimate of the degree of abnormality based on the control signal model 144 corresponds to the above-mentioned reliability calculation process.

[0279] The pattern division process of the present embodiment is performed before each of the model generation process and the reliability calculation process.

[0280] Fig.24 14 is a flowchart showing a process during learning for generating the control signal model 144 according to the present embodiment.

[0281] In step S601 , the pattern division unit 150 obtains data from the event signal information 147 .

[0282] In step S602, the mode segmentation unit 150 performs mode segmentation based on the data acquired in step S601. Fig.23 It's the same.

[0283] In step S301 , the model generation unit 110 obtains the relationship between the group of sensor signals and the group of control signals from the correspondence relationship information 141 .

[0284] In step S302 , the model generation unit 110 acquires the abnormality detection method applied to a specific sensor signal, information related to the input, and parameters of the abnormality detection method from the abnormality detection method information 142 .

[0285] In step S303 , the model generation unit 110 acquires the time series data of the control signal from the control signal time series data 143 .

[0286] Steps S301 to S303 and Fig.12 It's the same.

[0287] Step S304a is equivalent to Fig.12 Step S304, step S305a is equivalent to Fig.12 Step S305.

[0288] In step S304 a , the model generation unit 110 generates a model of each control signal as the control signal model 144 for each mode based on the data acquired in steps S602 , S301 , S302 , and S303 .

[0289] In step S305 a , the model generation unit 110 stores the control signal model 144 for all control signals for each mode obtained in step S304 a in the storage unit 140 .

[0290] Step S304a and step S305a are repeatedly performed for all patterns (step S603).

[0291] Fig.25 1 is a flowchart showing the process of obtaining an estimate of the degree of abnormality based on the control signal model 144 of the present embodiment.

[0292] Step S601 and step S602 are Fig.24 Step S601 and step S602 are the same.

[0293] In step S401 , the reliability calculation unit 120 obtains the relationship between the group of sensor signals and the group of control signals from the correspondence relationship information 141 .

[0294] In step S402 , the reliability calculation unit 120 obtains the abnormality detection method applied to a specific sensor signal, information related to the input, and parameters of the abnormality detection method from the abnormality detection method information 142 .

[0295] In step S403 , the reliability calculation unit 120 acquires the time series data of the control signal from the control signal time series data 143 .

[0296] Steps S401 to S403 and Fig.13 It's the same.

[0297] Step S404a is equivalent to Fig.13 Step S404, step S405a is equivalent to Fig.13Step S405.

[0298] In step S404 a , the reliability calculation unit 120 obtains a model for each mode corresponding to the control signal from the control signal model 144 .

[0299] In step S405 a , the reliability calculation unit 120 calculates the reliability for each pattern with respect to the abnormality detection results of all sensor signals based on the data acquired in steps S602 , S401 , S402 , S403 , and S404 a .

[0300] Step S406 and step S407 are Fig.13 Step S406 and step S407 are the same.

[0301] **Description of the Effects of the Embodiment**

[0302] The abnormality detection device of this embodiment is based on a pattern segmentation unit, which segments the control signal time series data according to each control state based on event signal information, wherein the event signal information stores log data of the event signal indicating the control state or alarm of the device. The model generation unit generates a model based on the control signal according to each pattern.

[0303] According to the abnormality detection device of this embodiment, even when imbalance occurs between modes, the same effects as those of the first and second embodiments can be obtained.

[0304] In the above embodiments 1 to 3, each part of the abnormality detection device is described as an independent functional block. However, the structure of the abnormality detection device may not be the structure of the above embodiment. The functional block of the abnormality detection device can be any structure as long as it can realize the functions described in the above embodiment. In addition, the abnormality detection device may not be a single device, but a system composed of multiple devices.

[0305] For example, the abnormality detection device may not be equipped with a model generation unit. The model generation unit may also be installed on a server or other device, and the control signal model may be generated by the server or other device. Alternatively, the abnormality detection device may use a pre-prepared control signal model.

[0306] In addition, a plurality of parts in Embodiments 1 to 3 may be implemented in combination. Alternatively, one part of these embodiments may be implemented. In addition, these embodiments may be implemented in combination in whole or in part.

[0307] That is, in Embodiments 1 to 3, it is possible to freely combine the embodiments or to modify any constituent element of the embodiments, or to omit any constituent element in the embodiments.

[0308] In addition, the above-mentioned embodiments are essentially preferred examples and are not intended to limit the scope of the present disclosure, the scope of the application objects of the present disclosure, and the scope of the use of the present disclosure. The above-mentioned embodiments can be modified in various ways as needed. For example, the process described using a flowchart or a timing diagram can also be appropriately modified.

[0309] Description of Reference Numerals

[0310] 100 abnormality detection device, 110 model generation unit, 120 reliability calculation unit, 130 abnormality diagnosis unit, 140 storage unit, 150 pattern segmentation unit, 141 corresponding relationship information, 142 abnormality detection technique information, 143 control signal time series data, 144 control signal model, 145 abnormality detection result information, 146 abnormality diagnosis result, 147 event signal information, 909 electronic circuit, 910 processor, 921 memory, 922 auxiliary storage device, 930 input interface, 940 output interface, 950 communication device.

Claims

1. An abnormality detection device, comprising a sensor for detecting an abnormality in a device controlled by a control signal, wherein: The abnormality detection device comprises: a reliability calculation unit that inputs time series data of a control signal within a specific time range of an abnormality detection result detected in a sensor signal obtained from the sensor as verification data into a control signal model, thereby obtaining a degree of abnormality of the control signal within the specific time range, and calculating the reliability of the control signal within the specific time range, the control signal model being a model that outputs the degree of abnormality of each control signal based on the time series data of the control signal generated based on training data, the training data being a feature amount of the time series data of the control signal in the past; as well as The abnormality diagnosis unit outputs information obtained by adding the reliability of the control signal within the specific time range to the abnormality detection result within the specific time range as the abnormality diagnosis result.

2. The abnormality detection device according to claim 1, wherein: The abnormality detection device includes a model generation unit that extracts a feature amount of time series data of the past control signal as the training data and generates the control signal model based on the training data.

3. The abnormality detection device according to claim 2, wherein: The model generation unit determines the abnormality detection method applied to each control signal set in the correspondence information based on the correspondence information, the abnormality detection method information, and the time series data of the control signal. The correspondence information is obtained by corresponding the sensor signal with the control signal that affects the sensor signal. The sensor signal, the abnormality detection method applied to the sensor signal, and the data format of the sensor signal input in the abnormality detection method are set in the abnormality detection method information. The model generation unit generates the training data input in the abnormality detection method by replacing the data format of the sensor signal input in the abnormality detection method with the data format of the control signal, and generates the control signal model by inputting the training data into the abnormality detection method.

4. The abnormality detection device according to claim 3, wherein: The reliability calculation unit generates the verification data according to the time series data of the control signal based on the correspondence information, the abnormality detection technique information, the control signal model, the time series data of the control signal and the specific time range of the abnormality detection result, and obtains the degree of abnormality of the control signal by inputting the verification data into the control signal model.

5. The abnormality detection device according to claim 3 or 4, wherein: In the case where a group of control signals and one sensor signal are associated with each other in the correspondence information and a combination method of the group of control signals is set, the model generation unit applies an abnormality detection method corresponding to the sensor signal to each group of control signals to generate the control signal model. The reliability calculation section calculates the reliability of the group of control signals using a combination method of the group of control signals.

6. The abnormality detection device according to any one of claims 2 to 5, wherein: The abnormality detection device includes a pattern segmentation unit that segments the time series data of the control signal for each control state based on event signal information, wherein the event signal information stores log data of the event signal indicating the control state or alarm of the device, The model generation unit generates the control signal model based on time series data of the control signal for each mode.

7. An abnormality detection method, used in an abnormality detection device, the abnormality detection device having a sensor, detecting an abnormality in a device controlled by a control signal, wherein: The computer inputs the time series data of the control signal within a specific time range of the abnormal detection result detected in the sensor signal obtained from the sensor as verification data into the control signal model, thereby obtaining the abnormality degree of the control signal within the specific time range, and calculating the reliability of the control signal within the specific time range. The control signal model is a model that outputs the abnormality degree of each control signal based on the time series data of the control signal, which is generated based on training data. The training data is a feature amount of the time series data of the control signal in the past. The computer outputs information obtained by adding the reliability of the control signal within the specific time range to the abnormality detection result within the specific time range as the abnormality diagnosis result.

8. An abnormality detection program for an abnormality detection device, the abnormality detection device having a sensor for detecting an abnormality in a device controlled by a control signal, wherein: The anomaly detection program causes the computer to execute: Reliability calculation processing, inputting the time series data of the control signal within a specific time range of the abnormality detection result detected in the sensor signal obtained from the sensor as verification data into the control signal model, thereby obtaining the abnormality degree of the control signal within the specific time range, and calculating the reliability of the control signal within the specific time range, the control signal model is a model that outputs the abnormality degree of each control signal based on the time series data of the control signal generated based on training data, and the training data is a feature amount of the time series data of the control signal in the past; as well as The abnormality diagnosis process outputs information obtained by adding the reliability of the control signal within the specific time range to the abnormality detection result within the specific time range as the abnormality diagnosis result.

Images