JP7246330B2 - How to monitor monitored data - Google Patents

Description

The present invention relates to a method for monitoring monitored data.

In various work sites such as manufacturing industries, stores, and warehouses, the Internet of Things (IoT) is progressing, and various data are collected and accumulated from various sensors installed in devices, people, and the like. A technology is known in which state monitoring is performed by learning data obtained in advance during stable operation and comparing the data during the monitoring period with the learned data. Also, in recent years, methods of detecting anomalies using autoencoders have been proposed in the fields of images and communications. In the anomaly detection using an autoencoder, the autoencoder is generated by learning the data during stable operation, and the output data obtained by inputting the monitoring time period data and the monitoring time period data into the generated autoencoder. Detect anomalies based on the difference between

In Patent Literature 1, the recovery error of the autoencoder (for example, the mean square error between input data x and output data y) is calculated as the degree of anomaly, and an anomaly is detected when it exceeds a preset threshold.

JP 2019-49778 A

In Patent Document 1, one degree of abnormality is calculated for input data x and output data y composed of a plurality of elements, and the amplitude of each element of the input data is not taken into consideration. For this reason, when the fluctuation width of the value in the normal state differs for each element, it is difficult to identify the state corresponding to each fluctuation width.

The present invention has been made in view of such a background, and an object of one aspect thereof is to provide a data group composed of a plurality of elements, even if the fluctuation range of values in normal times differs between elements, individual To provide a technology that can be evaluated corresponding to the amplitude.

One aspect of the present invention is a method for monitoring monitoring target data by a computer system, wherein the computer system stores an analysis model, and the analysis model is used for the input data composed of a plurality of elements. The computer system outputs restored data having the same number of elements as the data, and learns so that the difference between the input normal data and the restored data with respect to the normal data is small. generating a difference distribution for each element between the normal data for generation and the restored data obtained by inputting the plurality of normal data for generating difference distribution into the analysis model, and analyzing the monitoring target data and the monitoring target data; A difference for each element from restored data obtained by inputting to the model is compared with the distribution of the difference for each element, and an index value for each element is calculated based on the comparison.

According to one aspect of the present invention configured as described above, input data composed of a plurality of elements can be evaluated in consideration of the amplitude of each element.

1 is a schematic configuration diagram of a system including a state monitoring system; FIG. It is a figure which shows an example of a change of input data. It is a figure which shows an example of the information processing apparatus which can be used as a component of a condition-monitoring system. 3 is a diagram showing an overview of functions and data of a data storage device; FIG. It is an example of measurement data as a data group. It is a figure which shows the function of a data-analysis apparatus, and the outline|summary of data. It is an example of post-preprocessing data. It is an example of difference distribution data. It is an example of analysis result data. FIG. 10 is a sequence diagram showing processing during learning; 4 is a flowchart showing learning processing; FIG. 4 is a diagram showing an outline of a method of calculating a difference for each element of data before and after auto-encoding; FIG. 10 is a sequence diagram showing analysis processing during state monitoring; 4 is a flowchart showing analysis processing; FIG. 10 is a diagram showing the relationship between the difference for each element, the standardized value of the difference, and the probability of occurrence of an abnormality; It is a figure which shows an example of a parameter setting screen. It is a figure which shows an example of an analysis result display screen. 1 is a schematic configuration diagram of a state monitoring system that performs analysis processing on an edge side; FIG. FIG. 4 is a sequence diagram showing real-time analysis processing; It is a figure which shows the outline of two-dimensional arraying of input data. It is a figure which shows the input data arranged two-dimensionally. FIG. 4 is a diagram showing an outline of a method of calculating a difference between before and after auto-encoding two-dimensional array data; It is a figure which shows an example of two-dimensional arraying of input data. FIG. 10 is a diagram showing another example of two-dimensional arraying of input data; FIG. 10 is a diagram showing another example of two-dimensional arraying of input data; FIG. 10 is a diagram showing another example of two-dimensional arraying of input data; It is a figure which shows the example which divides|segments input data into two or more area|regions. FIG. 10 is a diagram showing an example of display of analysis results; FIG. 10 is a diagram showing another example of display of analysis results; FIG. 10 is a diagram showing an example of display of a determination result of an abnormality type; FIG. 10 is a diagram showing an example of a display of state identification results; FIG. 10 is a diagram showing an example of a display of state identification results; FIG. 10 is a diagram showing an example of a display of state identification results; It is a figure which shows an example of a structure of an autoencoder. FIG. 10 is a diagram showing an example of comparison of difference distributions before and after autoencoding for one or more autoencoders;

BEST MODE FOR CARRYING OUT THE INVENTION An embodiment of the present invention will be described below based on the drawings. The state monitoring method or state monitoring system according to the present embodiment provides input data composed of two or more elements. Perform corresponding condition monitoring. The state monitoring technology according to the present embodiment can use IoT (Internet of Things) in manufacturing, warehouses, stores, and the like, for example.

A first embodiment will be described with reference to FIGS. 1 to 17. FIG. In this embodiment, for example, data collected from various sensors installed in factory equipment is analyzed, and normal data collected during stable operation (that is, normal time) is compared with data collected during the monitoring period to obtain An example of detecting that the device to be monitored is in a state different from the learned normal state will be described. In this embodiment, a state different from the normal state is detected as an anomaly (that is, a failure or failure and its sign).

FIG. 1 shows an example of a system including a condition monitoring system 1 in this embodiment. The condition monitoring system 1 includes a data analysis device 102, a data storage device 101, and an input/output device 104, for example. A sensor 103 is installed on or around the analysis target 105 . Data storage device 101 , data analysis device 102 , input/output device 104 and sensor 103 are connected via communication network 106 . Thereby, the data analysis system 1 collects and analyzes measurement data from the sensor 103 .

The communication network 106 is composed of a wired or wireless communication infrastructure such as a LAN (Local Area Network), a WAN (Wide Area Network), the Internet, an intranet, a dedicated line, a mobile phone network, an optical fiber, or the like.

Note that the input/output device 104 may be a device connected to each device (that is, the sensor 103, the data storage device 101, and the data analysis device 102) via the network 106 as shown in FIG. , devices that are directly connected to each device (that is, the sensor 103, the data storage device 101, and the data analysis device 102).

The analysis target 105 is, for example, a machine tool or manufacturing device (for example, a press machine, an NC machine tool, a robot arm, a molding machine, a vulcanizing machine, or a three-dimensional printer) installed in a product manufacturing factory. . The analysis target 105 is not limited to machine tools or manufacturing equipment, and may be, for example, trains, equipment related to trains, lighting devices, internal combustion engines, elevators, escalators, or air conditioners. The analysis target 105 does not necessarily have to be connected to the communication network 106 .

The sensor 103 may be provided on the analysis target 105 or may be provided in the vicinity of the analysis target 105, for example. The state monitoring system 1 can also use the sensor 103 provided in another device (not shown) that works in conjunction with the analysis target 105 .

The type of sensor 103 does not matter. The sensor 103 includes, for example, a vibration sensor, a speed sensor, an acceleration sensor, a current sensor, a voltage sensor, a temperature sensor, a humidity sensor, a pressure sensor, a load sensor, a torque sensor, a strain sensor, a position sensor, an angle sensor, a gyro sensor, a sound sensor, An optical sensor, a gas sensor, a distance sensor, a color sensor, or the like.

The data storage device 101 acquires data output from the sensor 103 via the communication network 106, for example. Data acquired by the data storage device 101 (that is, sensor data) is hereinafter referred to as acquired data. This acquired data is an example of a "data group" or "data series" and can also be referred to as measurement data. However, the acquired data is not limited to time-series data. The data storage device 101 stores acquired data and transmits requested data to the data analysis device 102 in response to a request from the data analysis device 102 .

The data analysis device 102 performs learning processing and analysis processing, which will be described later, on sensor data acquired from the sensor 103 to monitor the state of the device and detect an abnormality. The input/output device 104 provides a user interface for users to access the data storage device 101 and the data analysis device 102 . Via the input/output device 104 , the user can input data, make various settings, and perform analysis processing with respect to the data storage device 101 and the data analysis device 102 . A user can refer to or acquire information output from the data storage device 101 and the data analysis device 102 via the input/output device 104 .

Input/output to/from the data storage device 101 and the data analysis device 102 may be performed via their respective input devices and output devices. The data storage device 101, the data analysis device 102, and the input/output device 104 are all information processing devices. These may be independent hardware, or two or more of them may be configured with common hardware.

All or part of each of these devices may be configured using virtual information processing resources such as a cloud server in a cloud system. Moreover, the owners of the sensor 103, the data storage device 101, and the data analysis device 102 may be different, or two or more of them may be owned by the same person. For example, the owner of the analysis object 105 owns the sensor 103, the owner of the analysis technology owns the data analysis device 102, and the owner of the data storage device 101 acts as an intermediary between the owner of the sensor and the owner of the analysis technology. , may collect and manage data.

Further, there may be two or more analysis targets 105 and sensors 103, and each sensor 103 may be owned by a different person. Moreover, the state monitoring system 1 may include two or more data storage devices 101 and data analysis devices 102 to perform distributed processing and distributed management of data.

FIG. 2 is an image diagram showing the relationship between each element of analysis target data and measured values. For example, in the case of performing analysis using spectral density on measurement data measured in time series by a vibration sensor, an acceleration sensor, or a sound sensor, for example, a fast Fourier transform is used for each predetermined unit time. to calculate the spectral density.

A graph 201 is a graph showing frequency and spectral density, with element IDs corresponding to frequencies and measured values corresponding to spectral densities. Graph 201 shows upper limit 211 , lower limit 212 , normal range 213 and average 214 . In normal times, similar spectral density distributions are obtained for each unit time, but there are cases where the normal range (ie, amplitude) 213 of the spectral density differs depending on the frequency. Factors that cause the amplitude to vary include the characteristics of the device itself, and external factors (for example, temperature, humidity, operating conditions of nearby devices, or other factors).

FIG. 3 is a hardware configuration example of an information processing device that can be used as each device of the data analysis device 102 , the data storage device 101 , and the input/output device 104 . As shown in FIG. 3, the information processing device 300 includes a processor 301, a main memory device 302, an auxiliary memory device 303, an input device 304, an output device 305, and a communication device 306, for example. These are communicably connected to each other via communication means such as a bus (not shown).

The processor 301 is configured using, for example, a CPU (Central Processing Unit) or an MPU (Micro Processing Unit). Various functions of the data analysis device 102, the data storage device 101, and the input/output device 104 are realized by the processor 301 reading and executing the computer programs stored in the main storage device 302. FIG.

The main memory device 302 is a device that stores computer programs and data, and is, for example, a ROM (Read Only Memory), a RAM (Random Access Memory), a nonvolatile semiconductor memory, or the like.

The auxiliary storage device 303 includes, for example, a hard disk drive, an SSD (Solid State Drive), an optical storage medium (that is, a CD (Compact Disc), a DVD (Digital Versatile Disc), etc.), a storage system, and an IC card (Integrated Circuit Card). ), a read/write device for recording media such as an SD memory card, and a storage area of a cloud server. Computer programs and data stored in the auxiliary storage device 303 are read into the main storage device 302 as needed.

The input device 304 is, for example, a keyboard, mouse, touch panel, card reader, voice input device, or the like. The output device 305 is a user interface that provides various information such as processing progress and processing results to the user. The output device 305 is, for example, a screen display device (ie, liquid crystal monitor, LCD (Liquid Crystal Display), graphic card, etc.), an audio output device (ie, speaker, etc.), or a printing device. For example, the information processing device 300 may input/output information to/from another device via the communication device 306 .

The communication device 306 is a wired or wireless communication interface that realizes communication with another device via a communication means such as a LAN or the Internet. The communication device 306 is, for example, a NIC (Network Interface Card), a wireless communication module, a USB (Universal Serial Bus) module, or a serial communication module.

Various functions provided by the data storage device 101, the data analysis device 102, and the input/output device 104 are, for example, a computer program read by the processor 301 into the main storage device 302 (from the auxiliary storage device 303 to the main storage device). 302 includes a computer program). The main storage device 302, auxiliary storage device 303, and combinations thereof are storage devices each including a non-transitory storage medium. Each function of the data analysis device 102, the data storage device 101, and the input/output device 104 can be implemented in one or a plurality of information processing devices 300, and one information processing device 300 may include the functions of different devices. good. Thus, the information monitoring system can be configured with a computer system including one or more processors and one or more storage devices.

FIG. 4 shows main functions of the data storage device 101 and main data stored in the data storage device 101 . The data storage device 101 has a data recording function 401 and a data reading function 402 and records measurement data 403 . A data recording function 401 records the data received from the sensor 103 into the measurement data 403 . Data reading function 402 reads data specified by data analysis device 102 from measurement data 403 and transmits the data to data analysis device 102 .

An example of the measurement data 403 is shown in FIG. Each data recorded by the data storage device 101 and the data analysis device 102 is recorded, for example, in a table format. The measurement data 403 stores, for example, a record having items of a sensor identifier 501 , measurement date and time 502 , and a measurement value 503 . An identifier of the sensor 103 is recorded in the sensor identifier 501 . The sensor identifier 501 is information for identifying the sensor 103 that performed the measurement. The sensor identifier 501 may be, for example, the name of the building in which the sensor 103 is installed, the floor, the room number, the device name, the sensor type, the serial number, or the like, or an identifier generated by combining them.

The measurement date and time 502 records the date and time when the data was measured. The date and time may be recorded in hours, minutes, seconds, or fractions of a second, as appropriate. A value measured by the sensor 103 is recorded in the measured value 503 . Note that this example is just an example of measurement data, and the operating angle of the device, the elapsed time and distance from the base point, and the like may also be recorded.

FIG. 6 shows main functions provided by the data analysis device 102 and main data stored by the data analysis device 102 . The data analysis device 102 has, for example, a parameter setting function 601, a data preprocessing function 602, a model generation function 603, a model calculation function 604, a distribution calculation function 605, an abnormality probability calculation function 606, and a result display function 607. The data analysis device 102 also stores post-preprocessing data 608 , difference distribution data 609 , analysis result data 610 , and an analysis model 611 .

A parameter setting function 601 displays a screen for setting parameters necessary for learning and monitoring, and reads input information. The data preprocessing function 602 performs preprocessing on the measurement data received from the data storage device 101 and records the preprocessed data 608 . Preprocessing is, for example, calculation of average value, median value, mode value, effective value, maximum value, or minimum value per predetermined unit time, calculation of spectral density, etc. Further, by combining these, For example, spectral density may be calculated for each unit time, and an average value may be calculated for each frequency using a plurality of spectral density data. Also, smoothing may be performed using a moving average value or the like.

A model generation function 603 generates an autoencoder that receives preprocessed data as an input and records it in an analysis model 611 . An autoencoder generates intermediate data with the number of elements reduced from input data, and generates and outputs restored data with the same number of elements as the input data from the intermediate data. Note that an analysis model that generates restored data with the same number of elements as the input data, which is of a type different from the autoencoder, may be used.

The model computation function 604 inputs the preprocessed data to the autoencoder generated by the model generation function 603 and outputs the autoencoded data. A distribution calculation function 605 generates a difference distribution before and after auto-encoding for each element based on comparison between the learning data and the data after auto-encoding output by the model calculation function 604, and records it in the difference distribution data 609. do.

The abnormality probability calculation function 606 calculates the difference between before and after autoencoding for each element based on comparison between the preprocessed monitoring target data and the output data obtained by inputting the monitoring target data to the autoencoder, By comparing with the distribution of the difference before and after the auto-encoding described above, the probability of abnormality occurrence is calculated and recorded in the analysis result data 610 . A result display function 607 displays the calculated abnormality occurrence probability on the input/output device 104 or the output device 305 .

FIG. 7 shows an example of post-preprocessing data 608 . The preprocessed data 608 includes a sensor identifier 701 , a measurement date and time 702 , and a preprocessed value 703 . The values after preprocessing are recorded corresponding to element IDs from 1 to n. The element ID corresponds to, for example, the frequency, the operating angle of the device, or the elapsed time or distance from the base point. The preprocessed data value 703 (Di_1, Di_2, . . . Di_n) corresponds to one entry of the input data. For example, in the case of calculating the spectral density, the spectral density corresponding to each frequency is recorded for each predetermined unit time.

The measurement date and time 702 is, for example, the first time of the unit time. In addition, when measuring the distance from the reference point, the elapsed time, or the operating angle of the rotating equipment, etc., are recorded together with the measured values, the distance, elapsed time, or operating angle, etc., will be recorded for each operation of the equipment. Measured data corresponding to predetermined values of the axis of may be recorded in 703 . In this case, for example, the operation start date and time is recorded as the measurement date and time. The data preprocessing function 602 extracts data for each operation and calculates measurement data corresponding to a predetermined value.

FIG. 8 shows an example of the difference distribution data 609. As shown in FIG. Difference distribution data 609 includes sensor identifier 801 , standard deviation 802 and 95% value 803 . A standard deviation 802 and a 95% value 803 are recorded for each element of the input data. Besides the 95% value 803, an average value, a 90% value, etc. may be recorded. One entry is recorded for each analysis model. FIG. 8 shows, for example, an example in which one analysis model is generated for each sensor, and one entry is recorded for each sensor.

FIG. 9 shows an example of analysis result data 610. As shown in FIG. Analysis result data 610 includes sensor identifier 901 , date and time 902 , abnormality probability 903 , and difference standardization 904 .

FIG. 10 shows an example of a learning sequence. Communication between the sensor 103 and the data storage device 101 and communication between the data storage device 101 and the data analysis device 102 are performed via the communication network 106, but the communication network 106 is omitted in FIG. 10 for simplicity.

The sensor 103 measures the analysis target 105 at preset intervals (S1001), and transmits the measurement results (including, for example, the sensor identifier, the date and time of measurement, and the measured value) to the data storage device 101 (S1002). The data recording function 401 of the data storage device 101 records the data received from the sensor 103 as part of the measurement data 403 (S1003). Measurement by the sensor 103 and recording of measurement data by the data recording function 401 are performed by the sensor 103 and the data recording function 401 as needed.

On the other hand, the parameter setting function 601 of the data analysis device 102 displays the parameter setting screen on the output device (S1004). A user who uses the condition monitoring system 1 inputs information to be used for learning from the input device 304 to the parameter setting screen. The parameter setting function 601 accepts a learning instruction input from the input device 304 (S1005), and requests data of the instructed learning period from the data storage device 101 (S1006).

When the data request is received, the data reading function 402 of the data storage device 101 extracts the relevant data from the measurement data 403 (S1007) and transmits it to the data analysis device 102 (S1008). Upon receiving the learning data from the data storage device 101, the data analysis device 102 executes learning processing (S1009).

FIG. 11 shows an example of the flow of the learning process S1009. The data preprocessing function 602 performs predetermined data preprocessing on the learning data received from the data storage device 101, and records it in preprocessed data 608 (S1102). Here, the predetermined data preprocessing is, for example, data smoothing, representative value calculation, Fourier expansion, or the like.

The model generation function 603 divides the preprocessed learning data into model generation data and difference distribution generation data (S1103). For this division, for example, data for model generation is randomly extracted, and the remainder is used as difference distribution generation data. The difference distribution generation data may be randomly extracted, and the rest may be model generation data. Also, although the division ratio is assumed to be 1:1, it does not have to be 1:1 as long as the autoencoder can learn and generate a difference distribution sufficiently.

Next, the model generation function 603 generates an autoencoder by learning model generation data, and records the generated autoencoder in the analysis model 611 (S1104). The model calculation function 604 inputs the difference distribution generation data to the autoencoder recorded in the analysis model 611 to obtain output data (S1105).

The distribution calculation function 605 calculates the difference distribution for each element of the preprocessed data by the difference distribution calculation process described later (S1106), and records the standard deviation and 95% value in the difference distribution data 609 (S1107). . Note that the 95% value is just one example, and for example, the 90% value and the 80% value may be used, and these values may be written together.

FIG. 12 is a diagram showing an image of the difference distribution calculation processing S1106. The distribution calculation function 605 calculates the absolute value of the difference between each data 1201 for differential distribution calculation and the corresponding auto-encoded output data 1202 for each element (1203). Here, let “Diff(i)m” be the absolute value of the difference of the element i between the m-th difference distribution generation data and the corresponding output data.

Distribution calculation function 605 collects "Diff(i)m" for all m for each i, calculates standard deviation (σi) and 95% value (Vi), and records in difference distribution data 609 (1204). Note that the 95% value is just one example, and for example, a 90% value or an 85% value may be used, or a plurality of values may be calculated and recorded. Although the absolute value is used as the difference for each element in this example, for example, the square value of the difference may be used instead of the absolute value.

An analysis sequence will be described with reference to FIG. FIG. 13 shows an example in which analysis is performed periodically (that is, at predetermined time intervals). The sensor 103 performs measurements at predetermined intervals (S1301) and transmits measurement data to the data storage device 101 (S1302). The data recording function 401 of the data storage device 101 records the received measurement data in the measurement data 403 (S1303). This processing is the same as the measurement during learning (S1001) and the measurement data recording (S1003), and is executed as needed.

The abnormality probability calculation function 606 of the data analysis device 102 periodically (that is, at predetermined time intervals) starts analysis (S1304), and requests data of the analysis target period from the data storage device 101 (S1305). The analysis target period is, for example, from the previous analysis date and time to the current date and time.

The data reading function 402 of the data storage device 101 extracts the relevant data from the measurement data 403 according to the received data request (S1306) and transmits it to the data analysis device 102 (S1307). The abnormality probability calculation function 606, upon receiving the data of the analysis target period from the data storage device 101, executes the analysis processing described later (S1308), and displays the analysis result on the input/output device 104 or the output device 305 (S1309). .

FIG. 14 shows an example of the flow of analysis processing S1308. The data preprocessing function 602 performs predetermined preprocessing on the acquired data (S1402). Here, the predetermined preprocessing is the same processing as the preprocessing in the processing flow during learning shown in FIG. 11 . The model calculation function 604 inputs the preprocessed data to the autoencoder recorded in the analysis model 611 to obtain output data (S1403).

The abnormality probability calculation function 606 compares the preprocessed data with the auto-encoded output data, and calculates a difference Diff(i) for an arbitrary element i (0<i<=n) ( S1404). Next, the abnormality probability calculation function 606 standardizes the difference Diff(i) calculated for each element using the following formula 1 (S1405).
Ds_i=max(Diff(i)-Vi,0)/σi/B*100 (Equation 1)
Here, σi and Vi are the standard deviation and 95% value of the difference recorded in the difference distribution data 609, respectively. Also, B is a threshold value set when parameters are set.

Next, the abnormality probability calculation function 606 calculates the abnormality probability O_i by the following Equation 2 (S1406).
O_i=min(DS_i, 100) (Equation 2)

Next, the abnormality probability calculation function 606 records the analysis result (that is, the standardized value of the difference (Ds_i) and the abnormality occurrence probability (O_i)) in the analysis result data 610 (S1407). Then, the abnormality probability and the alert threshold are compared (S1408), and if the abnormality probability is greater than a predetermined alert threshold (YES in S1408), an alert is issued (S1409).

Although the analysis sequence shown in FIG. 13 shows an example in which analysis processing is periodically performed at predetermined intervals, the user of the state monitoring system 1 may explicitly instruct the start of analysis. The data analysis device starts analysis upon receiving the analysis start instruction.

FIG. 15 shows the relationship between the difference Diff(i), the standardized difference (Ds_i), and the abnormality probability (O_i). If the difference Diff(i) is less than the 95% value (Vi) of the normal time, it is considered to be within the normal range and the probability of abnormality is 0%. Then, when the difference Diff(i) rises above the 95% value (Vi) of the normal time, the abnormality probability also rises linearly. Also, when the difference Diff(i) is equal to or greater than the value obtained by adding B times the standard deviation (σi) to the normal 95% value (Vi) (that is, the difference Diff(i)>=Vi+σi*B), the probability of abnormality Consider (O_i) to be 100%. Note that the standardized difference (Ds_i) increases linearly even in the region of Diff(i)>=Vi+σi*B.

FIG. 16 is an example of a parameter setting screen displayed by the parameter setting function. The parameter setting screen 1600 displays learning data 1601, data preprocessing 1602, abnormality determination parameters 1603, an OK button 1604, a cancel button 1605, and the like. A user of the condition monitoring system 1 uses the input device 304 to input various parameters.

The user designates a sensor 1606 to be analyzed and a learning target period 1607 in the learning data 1601 . In data preprocessing 1602, the user designates a data preprocessing method 1608 and various parameters required for preprocessing (unit time 1609 in the example of FIG. 16). After data preprocessing 1602 reads the input to preprocessing method 1608, it may optionally display one or more accompanying parameters 1609 (units of time in the example of FIG. 16). Also, the pretreatment may not be performed.

The user inputs a normal range 1610 , a threshold 1611 , and an alert threshold 1612 in the abnormality determination parameters 1603 . In the normal range 1610, the user inputs a difference reference value used when generating the difference distribution and when calculating the abnormality probability. Although the normal range is set to 95% in this embodiment, the normal range may be any other value. For example, the user may specify a 90% value, an 85% value, an average value, etc., and the difference distribution calculation process may calculate the corresponding values and record them in the difference distribution data 609 . A threshold 1611 is the threshold B used in Equation 1 when calculating the probability of abnormality. The alert threshold 1612 is a threshold for determining whether an alert is required in analysis processing.

The parameter setting screen may display more items than the example shown in FIG. 16, or may display fewer items. Default values may be set for the data preprocessing 1602 and the abnormality determination parameter 1603 .

FIG. 17 shows an example of an analysis result display screen displayed by the result display function 607. As shown in FIG. The analysis result display screen 1700 displays parameters 1701, a decision button 1706, a cancel button 1707, and an abnormality probability 1708, for example.

The parameters 1701 display the sensor 1702 used for analysis, the normal range 1703, the threshold 1704, and the display period 1705 of the analysis result on the screen. The abnormality probability 1708 displays the abnormality probability corresponding to the combination of the measurement date and time and the element ID by associating the abnormality probability with the color. The user can know the abnormality probability from the difference in hue and shade (that is, the difference in lightness). The element ID is, for example, frequency, time or distance from the base point, or operating angle of the device. Also, when the probability of abnormality is high, an explicit warning may be displayed.

Also, the parameters shown in the parameter 1701 may be resettable. In cases where resetting is possible, when these parameters are entered and the decision button 1706 is pressed, the distribution calculation function 605 recalculates the values (Vi) corresponding to the entered normal ranges as necessary. Further, the abnormality probability calculation function 606 also recalculates Equations 1 and 2. The result display function 607 updates the abnormality probability 1708 using the newly calculated abnormality probability. This makes it possible to adjust the sensitivity of anomaly detection.

The analysis result display screen may display more items than the example shown in FIG. 17, or may display fewer items.

In this embodiment, data measured by the sensor 103 is stored in the data storage device 101 and then read out from the data storage device 101 by the data analysis device 102 at the start of learning processing or analysis processing. Alternatively, for example, data measured by sensor 103 may be stored in an external storage medium (not shown). The data analysis device 102 can also read measurement data from an external storage medium and perform learning and analysis. Measured data may also be transmitted directly from the sensor 103 to the data analysis device 102 .

Also, the monitoring system does not necessarily have the data preprocessing function 602 , and preprocessed data may be input to the data analysis device 102 . For example, data that requires complicated preprocessing may be preprocessed in a computer (not shown) before being input to the state monitoring system 1 so that the state monitoring system 1 can be used.

Also, in this embodiment, table notation is used as an example of indicating each data, but this does not limit the recording method of each data to the table type, and data can be recorded in various ways such as list or chain. may be used. Also, the recorded elements may be represented in various forms, such as numbers, symbols, or mathematical expressions.

Also, in this embodiment, each data is collected by a single sensor, but for example, each element of the input data may be obtained by two or more sensors. may correspond to By learning and analyzing input data composed of elements arranged in a predetermined order, it is possible to detect anomalies according to the amplitude of measurement values for each sensor.

Also, the normalization of the difference 904 may be visualized and displayed. It is also possible to detect an element with a particularly large Ds_i from among the element IDs whose abnormality probability exceeds 100%, and use it to identify the failure location. Further, if the abnormality probability 903 or difference standardization 904 of the element and the abnormality type or failure location are linked and recorded, after identifying the element with the abnormality probability 903 or the difference standardization 904 high, the abnormality type or failure can be estimated. Also, the threshold value B shown in Equation 1 may be set for each element.

According to this embodiment configured in this way, even in cases where the amplitudes of the individual elements of the input data differ, the condition monitoring system 1 can detect abnormalities corresponding to the amplitudes of the individual elements. state monitoring can be performed.

In addition, in this method, the learning data is divided, one of the learning data is used to generate an autoencoder, and the other learning data is used to calculate the distribution of the difference for each element of the input/output data of the autoencoder. Abnormality determination of monitoring target data is performed using the distribution. For this reason, the restoration accuracy required for the autoencoder is not necessarily high, and an analysis model can be generated even when the number of learning data is small.

In this embodiment, an example will be shown in which data analysis devices are arranged on the edge and the cloud, respectively, and analysis is performed by the data analysis device on the edge side. FIG. 18 is a diagram showing an outline of the system configuration in this embodiment. An edge-side data analysis device 1801 is arranged on the edge, and a cloud-side data analysis device 1802 is arranged on the cloud. The edge side data analysis device 1801 and the cloud side data analysis device 1802 can communicate via the network 106 .

The edge-side data analysis device 1801 has some or all of the functions and data provided by the data storage device 101 of the state monitoring system 1 and the functions and data provided by the data analysis device 102 . The sensor 103 and the edge-side data analysis device 1801 are installed in close physical or network areas, such as in the same building or in the same company's network, and are connected via a network 1803 . Note that the learning sequence is the same as in Example 1 and is performed before analysis.

FIG. 19 shows an analysis sequence in this configuration. The cloud-side data analysis device 1802 transmits the learned analysis model 611 (that is, autoencoder) and difference distribution data 609 to the edge-side data analysis device 1801 (S1901). The edge-side data analysis device 1801 records the received autoencoder and difference distribution data in the analysis model 611 and the difference distribution data 609, respectively (S1902).

The sensor 103 measures data at predetermined intervals (S1903), and transmits the measured data to the edge side data analysis device 1801 (S1904). The data recording function 401 of the edge side data analysis device 1801 records the received measurement data in the measurement data 403 (S1905). Next, the abnormality probability calculation function 606 of the edge side data analysis device 1801 executes analysis processing similar to that of the first embodiment (S1906), and the result display function 607 of the edge side data analysis device 1801 displays the result (S1907). .

According to this embodiment, real-time analysis by the edge-side data analysis device 1801 becomes possible, and the time from the occurrence of an abnormality to detection can be reduced. Also, by providing the data analysis device 102 on the edge side, it becomes possible to analyze the monitored data without sending it to the outside. It is also possible to reduce the amount of data flowing into the network 106 . In addition, since the cloud-side data analysis device 1802 generates autoencoders, the edge-side data analysis device 1801 can perform real-time analysis on the edge side without having the high computing power required to execute Deep Learning.

Note that this state monitoring method may be operated on-premises. In the on-premise type operation, the data analysis device 1801 on the edge side performs learning processing (that is, analysis model generation and difference distribution calculation). This makes it possible to perform learning and analysis processing without exposing measurement data, including data for learning.

In Example 1, an analysis example using an autoencoder was shown for a case in which input data after preprocessing is represented by a one-dimensional array with a fixed number of elements. In this embodiment, an analysis example that can deal with various input data will be shown.

For example, each element of the input data is represented by a (Xi, Yi) pair, where Xi is the frequency, the rotation angle of the device, or the distance or time from the base point, and Yi is the measured value linked to Xi. , the same Xi does not necessarily exist in all the input data, and the number of input data in all the input data does not necessarily become the same. Also, the same Xi may exist in one piece of input data.

Also, for example, when the area of Xi is wide, the representative value is calculated for each unit interval to reduce the number of elements, and this is used as input data, the fluctuation range of Yi is large, the average value, the maximum value, Alternatively, there is a possibility that it cannot be expressed by a single value such as the minimum value. Therefore, in this embodiment, by converting the input data into two-dimensional array data, an analysis example that can deal with input data in various cases will be shown.

A method of converting input data into two-dimensional array data and an analysis method when applying this method to two-dimensional array data are described below. Note that the description of this embodiment will focus on the points that are different from those of the first embodiment.

The data analysis device 102 has a two-dimensional arraying function (not shown). After the preprocessing S1102 shown in FIG. 11 and the preprocessing S1402 shown in FIG. 14, the data analysis device 102 performs two-dimensional array processing of the preprocessed data.

The two-dimensional array processing of input data will be described with reference to FIG. Focusing on the measurement date and time T1 of the preprocessed data shown in FIG. 7, an example of two-dimensionally arraying the values (D1_1, D1_2, . Element IDs 1 to n correspond to predetermined numerical values F1 to Fn, and values after preprocessing are recorded corresponding to numerical values F1 to Fn.

A graph 2001 is a graph of data on the measurement date and time T1. An image 2002 is obtained by mapping the graph 2001 represented by the range of F1 to Fn on the horizontal axis and the range of y_min to y_max on the vertical axis to an image of size n×Y_high. Each element (i, k) of the image 2002 is represented by Equations 3 and 4 below for arbitrary i (0<i<=n) and arbitrary k (0<k<=Y_hign). In the formula below, 0 indicates black in the image and 1 indicates white in the image.
(i, k)=1 (k=yi) (Formula 3)
(i, k)=0 (k≠yi) (Formula 4)

Here, yi is defined by Equation 5 below.
yi = int {(D1_i−y_min)/(y_max−y_min)
*Y_high} (Formula 5)
However, when yi<1, yi=1, and when yi>Y_high, yi=Y_high.

Here, n is the number of elements of the input data and the size of the first dimension of the two-dimensional array data. Y_high is the size of the second dimension of the two-dimensional array data, and is set, for example, when setting parameters. Also, int is a process of converting a real number into an integer by a predetermined means such as rounding up, rounding down, or rounding off after the decimal point. Each element (i, k) obtained by Equations 3 to 5 above becomes each element of the two-dimensional array data.

FIG. 21 shows an image of two-dimensional array data. The two-dimensional array data are recorded in the data analysis device 102 . In addition, for an autoencoder that inputs and outputs two-dimensional array data, each element of the output data from the autoencoder is output as 0 to 1 by using, for example, a sigmoid function as the activation function of the output layer. can be done.

FIG. 22 is a diagram showing an image of difference distribution calculation processing S1106 for two-dimensional array data in the learning processing flow. The distribution calculation function 605 calculates the absolute value 2203 of the difference between each data 2201 for difference distribution calculation and the corresponding auto-encoded output data 2202 for each element (that is, for each square of the two-dimensional array data 2101 and 2102). calculate.

Here, "Diff(i, k)m" is the absolute value of the difference in element (i, k) between the m-th difference distribution generation data and the corresponding output data. Next, the distribution calculation function 605 calculates the total value “Diff(i)m” 2204 for each column of differences for arbitrary i and m using Equation (6).
Diff(i)m=Σk ₌₁ ^Y_high Diff(i,k)m (Formula 6)

Next, the distribution calculation function 605 obtains the distribution 2205 of "Diff(i)m" for each i, calculates the standard deviation (σi) and the 95% value (Vi), and records them in the differential distribution data 609. . Note that the 95% value is just one example, and for example, a 90% value or an 85% value may be used, or a plurality of values may be calculated and recorded.

In the process S1404 in the analysis process flow, the two-dimensional array data to be analyzed is compared with the output data after autoencoding, and for any i (0<i<=n) (that is, the column of the two-dimensional array ), the total value Diff(i) of the differences is calculated using the same method as during learning, that is, using Equation (6). The normalization of the difference and the abnormality probability are calculated by Equations 1 and 2 as in the first embodiment. As described above, input data can be converted into two-dimensional array data and analyzed.

Among factory equipment, there are many equipment such as a press that repeats the same operation, and similar waveforms often appear repeatedly in measurement data. As the horizontal axis of the graph 2001, for example, it is possible to use the angle of the axis of the rotating device, or it is also possible to use time or distance from the base point. The two-dimensional arraying of input data will be further described with an example.

Figures 23A-23D show two-dimensional array examples 2301-2304, respectively. Images 2301 to 2304 are visualizations of two-dimensional array data. As shown in the image 2301 of FIG. 23A, the two-dimensional array data may have one element of 1 per column (that is, one cell in the corresponding column of the image may be white).

As shown in images 2302-2304 of FIGS. 23B-23D, multiple elements per column may be 1 (ie, multiple cells are white in corresponding columns of the image). Also, there may be columns with all elements zero (ie, the corresponding column of the image is all black).

(A)
An image 2301 in FIG. 23A is an image converted into two-dimensional array data according to Equations 3-5. This is an example in which input data is mapped to a two-dimensional array so that one element in each column is 1 (that is, white in the image).

(B)
Two-dimensional array data does not necessarily have one element of 1 for each column. For example, a two-dimensional array can be obtained by changing Equations 3 and 4 to Equations 7 and 8 below.
(i, k)=1 (k<=yi) (Formula 7)
(i, k)=0 (k>yi) (Formula 8)

The two-dimensional array calculated using Equations 7 and 8 is visualized in image 2302 of Figure 23B.

(C)
The number of data in the first dimension (that is, the horizontal axis) of the two-dimensional array and the number of elements in the preprocessed data do not necessarily match. Also, each element of the input data may be recorded as a set of (x, y).

As an example of a two-dimensional array of input data, the operating angle of the rotating device and the measured value (for example, the value obtained from a strain sensor or a load sensor) are measured together, and the i-th element of the input data is (angle i , the two-dimensional arraying of the input data in the case of being recorded in sets of measurements i). where angle i is the operating angle of the instrument, 0<=angle i<360, for any i.

The data preprocessing function divides the measurement data into a plurality of pieces of input data by dividing the measurement data by a predetermined time. The two-dimensional array function prepares a two-dimensional array with a size of 360×Y_high and sets all elements to zero. Next, the two-dimensional array function calculates the position (Xi, Yi) of the corresponding two-dimensional array element for each element (angle i, measured value i) of the input data. Xi is obtained by truncating the decimal part of the angle i, converting it to an integer, and adding 1 to it. Also, Yi is obtained by Equation (5).

The two-dimensional array function sets (Xi, Yi)=1 for any i. In this case, as shown in image 2303 of FIG. 23C, there may be two or more white cells in one column, or an entire column may be black.

According to this method, two-dimensional arraying is possible even when the number of elements of input data is variable. In addition, even if the elements of the input data are not arranged in ascending or descending order of the axis data, two-dimensional arraying is possible.

(D)
In the example of (A), the elements of the input data and the first dimension of the two-dimensional array are mapped on a one-to-one basis, with n being the first dimension of the two-dimensional array for the elements 1 to n. However, there are cases where n is large in actual data and one-to-one mapping is impossible. Therefore, as an example of two-dimensional arraying of input data, a compression method in the direction of the horizontal axis (that is, frequency, time, distance, etc.) will be shown.

For example, it is possible to calculate an average value for each predetermined unit interval and perform two-dimensional arraying based on the average value. For example, if the average value is calculated every 10 degrees for an angle of 360 degrees, and the order of the first dimension of the two-dimensional array is 36 (=360/10), then a two-dimensional array with a size of 36 x Y_high is possible.

However, in cases where the amplitude of measured values is large, it is conceivable that a single value, such as an average value or maximum value, cannot fully express the characteristics. In this case, for example, the minimum value and the maximum value are calculated for each unit interval, and the area sandwiched between the minimum value and the maximum value is set to 1 (that is, displayed in white in the image). Thus, it becomes possible to form a two-dimensional array corresponding to the amplitude of the measured values.

Note that the predetermined unit interval does not necessarily have to be equal on a linear scale. You can set it. This makes it possible to transform the input data into two-dimensional array data without increasing the size of the two-dimensional array.

Although one input data is converted into one two-dimensional array in FIGS. 23A to 23D, it is not necessary to convert one input data into one two-dimensional array data. It may be divided and each region may be converted into two-dimensional array data for analysis. In cases where the data axis (i.e. frequency, time, distance, etc.) is wide, or when you want to change the granularity of analysis according to the area, set two or more axes (i.e., frequency, time, distance, etc.) regions, and learning and analysis processing may be performed for each of the divided regions. Also, for each region after division, the scale of the axis may be selected individually.

FIG. 24 shows an example of dividing the data area into two or more. In the example of FIG. 24, for example, an axis (that is, frequency, time, distance, etc.) is divided into a region 2401 of 100 or less and a region 2402 of 100 to 10000, and the region 2401 of 100 or less is linearly scaled in a two-dimensional array 2403 , and maps 100 or more regions 2402 to a two-dimensional array 2404 on a log scale. This makes it possible to perform analysis with a granularity suitable for the region to be analyzed.

In addition, in the case where the measured value is small and the variation of the value can be easily understood by using a log scale for the vertical axis, the logarithmic value of the measured value may be used at the time of analysis.

Next, a method of calculating y_max and y_min used in Equation 5 will be described. These parameters may be explicitly set at the time of parameter setting, or may be calculated by a calculation algorithm. An example of an algorithm for calculating y_max and y_min is shown below.

The two-dimensional array function reads learning data, detects the maximum value (hereinafter referred to as Tmp_max) and minimum value (hereinafter referred to as Tmp_min) of measured values, and obtains the average value (hereinafter referred to as Tmp_ave). Equations 9 and 10 below define y_max and y_min.
y_max=Tmp_ave+α*(Tmp_max−Tmp_ave) (Formula 9)
y_min=Tmp_ave−α*(Tmp_ave−Tmp_min) (Formula 10)

Here, α is a predetermined positive constant. α may be set when parameters are set, or may be set to a default value (eg, α=1.2). Also, y_max and y_min may be approximated by the factorial of 10 or its integral multiple.

When the measurements are log-scaled and mapped onto a two-dimensional array, y_max and y_min can be calculated using the logarithms of the measurements, and the analysis can also be performed using the logarithms of the measurements.

Equations 9 and 10 above allow y_max and y_min to be calculated without explicitly specifying them.

Note that the two-dimensional arraying of input data is not limited to the examples given above. In addition, for example, two-dimensional arraying may be performed after performing various preprocessing such as data smoothing, or an envelope may be calculated and the calculated envelope may be mapped to two-dimensional array data. good. Alternatively, the upper envelope and the lower envelope may be calculated separately, and 1 may be assigned to the sandwiched region. Also, the two-dimensional arraying method and various parameters may be specified on the parameter setting screen.

By forming a two-dimensional array as described above, it becomes possible to deal with various types of input data, and it becomes possible to perform analysis that takes advantage of the characteristics of the input data. According to the above method, the number of elements of the input data may be fixed or may vary.

In Example 3, a method of detecting an abnormality by converting data in which each element of input data is represented by a set of two values, ie, (x, y), into two-dimensional array data was shown. This embodiment shows an example of converting input data into an n-dimensional array for analysis.

For example, in the case where each element of the input data consists of a set of three values, i.e. (x, y, z), is converted into a three-dimensional array to detect anomalies, arbitrary i and h, i.e. arbitrary (xi, yh), z_ih is calculated by, for example, the same calculation as Equation 5, and a three-dimensional array is generated according to the following.
(i, h, k)=1 (k=z_ih) (Formula 11)
(i, h, k)=0 (k≠z_ih) (Formula 12)

As in the third embodiment, the absolute value of the difference between the input data and the output data to the autoencoder is calculated for each element (that is, each square of (i, h, k)), and the difference before and after autoencoding is (xi, yh) (i.e., one-dimensional space) by calculating the total value of the differences and performing analysis processing using this, it becomes possible to calculate the abnormality probability for each pair of (xi, yh) . Further, as the difference before and after auto-encoding, the sum of the differences in the two-dimensional space obtained by fixing xi or yh may be used to calculate the abnormality probability. For example, when xi is fixed, the abnormality probability for each xi is calculated by calculating the total value of the differences of all yh for each xi. This makes it possible to calculate the anomaly probability for any space of two dimensions or less.

By using this method, for example, for measurement data obtained from a vibration sensor, it is possible to analyze the time variation of the frequency density from a starting point, where x is the elapsed time, y is the frequency, and z is the frequency density. . For example, for measurement data collected from vibration sensors and sound sensors installed at predetermined intervals on a straight line, x is the position of the sensor (that is, equivalent to the sensor ID), y is the frequency, and z is the spectral density. Monitoring is also possible. It is also possible to install pressure sensors, strain sensors, etc. on a flat or curved surface at regular intervals, and to monitor the condition using (x, y) as the sensor position and z as the measured value. Thus, this method can be applied to various cases.

25A and 25B are diagrams showing display examples of anomaly probabilities. As a display method of the analysis result, for example, as shown in FIG. 25A, it is possible to display the abnormality probability for a combination of date and time. The anomaly probability may be displayed for an arbitrary space, or may be displayed for a part of the space extracted by designating conditions. For example, a space with a high probability of abnormality may be extracted and displayed.

Also, anomaly probability may be displayed for a set of (x, y). For example, in a case where x corresponds to the sensor ID and y corresponds to the frequency, as shown in FIG. 25B, the abnormality probability can be displayed with the sensor ID on the horizontal axis and the frequency on the vertical axis. These displays allow the user of the condition monitoring system 1 to easily detect locations with a high probability of abnormality.

Note that this method can be extended to the case where each element of the input data is represented by a set of four or more values. It becomes possible to convert to dimensional array data and perform analysis. It should be noted that the abnormality probability can be calculated for an arbitrary space of n-1 dimensions or less according to the selection of elements to be fixed. A sum of differences is calculated for each set of fixed elements (including a set consisting of one element). As a result, it is possible to analyze various types of input data while taking into account the characteristics of each space (that is, the amplitude).

A plurality of methods for converting input data into n-dimensional array data may be prepared, and user designation of the conversion method may be accepted. The data analysis device 102 converts input data into n-dimensional array data by a method specified by the input/output device 104 . As a result, various types of data can be more appropriately converted into n-dimensional array data.

An example of performing pooling and unpooling when calculating the difference before and after auto-encoding described in the third embodiment is shown.

The distribution calculation function 605 performs pooling on each of the image 2201 (that is, two-dimensionally arrayed input data) and the image 2202 (that is, output data from the autoencoder) in FIG. Here, max-pooling is performed with a filter size of 3×3 and a slide width of 1, for example.

Next, the distribution calculation function 605 calculates the difference between the input data after pooling and the output data for each element, and obtains an array of the difference. Next, the distribution calculation function 605 unpools the array of calculated differences. Note that unpooling uses the same filter size and slide width as the previously executed max-pooling. Using this as the difference 2203 for each element, the same processing as in the first embodiment is performed thereafter to calculate the total value of the difference for each column. These processes are performed both when calculating the difference distribution in the learning process (S1106) and in the analysis process (S1404).

Note that the filter size and slide width may be set freely. Also, similar to two-dimensional array data, n-dimensional array data may also be analyzed using pooling and unpooling.

According to this embodiment, by performing pooling and unpooling, it is possible to perform analysis that absorbs fluctuations during measurement. As described above, it is possible to appropriately analyze various types of input data.

In Example 1, an autoencoder was generated using data during stable operation, that is, during normal operation. Further, in the monitoring time period data, the data determined to be abnormal by the method of the first embodiment may be collected and learned to generate a new analysis model.

The abnormality probability calculation function 606 records input data determined to be abnormal (hereinafter referred to as abnormality determination data) as abnormality data (not shown). At this time, for example, the type of abnormality is also recorded. The model generation function 603 reads the abnormality determination data classified into the same type of abnormality, generates the autoencoder 2 and the difference distribution 2 in the same manner as in the first embodiment, and records them in the abnormality model and the abnormality difference distribution data (not shown). do. These are generated for each type of anomaly.

In the case where an abnormality model is recorded, the abnormality probability calculation function 606 inputs the input data determined to be abnormal in the abnormality determination process (S1408) of the first embodiment to the autoencoder 2, and calculates the difference a before and after autoencoding. is calculated, and the difference a is compared with the difference distribution 2. In this embodiment, the degree of abnormality (O_i) obtained by Equation 2 of Embodiment 1 is considered as an index value indicating the degree of matching with each state (that is, normal and abnormal). If the difference a is close to any anomaly type, then that anomaly type can be considered the anomaly type of the data under analysis.

FIG. 26 is an example showing the results of analyzing data determined to be abnormal (that is, data to be analyzed) using an analysis model and one or more abnormal models. For each element, as a result of analysis by the analytical model (or abnormal model), when the index value is 0%, it matches the analytical model (or abnormal model), and when the index value is 100%, it is the analytical model (or abnormal model). The abnormality type determination 2601 is displayed as mismatch.

If the type of anomaly in the data to be analyzed is known (that is, an anomaly model has already been generated), one of the anomaly models will match the analysis result, and the type of anomaly can be inferred. However, if, for example, anomaly has multiple factors (for example, anomaly 2 and anomaly 3 occur at the same time), the analysis results of the analysis target data will match anomaly 2 in the elements that show the characteristics of anomaly 2, and the characteristics of anomaly 3. It is also conceivable that the shown element matches anomaly 3.

As a result, in the abnormality type determination 2601, the abnormality type can be estimated based on the comparison between the analysis result of the analysis target data and each abnormality model. Note that the abnormality type determination 2601 may be displayed for one monitoring target data, or may be displayed for a plurality of monitoring target data.

Also, the element ID may correspond to the frequency, the rotation angle of the device, or the time or distance from the base point, or may be an ID that identifies two or more data series. Also, the data series may be collected by more than one sensor. Also, the analysis using the autoencoder generated by utilizing the normal data and the analysis using the autoencoder 2 generated using the abnormality determination data may be performed in parallel.

Further, the analysis model and the abnormality model obtained by monitoring the equipment of the same type as the equipment to be monitored may be obtained and utilized. Further, an abnormality model may be generated by collecting and learning abnormality data of the same type of abnormality by monitoring a plurality of devices of the same type. This enables efficient collection of even difficult-to-collect abnormal data. According to this embodiment, it is possible to guess the type of abnormality when an abnormality occurs.

In the present embodiment, an example of identifying states for a case where there are two or more normal states is shown. In the processing during learning, the data analysis device 102 generates and records an autoencoder for each state and the difference distribution for each element by the same processing as in the first embodiment. In the monitoring process, the data analysis device 102 analyzes the monitoring target data using the autoencoder and difference distribution for each state by the same processing as in the first embodiment, and calculates Equation 2 for each state and for each element. to calculate the index value. As a result, the degree of matching with each state is quantified for each element.

Furthermore, in the case where data in an abnormal state is also collected, an autoencoder in an abnormal state and a distribution of differences for each element are generated in the same way as in the normal state, and the generated autoencoder in an abnormal state and the distribution of differences are used to An index value may be calculated for each element.

FIG. 27 is an example 2701 of the image of the state identification result. As in the sixth embodiment, an index value for each element of monitoring target data is displayed for each state. This makes it possible to identify the state of the monitored data. Also, if it is different from any state, it can be identified as an unknown state. FIG. 28 is another example 2801 of a state identification result image. As shown in FIG. 28, index values for each state and each element may be displayed in chronological order.

FIG. 29 is another example 2901 of the state identification result image. For example, an average index value of all elements may be calculated for an arbitrary date and time for each state and displayed as the probability of being in each state. Alternatively, each element may be weighted to calculate the average value.

According to this embodiment, it is possible to identify which state the monitoring target data is in when there are multiple states.

This embodiment can also be applied to the case of converting input data into an n-dimensional array for analysis. The state identification result can be displayed, for example, in the designated space in the same manner as in FIGS. However, at this time, a space identifier is used instead of the element ID.

This embodiment shows an example of parameter selection when generating an autoencoder. In addition, in the present embodiment, the explanation will focus on the processing different from that of the first embodiment.

FIG. 30 shows an example of the network configuration of the autoencoder. In this example, the autoencoder includes input nodes 3002 , intermediate nodes 3003 , and output nodes 3004 . The number of nodes N in the input node group 3002 and the output node group 3004 is the number of input elements to the autoencoder. xY_high). The number M of nodes in one layer of the intermediate node group 3003 (that is, one vertical row of the intermediate node group 3003) is an integer that satisfies 1<M<N.

In this embodiment, the number M of the intermediate node group 3003 of the autoencoder for one layer (hereinafter referred to as the number of intermediate nodes M) and the learning amount L will be described as examples of parameters when generating the autoencoder. conduct. Here, the learning amount L is, for example, the number of times the model generation function 603 repeats and learns model generation data when generating an autoencoder.

The number of intermediate nodes M and the amount of learning L may be input on the parameter setting screen 1600, for example. The value of the number M of intermediate nodes may be set in proportion to the number of input nodes, such as 1/2 or 1/4 of the number N of input nodes.

The model generation function 603 generates an autoencoder for an arbitrary combination (Mu, Lv) of the number of intermediate nodes M and the amount of learning L. The model calculation function 604 inputs difference distribution generation data to each generated autoencoder and obtains output data. The distribution calculation function 605 calculates the difference for each element (however, for each column when using a two-dimensional array; the same applies hereinafter) for each combination (Mu, Lv) of the number of intermediate nodes M and the amount of learning L. .

In nature, it is known that many non-negative distributions are approximated by the lognormal distribution. Therefore, the distribution calculation function 605 compares the calculated distribution of each difference with the logarithmic normal distribution. Specifically, the distribution calculation function 605 confirms the normality of the distribution of logarithmic values of differences for each element.

For example, the distribution calculation function 605 uses the Shapiro-Wilk test to calculate the test statistic of the distribution of the logarithm of the difference for each element for each combination (Mu, Lv) of the number of intermediate nodes M and the amount of learning L. do. Next, the distribution calculation function 605 calculates the average value of the test statistics calculated for each element.

FIG. 31 shows an image of the average value of test statistics calculated for each combination of the number of intermediate nodes M and the amount of learning L. In FIG. When the mean value of the test statistic is sufficiently large in a certain combination (Mq, Lp) of the number of intermediate nodes Mq and the amount of learning Lp, the distribution of the logarithmic value of the difference of each element approaches the normal distribution.

When there is a combination of (Mq, Lp) in which the average value of test statistics exceeds a preset threshold value, the distribution calculation function 605 selects this combination. If there are multiple combinations of (Mq, Lp) for which the average value of the test statistic exceeds the threshold, for example, the combination that maximizes the average value of the test statistic may be selected.

The distribution calculation function 605 records the corresponding autoencoder in the analysis model 611 for the selected combination (Mq, Lp) of the number of intermediate nodes M and the amount of learning L, and calculates the distribution of the logarithmic value of the difference for an arbitrary element i. The average value M_Ln(i) and the standard deviation σ_Ln(i) of are calculated and recorded in the difference distribution data 609 .

Next, analysis processing will be described. The abnormality probability calculation function 606 calculates the logarithmic value Diff_Ln(i) of the difference before and after auto-encoding for an arbitrary element i. Next, the abnormality probability calculation function 606 standardizes the logarithmic value Diff_Ln(i) of the difference for an arbitrary element i by Equation 13 using the average value M_Ln(i) and the standard deviation σ_Ln(i).
Dist(i)=(Diff_Ln(i)-M_Ln(i))/σ_Ln(i)
(Formula 13)

In the standard normal distribution, the probability of occurrence of a value z or greater is known (ie, the standard normal distribution table), and for Dist(i) its rarity can be obtained using the standard normal distribution table.

The result display function 607 displays Dist(i) corresponding to a set of measurement date and element ID (that is, i) in place of the abnormality probability 1708 on the analysis result display screen 1700 by associating Dist(i) with a color. You may At this time, in Equation 13, when Dist(i)<0, Dist(i)=0 may be set.

In such a case where the distribution of the difference approaches the lognormal distribution, by treating the distribution of the logarithmic value of the difference as a normal distribution, the rarity of any element i can be calculated based on the standard normal distribution table. is also possible.

Note that the autoencoder shown in FIG. 30 is just an example, and the configuration is not limited to this, and for example, it is possible to increase the number of intermediate node layers. Also, in this embodiment, an example in which the number of intermediate nodes M and the learning amount L are varied as parameters is shown, but other parameters may be varied, for example, the number of layers of intermediate nodes may be varied as a parameter. I do not care.

Although the present invention has been specifically described above based on the embodiments, the present invention is not limited to the above embodiments, and can be variously modified without departing from the scope of the invention. For example, the above embodiments have been described in detail in order to explain the present invention in an easy-to-understand manner, and are not necessarily limited to those having all the described configurations. In addition, addition, deletion, or replacement of other configurations can be made to part of the configurations of the above embodiments.

Further, each of the above configurations, functions, processing means, and the like may be realized by hardware, for example, by designing a part or all of them using an integrated circuit. Further, each of the above configurations, functions, etc. may be realized by software by a processor interpreting and executing a program for realizing each function. Information such as programs, tables, and files that implement each function can be stored in recording devices such as memories, hard disks, SSDs (Solid State Drives), or recording media such as IC cards, SD cards, and DVDs.

Further, in each drawing, control lines and information lines are those considered to be necessary for explanation, and not all control lines and information lines for mounting are necessarily shown. For example, it may be considered that almost all structures are interconnected in practice.

Also, each function of the information processing apparatus described above and the arrangement form of each database are merely examples. The layout of each function and each database can be changed to an optimum layout from the viewpoint of hardware and software performance, processing efficiency, communication efficiency, and the like for each information processing apparatus.

Also, the configuration of each database described above can be flexibly changed from the viewpoint of efficient use of resources, improvement of processing efficiency, improvement of access efficiency, improvement of search efficiency, and the like.

The data recording format described above is not limited to the table structure, and may be recorded in other structures such as a queue structure and a list structure as long as each data can be appropriately associated and stored. In addition to numerical values, numerical formulas and the like may be used as a value recording method. Also, each item recorded in each table may be different depending on the application, and is not limited to the items described as examples. Also, the data may be distributed to a plurality of information processing devices.

In addition, in the above, when referring to the number of elements (e.g., number, numerical value, amount, range, etc.), the specific is not limited to the number of , and may be greater than or less than a specific number. Also, in the above explanation, the components (for example, each function, database, element step, etc.) are not necessarily essential unless otherwise specified or clearly considered essential in principle. do not have.

Also, each of the processes described above may be executed either in real time or in batch depending on the object to be monitored and the purpose. Moreover, each embodiment shown above may apply each independently, and may apply combining all or one part of several embodiment.

In addition, as an example of the analysis target, the equipment in the factory was given, but it is not necessarily limited to the equipment and devices in the factory. or strain sensors, etc.), applications other than factories, such as analyzing measurement data when a train passes to identify the state or detect anomalies, are also conceivable.

Also, in each embodiment, an example of collecting data using a sensor has been shown, but it is not necessary to limit the use of the sensor. , image data, and the like.

1: Condition monitoring system, 101: Data storage device, 102: Data analysis device, 103: Sensor, 104: Input/output device, 105: Analysis object, 609: Difference distribution data, 611: Analysis model

Claims (15)

A method for a computer system to monitor monitored data, comprising:
the computer system stores an analytical model;
The analysis model outputs restored data with the same number of elements as the input data for input data composed of a plurality of elements, so that the difference between the input normal data and the restored data for the normal data is small. is learned by
The method comprises: the computer system comprising:
generating a distribution of differences for each element between a plurality of normal data for generating a difference distribution and restored data obtained by inputting the plurality of normal data for generating a difference distribution into the analysis model;
comparing the difference for each element between the monitoring target data and the restored data obtained by inputting the monitoring target data into the analysis model and the distribution of the difference for each element;
A method of calculating an index value for each element based on the comparison. 2. The method of claim 1, wherein
The analysis model generates intermediate data with a reduced number of elements from the input data, generates and outputs the restored data of the input data from the intermediate data,
The method comprises: the computer system comprising:
Acquire multiple normal data for training,
dividing the plurality of normal data for learning into a plurality of normal data for model generation and the plurality of normal data for difference distribution generation;
The plurality of normal data for model generation are input to the analysis model, and the analysis model is adjusted so that each difference between the plurality of normal data for model generation and the restored data of the plurality of normal data for model generation is reduced. how to learn 2. The method of claim 1, wherein
The method according to claim 1, wherein each element of the input data is a value associated with a predetermined value and arranged in a predetermined order. 2. The method of claim 1, wherein
A method, wherein each element of the input data is a set of a first value and a second value associated with the first value. 2. The method of claim 1, wherein
The input data of the analysis model is n-dimensional array data in which n is an integer of 2 or more,
The analysis model generates intermediate data with a reduced number of elements from the input data, generates and outputs restored data with the same number of elements as the n-dimensional array data from the intermediate data,
The method comprises: the computer system comprising:
Acquiring a plurality of normal data for learning, which consists of a plurality of elements, each element consisting of n values,
converting the plurality of normal data for learning into an n-dimensional array data group;
dividing the n-dimensional array data group into a plurality of n-dimensional array data for model generation and a plurality of n-dimensional array data for difference distribution generation;
inputting the plurality of n-dimensional array data for model generation into the analysis model, learning the analysis model so that each difference between the plurality of n-dimensional array data for model generation and corresponding restored data becomes small;
inputting each of the plurality of n-dimensional array data for difference distribution generation into the analysis model to generate restored data for each of the plurality of n-dimensional array data for difference distribution generation;
calculating a difference between each of the plurality of n-dimensional array data for difference distribution generation and each of the restored data of the plurality of n-dimensional arrays for difference distribution generation, for each element of the n-dimensional array, and element for each space of dimension less than n calculating the total value of the differences calculated for each space, and generating a distribution of the total value of the differences for each of the spaces;
converting the monitoring target data into n-dimensional array data and inputting it to the analysis model to generate restored data of the monitoring target data;
calculating, for each element of the n-dimensional array, a difference between the monitoring target data after the n-dimensional arraying and the restored data of the monitoring target data, and calculating a total value of the differences for each element for each of the spaces; calculating an index value for each space based on a comparison with the distribution of the total value of the differences for each space;
A method of outputting the calculated index value. 6. The method of claim 5, wherein the computer system comprises:
Pooling the input data to the analysis model and the restored data, calculating an n-dimensional array for each,
calculating the difference between the input data after the pooling and the restored data for each element to calculate an n-dimensional array of the difference;
unpooling the n-dimensional array of the differences to calculate an n-dimensional array;
calculating the total value for each space of dimensions less than n for the n-dimensional array obtained by the unpooling;
A method of calculating the index value using a total value calculated for each space of the n-dimensional array obtained by the unpooling. 6. The method of claim 5, wherein the computer system comprises:
dividing the input data to the analysis model into a plurality of regions;
A method of calculating an index value by converting each of the divided regions into n-dimensional array data. 2. The method of claim 1, wherein
The method, wherein the analytical model is an autoencoder. 2. The method of claim 1, wherein the computer system comprises:
Set the threshold for abnormality judgment,
A method of determining abnormality of the index value calculated for each of the elements according to the threshold value. 2. The method of claim 1, wherein the computer system comprises:
generating the analysis model for each state with respect to a plurality of states;
generating a distribution for each element of the difference for each state using the analysis model;
inputting the monitoring target data into each analysis model for each state and calculating the difference for each element from the restored data;
calculating an index value for each element for each state based on a comparison between the difference for each element calculated for each state and the distribution of the difference for each element calculated for each state;
The method of outputting the calculated index value. 2. The method of claim 1, wherein
A method, wherein the computer system displays the index value calculated for each element in association with a color. 2. The method of claim 1, wherein the computer system comprises:
generating an analytical model for each combination of one or more parameters,
generating a logarithm distribution of the difference for each element for each analysis model of the analysis model;
Selecting a combination of parameters that makes the distribution of the logarithmic value of the difference close to a normal distribution,
selecting an analysis model corresponding to the selected combination of parameters from the analysis models;
The logarithm of the difference for each element between the monitoring target data and the restored data obtained by inputting the monitoring target data into the selected analysis model, and the logarithm of the difference for each element corresponding to the selected analysis model. make a comparison with the distribution,
A method of calculating an index value for each of the elements based on the comparison. A computer system for monitoring monitoring target data,
one or more processors;
one or more storage devices,
the one or more storage devices store analytical models;
The analysis model outputs restored data with the same number of elements as the input data for input data composed of a plurality of elements, so that the difference between the input normal data and the restored data for the normal data is small. is learned by
the one or more processors
generating a distribution of differences for each element between a plurality of normal data for generating a difference distribution and restored data obtained by inputting the plurality of normal data for generating a difference distribution into the analysis model;
comparing the difference for each element between the monitoring target data and the restored data obtained by inputting the monitoring target data into the analysis model and the distribution of the difference for each element;
A computer system that calculates an index value for each element based on the comparison. 14. The computer system of claim 13,
The analysis model generates intermediate data with a reduced number of elements from the input data, generates and outputs the restored data of the input data from the intermediate data,
The one or more processors
Acquire multiple normal data for training,
dividing the plurality of normal data for learning into a plurality of normal data for model generation and the plurality of normal data for difference distribution generation;
The plurality of normal data for model generation are input to the analysis model, and the analysis model is adjusted so that each difference between the plurality of normal data for model generation and the restored data of the plurality of normal data for model generation is reduced. A computer system that learns 14. The computer system of claim 13,
The input data of the analysis model is n-dimensional array data in which n is an integer of 2 or more,
The analysis model generates intermediate data with a reduced number of elements from the input data, generates and outputs restored data with the same number of elements as the n-dimensional array data from the intermediate data,
The one or more processors
Acquiring a plurality of normal data for learning, which consists of a plurality of elements, each element consisting of n values,
converting the plurality of normal data for learning into an n-dimensional array data group;
dividing the n-dimensional array data group into a plurality of n-dimensional array data for model generation and a plurality of n-dimensional array data for difference distribution generation;
inputting the plurality of n-dimensional array data for model generation into the analysis model, learning the analysis model so that each difference between the plurality of n-dimensional array data for model generation and corresponding restored data becomes small;
inputting each of the plurality of n-dimensional array data for difference distribution generation into the analysis model to generate restored data for each of the plurality of n-dimensional array data for difference distribution generation;
calculating a difference between each of the plurality of n-dimensional array data for difference distribution generation and each of the restored data of the plurality of n-dimensional arrays for difference distribution generation, for each element of the n-dimensional array, and element for each space of dimension less than n calculating the total value of the differences calculated for each space, and generating a distribution of the total value of the differences for each of the spaces;
converting the monitoring target data into n-dimensional array data and inputting it to the analysis model to generate restored data of the monitoring target data;
calculating, for each element of the n-dimensional array, a difference between the monitoring target data after the n-dimensional arraying and the restored data of the monitoring target data, and calculating a total value of the differences for each element for each of the spaces; calculating an index value for each space based on a comparison with the distribution of the total value of the differences for each space;
A computer system that outputs the calculated index value.

Images