JP6875179B2 - System analyzer and system analysis method - Google Patents

Description

The present invention relates to a system analyzer and a system analysis method.

Conventionally, failure detection of various devices provided in a predetermined system (hereinafter referred to as a system, etc.) including various devices, factories, power plants, plants, etc. as components is visually detected by a skilled person or from the devices. This was done by output noise, etc., but with the recent introduction of sensing technology, it has become possible to capture failures in equipment and the like as data and automatically detect these failures. Further, in recent years, with the introduction of analysis technology called machine learning, patterns of data output from various sensors during the period when the system etc. were operating normally are registered, and based on this, the system etc. It is becoming possible to quickly grasp the abnormality or its sign.

However, in this case, in order to detect failures and signs of equipment using analysis technology, it is necessary to apply advanced expertise (for example, knowledge of analysis by equipment and the result of the analysis to the equipment, etc.) and interpret (analyze) it. Knowledge) is required. However, since engineers with such knowledge are rare, it is necessary for engineers with individual knowledge to manually discover such failures and signs while exchanging opinions, and in some cases it is considerable. It took a long time, and a great deal of cost was sometimes incurred.

In this regard, Patent Document 1 defines a rule for the judgment of a person for each target device or plant when diagnosing the remaining life of the device or plant, and automatically analyzes the device by giving it to the system from the outside. The system is disclosed.

Japanese Unexamined Patent Publication No. 2006-302293

According to Patent Document 1, it is possible to reduce the man-hours for developing an application for performing diagnosis, but in order to construct the above rule, expertise in the system (for example, knowledge of analysis of equipment such as an engineer) And knowledge of technology in the field) is required. Since the technique of Patent Document 1 does not have a mechanism for deriving these rules, there is no guarantee that the system can be analyzed accurately.

The present invention has been made in view of such a background, and one object of the present invention is to provide a system analyzer and a system analysis method capable of accurately analyzing the state of a system.

One of the present inventions for solving the above-mentioned problems is a system analyzer including a processor and a memory, and in a predetermined system including a plurality of devices, the state of the devices is time-series. State data, which is the data to be shown, is acquired for each of the plurality of devices, and based on the acquired state data, each of the plurality of devices is divided into at least one or more groups according to the characteristics of the state change of the devices. Based on the acquired state data, the device classification unit that classifies into the above, the predecessor relationship estimation unit that estimates the temporal predecessor relationship of the processing performed by each device in the group, the classified group, and the estimated temporal relationship. An analysis model generation unit that generates an analysis model that predicts temporal changes in the state of each device in the group based on the relationship between the future and the future, and an analysis model output unit that outputs information indicating the contents of the generated analysis model. And.

According to the present invention, the state of the system can be accurately analyzed.

FIG. 1 is a diagram showing an example of the configuration of the analysis system 1 according to the present embodiment. FIG. 2 is a diagram showing an example of a hardware configuration included in the system analyzer 120. FIG. 3 is a diagram illustrating an example of the configuration of the history data 150. FIG. 4 is a diagram illustrating an outline of processing performed by the analysis system 1. FIG. 5 is a diagram illustrating an example of analysis processing. FIG. 6 is a diagram illustrating an example of analysis processing. FIG. 7 is a diagram illustrating an example of analysis processing. FIG. 8 is a diagram illustrating an example of a process of cleansing in the time axis direction. FIG. 9 is a diagram illustrating an example of processing related to exclusion of unnecessary sensors and sensor selection. FIG. 10 is a diagram illustrating an example of the contents of the time offset processing. FIG. 11 is a diagram illustrating an example of the content of the optimization process. FIG. 12 is a diagram showing an example of another configuration of the analysis system 1.

<System configuration>
FIG. 1 is a diagram showing an example of the configuration of the analysis system 1 according to the present embodiment. The analysis system 1 is, for example, a mining place for resources such as a factory, a power plant, an oil field, or a public infrastructure system (infrastructure. For example, water pipe, gas pipe, road, automobile, railroad, aircraft. This is an information system introduced into an entity that manages a predetermined system (hereinafter referred to as an analysis target system 10) provided with a plurality of devices 5 such as.).

The analysis system 1 is a system that enables analysis of the analysis target system 10 even by a user 110 who does not have expertise in the field related to the analysis target system 10 (for example, analysis expertise). For example, the analysis system 1 can analyze the possibility that the device 5 is damaged or fails due to its use or deterioration over time.

That is, the user 110 requests the system analysis device 120, which is an information processing device (computer) provided in the analysis system 1, with the processing data 102 including the operation data and the state data described later, and the user 110. The analysis request 100 including the analysis target data 104, which is information related to the analysis, is input. Then, the system analyzer 120 applies the analysis model 108 and data indicating the validity of the analysis model 108 (for example, information indicating system characteristics, an algorithm used for generating the analysis model 108, parameters applied to the algorithm, and application). The analysis result 106, which is the data including the basis data 105, which is the order and combination of the performed algorithms, is output and presented to the user 110.

The analysis target data 104 includes, for example, the past failure date and time or failure location of each device 5, the device 5 to be analyzed by the user 110 (hereinafter, referred to as the analysis target device), and the like.

Further, the device 5 is, for example, a device used in a factory or a power plant, a device used in an oil field or other resource mining site, a pipe, a road, an automobile, a railroad, an aircraft, etc., and each of the plurality of devices 5 Is performing the corresponding processing or operation independently or in cooperation with each other.

Further, each of the devices 5 has a predetermined sensor 6 (for example, a surveillance camera, a thermometer, a voltage) that monitors the state or operation of each device 5 at any time (for example, at a predetermined time interval or a predetermined timing). Meters, ammeters, vibrometers, speedometers, etc.) are provided.

<Configuration of system analyzer 120>
Next, the configuration of the system analyzer 120 will be described.
As shown in FIG. 1, in the system analysis device 120, for each device 5 in the analysis target system 10, data indicating whether or not the device 5 is in operation (hereinafter referred to as operation data) and each sensor 6 are used. By using the data (state data; hereinafter, also referred to as sensing data) indicating the state of the device 5 acquired from the device 5, the characteristics of the analysis target system 10 (hereinafter, referred to as system characteristics, for example, the analysis target system 10). (Characteristics such as scale, arrangement of each device 5, order of processing performed by device 5), and using the estimated system characteristics to predict and analyze the state of device 5 in the system to be analyzed 10. An analysis model 108, which is a model, is created.

The operation data is data indicating the period during which the device 5 was in operation or stopped for each device 5. In the present embodiment, the operation data is time-series data, 0 is set for the operation data during the period when the device 5 is stopped, and 0 is set for the operation data during the period when the device 5 is in operation. The value of is set. The operation data is, for example, the data of the history of power-on or power-off stored in the device 5.

The sensing data is time-series data acquired by the sensor 6, and is, for example, time-series data showing time changes such as the power supply state, production amount, device state, temperature, and voltage of the device 5, or speed, material. It is time series data which shows the time change of the input amount of.

The analysis model 108 is a numerical model generated by machine learning, and stores, for example, a mathematical formula representing the relationship between each device 5 and its failure time. As a result, the user 110 can input the analysis request 100 to the system analyzer 120 to know how close the current state of each device 5 is to the failure regardless of his / her specialty. it can. The process of utilizing such an analytical model 108 is called diagnosis in the field of machine learning.

Note that FIG. 2 is a diagram showing an example of the hardware configuration included in the system analyzer 120. As shown in the figure, the system analyzer 120 includes a processor 11 including a CPU (Central Processing Unit) and the like, and a RAM (Random Access Memory), a ROM (Read Only Memory), an NVRAM (Non-Volatile RAM), or the like. Main storage device 12, auxiliary storage device 13 such as hard disk drive and SSD (Solid State Drive), input device 14 such as touch panel and operation buttons, output device 15 such as liquid crystal display and printer, and others. A communication device 16 for communicating with the information processing device of the above is provided.

Next, as shown in FIG. 1, the system analyzer 120 includes an operating device extraction unit 125, a device classification unit 130 (fault involvement sensor output unit), a pre-post relationship estimation unit 135 (time correlation extraction unit), and an analysis model generation unit. It has 142 functions, an information input unit 143, an analysis model output unit 145, and a state prediction output unit 147.

First, in a predetermined system (analysis target system 10) configured to include a plurality of devices, the device classification unit 130 provides state data, which is data indicating the state of the devices in time series, to each of the plurality of devices. Based on the acquired state data, each of the plurality of devices is classified into at least one or more groups according to the characteristics of the state change of the devices.

Here, the operating device extraction unit 125 acquires operation data, which is data indicating that the device is operating or stopped, and is an analysis target based on the acquired operation data and the acquired state data. The state data is acquired. Then, the device classification unit 130 classifies each of the plurality of devices into at least one or more groups (hereinafter, referred to as a sensor group) based on the acquired state data of the analysis target.

The pre-post relationship estimation unit 135 estimates the temporal pre-post relationship of the processing performed by each of the devices in the group based on the acquired state data.

The analysis model generation unit 142 generates an analysis model (analysis model 108) which is a model for predicting the temporal change of the state of each device in the group based on the classified group and the estimated temporal predecessor relationship. ..

The analysis model generation unit 142 includes a parameter optimization unit 140 that calculates a characteristic value, which is a parameter that characterizes the predetermined system, based on the time-pre-post relationship estimated by the pre-post relationship estimation unit 135. The analysis model generation unit 142 generates the modified analysis model based on the calculated characteristic value.

For example, the parameter optimization unit 140 calculates a parameter indicating the temporal pre-post relationship of the processing between the devices as the characteristic value, and based on the calculated parameter, satisfies the pre-post relationship, the modified analysis model. To generate.

Further, for example, the parameter optimization unit 140 calculates, as the characteristic value, a parameter indicating the time required to complete all the processes in the predetermined system, and the calculated parameter exceeds a predetermined threshold value. In the case, the state data in a part of the time is specified, and the modified analysis model is calculated based on the specified state data.

Further, for example, the parameter optimization unit 140 calculates a parameter indicating the frequency with which the apparatus has reached a predetermined state as the characteristic value, and specifies the analysis period of the analysis model based on the calculated parameter. The analysis model in the specified analysis period is generated as the modified analysis model.

Further, for example, the parameter optimization unit 140 calculates, as the characteristic value, a parameter indicating a difference in the characteristics of the state data in a predetermined period before and after a predetermined event, and the calculated parameter satisfies a predetermined condition. In this case, the analysis model before and after the predetermined event is calculated as the modified analysis model, respectively.

The information input unit 143 receives input of information indicating a predetermined state in the past of the device.

The analysis model output unit 145 outputs information indicating the contents of the analysis model generated by the analysis model generation unit 142.

The state prediction output unit 147 estimates the time when the device reaches the predetermined state based on the input state information of the device and the generated analysis model, and outputs the information regarding the estimated time.

Further, the system analyzer 120 stores the history data 150.
<History data>
FIG. 3 is a diagram illustrating an example of the configuration of the history data 150. The history data 150 is a relational database including each database of the history data management table 510, the characteristic type management table 520, the data management table 530, the characteristic value management table 540, and the work management table 550.

Of these, the history data management table 510 is a table that stores the contents of the analysis request 100 and the analysis result 106, and is an identifier assigned to each input analysis request 100 (for each analysis process) (hereinafter, history ID). The analysis input immediately before the analysis request 100 of the history ID 511 that stores the history ID 511, the reference ID 512 that stores the information for associating with other tables (hereinafter referred to as the reference ID), and the history ID 511. The failure correlation degree (analysis result) calculated in response to the previous analysis ID 513 (“0” is set in the case of the first analysis) in which the history ID of the request 100 is stored and the analysis request 100 of the history ID 511. A value representing the quality of 106. Details will be described later. The failure correlation degree 514 and the adoption flag 515 (for example, use) in which the evaluation of the user 110 with respect to the failure correlation degree indicated by the failure correlation degree 514 is stored are stored. If the person 110 is satisfied with the analysis result 106, the user 110 or the like stores "1", and if the person 110 is not satisfied, "0" is stored. ).

The characteristic type management table 520 is a table for storing an expression for determining system characteristics (hereinafter, referred to as a characteristic determination expression) used when generating the analysis model 108. The characteristic type management table 520 contains the contents of the characteristic determination formulas (for example, calculation formulas) of the characteristic judgment formula ID 521 and the characteristic judgment formula ID 521 in which the identifiers assigned to each characteristic judgment formula (hereinafter referred to as the characteristic judgment formula ID) are stored. , A program, or a characteristic determination formula 522 in which an API (Application Programming Interface) for calling a program is stored, a characteristic division 523 in which information on a processing method of the characteristic determination expression stored in the characteristic determination expression 522 is stored, and , Information for specifying the process to be performed using the characteristic determination formula 522 stored in the characteristic determination formula 522 is stored. Post-judgment processing 524 (for example, API information for calling a calculation formula, a program, or a program is stored. Specifically, for example, information for identifying the operating device extraction unit 125, the device classification unit 130, or the predecessor-posterior relationship estimation unit 135 is stored).

The characteristic judgment formula 522 includes a characteristic judgment formula for calculating the inter-device order characteristic value, a characteristic judgment formula for calculating the inter-device distance characteristic value, a characteristic judgment formula for calculating the system scale characteristic value, and a failure frequency characteristic, which will be described later. A characteristic judgment formula for calculating a value, a characteristic judgment formula for calculating a characteristic change characteristic value, and the like are stored.

Further, in the characteristic category 523, for example, parameters used for a predetermined process repeatedly performed by the parameter optimization unit 140 are stored. For example, when the value calculated by the characteristic determination formula is different from the value calculated last time, information for specifying a case where both values can be regarded as the same value is stored. For example, when a predetermined value is stored in the characteristic category 523, it is considered that both of the above values match.

The data management table 530 is a table that manages data processed by the system analysis device 120 (for example, data used in each calculation process, set values related to the device 5, data output from the system analysis device 120, and the like). The data management table 530 stores data ID 531 in which an identifier assigned to each data (hereinafter referred to as a data ID) is stored, reference ID 532 in which the reference ID is stored, and information specifying a device for processing the data of the data ID 531. A functional unit that processes data of device type 533 and data ID 531 (specifically, for example, an operating device extraction unit 125, a device classification unit 130, a predecessor relationship estimation unit 135, an analysis model generation unit 142, and a parameter optimization unit). Function type 534 in which information for specifying 140) is stored, data type 535 in which information for specifying the data type of data ID 531 is stored, data content, storage location, or acquisition destination (for example, API) of data ID 531. It has each item of data 536 in which the information to be specified is stored.

The characteristic value management table 540 is a table for storing parameters (hereinafter, referred to as characteristic values) indicating system characteristics calculated by a specific determination formula. The characteristic value management table 540 contains a characteristic value ID 541 that stores an identifier assigned to each characteristic value (hereinafter referred to as a characteristic value ID), a reference ID 542 that stores a reference ID, and a characteristic that stores a characteristic determination expression ID. It has each item of the determination formula ID 543 and the characteristic value 544 in which the characteristic value in the characteristic judgment formula specified by the characteristic judgment formula ID 543 is stored.

<Characteristic value>
Here, a specific example of the characteristic value will be described. The characteristic value is a parameter calculated based on a characteristic judgment formula or the like, and the characteristic value includes, for example, an inter-device order characteristic value, an inter-device distance characteristic value, a system scale characteristic value, a failure frequency characteristic value, and a characteristic change characteristic. There is a value.

First, the order characteristic value between devices is temporally different when there is a time difference in the time change of the state data between the sensors 6 or when there is a time-preceding relationship in the time change of the state data between the sensor groups. It is a characteristic value representing the pre-post relationship (that is, the order between the devices 5).

The inter-device distance characteristic value is a characteristic value representing the distance between the devices 5 calculated from the time difference of the time change of the state data between the sensors 6 or between the sensor groups when the above-mentioned pre-existing relationship exists. Based on this characteristic value, it is presumed that the sensor 6 (device 5) is classified into a plurality of blocks and the analysis target system 10 is composed of the plurality of devices 5.

The system scale characteristic value is a characteristic value indicating the scale (for example, spatial size) of the analysis target system 10 or the device 5, and is analyzed from the time of inputting the material or energy into the device 5 of the analysis target system 10, for example. It is the time until the product or energy is output from the target system 10. The system scale characteristic value is calculated using, for example, the distance between the devices 5.

The failure frequency characteristic value is a characteristic value indicating the failure frequency of the device 5, and is a value indicating the frequency or the number of times that one or more failures have occurred before a predetermined date and time. Details of the method for calculating the failure frequency characteristic value will be described later.

The characteristic change characteristic value is a characteristic value indicating a difference in the behavior of the analysis target system 10 in a period before and after a predetermined event (for example, a stop period of the device 5, a predetermined time, a season, etc.). The characteristic change characteristic value is, for example, the state data of the sensor 6 or the sensor group (hereinafter referred to as pre-stop data) in the predetermined period before the predetermined event and the predetermined period after the predetermined event. , The value indicating the correlation with the state data of the sensor 6 or the sensor group (hereinafter, referred to as the data after stopping). Characteristic change The characteristic value is when the correlation coefficient between the pre-stop data and the post-stop data is greater than or equal to a predetermined threshold (for example, what was 0.5 or less before the stop is 0.8 or more after the stop). A predetermined value is set when the value is increased or when the value is 0.8 or more before the stop and becomes 0.5 or less after the stop.

Next, the work management table 550 is a table for storing parameters related to the optimization process repeatedly performed by the system analyzer 120. The work management table 550 includes a sub-history ID 551 in which an identifier (hereinafter referred to as a sub-history ID) assigned for each optimization process is stored, a history ID 552 in which the history ID is stored, and a reference ID 553 in which the reference ID is stored. , The previous analysis ID 554 in which the sub-history ID of the optimization process performed last time is stored, the failure correlation degree 555 in which the failure correlation degree calculated by the optimization process of the sub-history ID 551 is stored, and the optimization process of each time. Of these, the adoption flag 556 in which a predetermined value is stored when the result of the optimization processing of the sub-history ID 551 is determined to be the most appropriate (Note that the adoption flag 556 indicates the appropriateness of the result of the optimization processing. It has each item of (the order may be stored).

The above history data 150 includes processing performed by the operating device extraction unit 125, the device classification unit 130 (failure involvement sensor output unit), the pre-post relationship estimation unit 135 (time correlation extraction unit), and the analysis model generation unit 142. The data used for the processing may be further stored.

The above-mentioned functions of the system analyzer 120 are executed by the hardware of the system analyzer 120 or by the processor 11 of the system analyzer 120 by reading and executing the program stored in the main storage device 12 or the auxiliary storage device 13. It is realized by.

<Processing>
Next, the processing performed by the system analyzer 120 will be described. First, the outline of the processing of the analysis system 1 will be described.

<Outline of processing>
FIG. 4 is a diagram illustrating an outline of processing performed by the analysis system 1. First, the system analyzer 120 receives the input of the analysis request 100 and analyzes the content of the analysis request 100 that has received the input (s10). Then, the system analyzer 120 performs a process (hereinafter, referred to as data cleansing) for improving the quality of the data of the analysis request 100 based on the analyzed contents (s20).

Specifically, the system analyzer 120 first identifies a period during which the device 5 has been operating based on the state data and the like, and extracts the state data of the device 5 during that period (hereinafter referred to as time axis direction cleansing). ) Is performed (s21). This process is mainly performed by the operating device extraction unit 125.

Next, the system analyzer 120 performs a process of identifying the failed device 5 or the device 5 related to the failure (hereinafter, referred to as unnecessary sensor exclusion) among the devices 5 that have been cleansed in the time axis direction (hereinafter, referred to as unnecessary sensor exclusion). s22). This process is mainly performed by the device classification unit 130.

Further, the system analyzer 120 performs a process (hereinafter, referred to as sensor selection) of classifying the device 5 (sensor 6 corresponding to the device 5) from which unnecessary sensors have been excluded into at least one or more groups (sensor groups). Do (s23). This process is mainly performed by the device classification unit 130.

The system analyzer 120 performs a process (hereinafter, referred to as a time offset) for specifying the temporal relationship between the processes performed by each sensor group for the sensor groups classified by sensor selection (s24). It should be noted that this processing is mainly performed by the predecessor-posterior relationship estimation unit 135.

When the data cleansing is completed as described above, the system analyzer 120 analyzes the state data of each group for which the data cleansing has been performed, generates an analysis model 108, and optimizes the analysis model 108 (hereinafter,). , It is called optimization processing) (s30). The system analyzer 120 may repeatedly perform data cleansing based on the result of the optimization process. This optimization process is mainly performed by the analysis model generation unit 142 (parameter optimization unit 140).

When the optimization process is completed, the system analyzer 120 outputs the analysis result 106 including the basis data 105 and the analysis model 108 and presents it to the user 110 (s40).

Next, the processing performed by the system analyzer 120 will be described.
<Analysis processing>

5, FIG. 6 and FIG. 7 are diagrams illustrating an example of a process (hereinafter referred to as an analysis process) in which the system analyzer 120 generates an analysis model 108 based on the analysis request 100 and outputs the analysis result 106. is there. This process is started, for example, when a predetermined input is made to the system analyzer 120 by the user 110.

First, the information input unit 143 of the system analyzer 120 receives the input of the analysis request 100 from the user 110 (s210).

(Cleansing in the time axis direction)
Next, the operating device extraction unit 125 of the system analysis device 120 identifies all the periods during which the device 5 was operating or stopped by referring to the operating data in the analysis request 100 (s215). Specifically, for example, the operating device extraction unit 125 specifies a period in which the value is 0 or has a value separated from 0 by a predetermined value or more as operating data.

Next, the operating device extraction unit 125 acquires the state data of each device 5 at the failure date and time based on the failure date and time information of the device 5 included in the analysis request 100 (s220).

The operating device extraction unit 125 may allow the user 110 to select the sensor 6 for acquiring the operating data (s230). Further, the operating device extraction unit 125 may allow the user 110 to select the state data to be acquired.

Next, the operating device extraction unit 125 stores the processing results of s215 and s220 in the history data 150 (s227). Specifically, for example, the operating device extraction unit 125 generates a new record in the data management table 530, and in the generated record, information (for example,) that specifies the stop period, the operating period, and the failure date and time of each device 5. , Relationship between the boundary time of the operation period or the stop period and the failure date and time).

Further, the operating device extraction unit 125 stores various data and parameters used in the processing of s215 and s220 in the history data 150 (s227). Specifically, for example, the operating device extraction unit 125 generates a new record in the data management table 530 and stores the state data in the data 536 of the generated record.

Then, the operating device extraction unit 125 transmits the state data of each sensor 6 (device 5) extracted in s220 (for example, the state data of each sensor 6 (device 5) other than the past stop period) to the device classification unit 130. Send (s229).

Here, FIG. 8 is a diagram illustrating an example of the above time axis direction cleansing process. For example, the operating device extraction unit 125 identifies the period 801 when the device 5 was normally stopped, which is specified based on the operating data. Then, the operating device extraction unit 125 considers a period other than the period 801 as a period during which the device 5 was operating (operating period). The operating device extraction unit 125 extracts all the state data 802 of each device 5 during the operating period.

In this way, the operating device extraction unit 125 extracts the state data (sensing data) of the operating device 5. Sensing data is usually collected for the entire period from the data collection start time, but the analysis target system 10 is not always in operation in actual operation. For example, the device 5 may receive a planned stop, a stop due to a failure, a stop for energy saving practice, or the like. In such a case, when the sensing data for the entire period is used to generate the analysis model 108 described later, it is determined that the device 5 is stopped due to a failure even though it is normally stopped. This may reduce the accuracy of the analytical model 108. Therefore, it is desirable to exclude in advance the sensing data of the device 5 that is normally stopped, which reduces the accuracy of the analysis model 108. Conventionally, such exclusion processing has been performed manually by the user 110 on the premise that the user 110 has specialized knowledge. However, according to the operating device extraction unit 125, such exclusion processing can be automatically performed regardless of whether or not the user 110 has specialized knowledge.

(Exclusion of unnecessary sensors and center selection)
Next, as shown in FIG. 5, the device classification unit 130 analyzes the characteristic (pattern) of the temporal change of the state data of each sensor 6 received from the operating device extraction unit 125 (s235). Specifically, for example, the apparatus classification unit 130 analyzes the characteristics of the temporal change of the state data based on the regression analysis method, the principal component analysis method, and the like.

Then, the device classification unit 130 classifies each device 5 into a plurality of groups (sensor groups) based on the difference in each feature analyzed in s235 (for example, based on the similarity between the state data and the correlation coefficient). s240). Specifically, for example, the device classification unit 130 uses a regression analysis method to identify a group of devices 5 having a temporal change in state data having a high correlation coefficient (for example, the correlation coefficient exceeds a predetermined threshold value). It is classified into the sensor group of. Further, for example, the device classification unit 130 classifies devices 5 having similar characteristics of temporal changes in state data into the same sensor group based on the K-means method or the like.

If there is state data whose value is 0 during a predetermined period, the device classification unit 130 excludes the sensor 6 corresponding to the state data from the analysis target (sensor group target).

The device classification unit 130 sets a past period (hereinafter, referred to as a normal operation period) in which the device 5 of each sensor group has been operating normally (s245). Specifically, for example, the apparatus classification unit 130 sets a period from half a year before the failure occurrence date and time indicated by the information included in the analysis request 100 to a predetermined date and time one year or more before the failure occurrence date and time as a normal operation period. Set.

Then, the device classification unit 130 generates the analysis model 108 based on the state data of each sensor group in the normal operation period set in s245 and the system characteristics recorded in the characteristic value management table 540 (s250).

That is, the device classification unit 130 learns the normal operation period regarding the relationship between the state of the device 5 indicated by the state data of each sensor 6 and the time when the device 5 is in that state according to the characteristics of the system 10 to be analyzed. Machine learning is performed as a period, and an analysis model 108 based on this is generated.

The device classification unit 130 acquires the system characteristics from, for example, the characteristic type management table 520 or the characteristic value management table 540, or acquires the system characteristics from the characteristic values calculated by the optimization process described later. Further, the apparatus classification unit 130 may generate an analysis model 108 based on, for example, a regression analysis method, a K-means method, or a K-nearest neighbor method.

The device classification unit 130 determines the degree of abnormality of the sensor group for the entire period extracted by time axis cleansing based on the analysis model 108 generated in s250, the failure date and time information, and the state data in each sensor group. Perform the calculation process (hereinafter referred to as abnormality diagnosis) (s255). The device classification unit 130 performs an abnormality diagnosis by an algorithm similar to the algorithm adopted for generating the analysis model 108 in the s250.

The degree of anomaly is a value indicating the degree of dissociation between state data. For example, the degree of anomaly at a certain point in time can be defined as the difference between the value of the state data at a certain point in time in the target sensor group and the value of the state data in the normal operation period in all the sensor groups. For example, when the regression analysis method is used as the algorithm for diagnosing the abnormality, the device classification unit 130 calculates the degree of abnormality as a distance from the regression curve.

Next, as shown in s320 of FIG. 6, the device classification unit 130 is an index (hereinafter, referred to as failure correlation degree) indicating the high degree of involvement of each sensor group in the failure at a predetermined date and time based on the result of the abnormality diagnosis. Is calculated. The failure correlation degree calculated here is, for example, a value obtained by dividing the abnormality degree of the failure date and time by the average value of the abnormality degree of the learning period.

Then, the device classification unit 130 identifies the sensor group having the highest failure correlation degree (for example, the failure correlation degree exceeds a predetermined threshold value). It is presumed that the sensor group with a high degree of failure correlation has a large relationship with the failure of the device to be analyzed.

The value of the failure correlation degree calculated by this process differs depending on, for example, the contents of the parameters used in the exclusion of unnecessary sensors and the selection of sensors. For example, when the above regression analysis method is used, the degree of failure correlation varies depending on the period of the state data targeted for the pattern analysis of s235. Therefore, for example, when a high failure correlation degree (for example, a failure correlation degree of 10 or more) is not obtained in s320, the device classification unit 130 changes the period of the state data used for the pattern analysis of s235. The same process may be repeated again. Regarding the repetition of such analysis processing, if resources such as the system analyzer 120 have a margin, each analysis processing may be executed in parallel. Further, when the analysis process may take a sufficient time, the device classification unit 130 performs each analysis process until a higher condition is satisfied (for example, until a higher value of failure correlation degree is obtained). You may go. Further, the device classification unit 130 may use the sensor group classified in the analysis process in which the highest degree of failure correlation is obtained among the analysis processes performed a plurality of times as the final sensor group.

Further, in the processing in s245 to s320 (hereinafter referred to as this processing in this paragraph), each parameter such as the identifier of the sensor group, the learning period, the number of clusters in the classification of the sensor group, or the algorithm adopted in machine learning is used. Be done. Since these parameters do not depend on each other in this process, the apparatus classification unit 130 performs the present process by various combinations in which the values of these parameters are changed when resources such as the system analyzer 120 are available. The analysis model 108 may be generated by executing in parallel, and the analysis model 108 based on the combination of the parameters having the highest degree of failure correlation among the generated analysis models 108 may be finally selected. Further, the device classification unit 130 creates a ranking of the failure correlation degree calculated based on the above combination and presents it to the user 110, and causes the user 110 to select the above combination to obtain the final analysis model 108. It may be specified.

The device classification unit 130 stores the data, parameters, contents of the analysis model 108, etc. used in s235 to s320 in the history data 150 (s322). For example, the device classification unit 130 generates a record of a new history ID in the history data management table 510, and stores the failure correlation degree calculated in s320 in the failure correlation degree 514 of the generated record. Further, the device classification unit 130 generates a record of a new data ID in the history data management table 510, and stores the data used in s235 to s320 in the data 536 of the generated record.

The device classification unit 130 transmits the information and state data of the sensor 6 (device 5) grouped by s240 to the pre-post relationship estimation unit 135 (s323).

Note that FIG. 9 is a diagram illustrating an example of processing related to exclusion of unnecessary sensors and sensor selection. As shown in the figure, the state data 901 and the state data 903 having similar characteristics of the temporal change are classified into the sensor group of "group 1" by the sensor selection (the degree of correlation between the state data 901 and the state data 903 is high). (High), the state data 902 and the state data 904 are classified into the "group 2" sensor group (the degree of correlation between the state data 902 and the state data 904 is high). Further, by excluding unnecessary sensors, the state data 905 whose value is 0 is excluded from the sensor group (the degree of correlation with any state data is low).

In this way, the device classification unit 130 groups the sensors 6 (device 5) according to the characteristics of the temporal change of the state data. Since the sensors 6 are provided in the devices 5 at various positions in order to identify the cause of the failure, it is highly possible that the sensors 6 show the same data tendency when the analysis target system 10 is small. , When the device 5 is installed in a large-scale factory, even if the sensor 6 is the same type of sensor, the change in the state data may tend to differ depending on the location where the sensor 6 is installed. .. Therefore, in order to correctly grasp the state of the device 5, the sensors 6 are grouped based on the tendency of the state data of the sensor 6 instead of the position of the sensor 6, so that the device 5 to which the sensor 6 is attached is spatially. It is possible to grasp various features and functional features, and to effectively find a sign of failure of the device 5. Conventionally, such grouping of devices 5 (sensors 6) has been performed manually by a person having specialized knowledge in the field. However, the device classification unit 130 can automatically and appropriately group the devices 5 (sensors 6) without depending on such a person.

(Time offset)
Next, as shown in s325 of FIG. 6, the predecessor-posterior relationship estimation unit 135 has the sensor 6 (analysis request) having the highest contribution to failure for each sensor group based on the state data received by s323 and the information of each sensor group. A sensor 6) that has a high contribution to failure at the failure date and time specified by 100 is specified. The contribution of a certain sensor 6 is calculated based on, for example, the degree of similarity of the time change of the state data of the sensor 6 with respect to the time change of the average value of the state data of the sensor group to which the sensor 6 belongs.

Then, the pre-post relationship estimation unit 135 shifts each state data of each state data of each sensor 6 specified above by a predetermined time in the positive or negative time axis direction (hereinafter, post-shift state data). ) Is generated, and the degree of correlation (for example, correlation coefficient) between these post-shift state data is calculated.

The pro-posterior relationship estimation unit 135 performs calculations that are shifted by various times in the positive or negative time axis direction a plurality of times in this way, and among them, the calculation with the highest correlation coefficient (for example, 0.8 or more). Each state data generated by the shift in (calculation when the correlation coefficient of) is obtained (that is, the state data with time offset. Other state data that are not the target of the shift are also shifted. You may do this), and other data and parameters used in the processing of s325 are stored in the history data 150 (s327). Further, the pre-post relationship estimation unit 135 transmits the state data (time-shifted state data) generated in s325 to the parameter optimization unit 140 (s328).

The significance of performing the above time offset is as follows. That is, depending on the scale of the system 10 to be analyzed and the like, it may take a certain amount of time for the influence of the failure generated in one device 5 to appear in another device 5. In this case, even if the correlation of the state data between the devices 5 at a certain time is not so high, the correlation may be high by considering the time required for this expression. The time offset is a process corresponding to such a case.

Since the above-mentioned correlation calculations can be performed independently, if resources such as the system analyzer 120 have a margin, the pre-post relationship estimation unit 135 may execute these calculations in parallel. Good. Further, the pre-post relationship estimation unit 135 may output these calculation results and present them to the user 110.

Here, FIG. 10 is a diagram illustrating an example of the content of the time offset processing. As shown in the figure, the sensor group 1001 (group 1) and the sensor group 1002 (group 2) are sensor groups related to the sensor 6 attached to the device a, which is a device that performs processing at time x, and are sensors. It is assumed that the group 1003 (group 3) and the sensor group 1004 (group 4) are groups related to the sensor 6 attached to the device n, which is a device that performs processing at time x + t. In this case, if the time change of the state data of the sensor group 1001 and the sensor group 1002 is shifted by time t, it resembles the time change of the state data of the sensor group 1003 and the sensor group 1004, respectively. Therefore, the pre-post relationship estimation unit 135 can set that group 1 and group 3 are the same sensor group, and group 2 and group 4 are the same sensor group.

In this way, the pre-post relationship estimation unit 135 shifts the state data of each sensor group (sensor 6) in the time axis direction. That is, when the system 10 to be analyzed is small and the distance between the sensors 6 is short, the time when the failure of each device 5 actually occurs is the same as the time when the state data observed by each sensor 6 fluctuates, or In the same sampling section, but when the system 10 to be analyzed is large-scale (for example, in the case of devices distributed in a factory or the like), a failure that occurred in one device 5 and an observation in another device 5 were observed. Failures are not observed by each sensor 6 at the same time or within the same sampling interval, and as a result, there is a possibility that the time of failure of the device 5 cannot be correctly specified. Therefore, when determining whether or not the device 5 has a failure, it may be necessary to consider the time required for propagation of the failure (considering the time difference). The pre-post relationship estimation unit 135 can automatically consider and adjust such a time difference by shifting the state data.

(Optimization process)
Next, as shown in s330 of FIG. 6, the parameter optimization unit 140 refers to the history data management table 510 to calculate the failure correlation degree this time and the failure correlation degree (history) calculated in the previous processing. The failure correlation degree obtained from the data management table 510) is compared. Then, when the failure correlation degree calculated this time is lower than the failure correlation degree calculated in the previous processing, the parameter optimization unit 140 performs the following optimization processing.

That is, the parameter optimization unit 140 calculates the characteristic value in the characteristic determination formula by referring to the characteristic type management table 520, and performs the optimization process based on the calculated characteristic value.

For example, in the case of the inter-device order characteristic value, the parameter optimization unit 140 analyzes the time-varying characteristics of the state data of each sensor group (or sensor 6), and each sensor group (or the sensor 6) in the order in which the analyzed time-changing characteristics appear. Alternatively, the inter-device order characteristic value is calculated by arranging the sensors 6). Then, the parameter optimization unit 140 causes the data cleansing (specifically, for example, the processing after the device classification unit 130) to be executed again based on this characteristic value.

Further, in the case of the inter-device distance characteristic value, the parameter optimization unit 140 sets the time difference of the time change of the state data between the sensors 6 or between the sensor groups based on the calculated inter-device order characteristic value. Calculate as. Then, the parameter optimization unit 140 causes the data cleansing (specifically, for example, the processing after the device classification unit 130) to be executed again based on this characteristic value.

Further, in the case of the failure frequency characteristic value, the parameter optimization unit 140 is a sensor group having a failure correlation degree equal to or higher than a predetermined threshold value before the failure date and time indicated by the analysis request 100 among the sensor groups calculated by the device classification unit 130. All the times (time zones) that existed are specified, and the number of specified times (time zones) is calculated as the failure frequency characteristic value. Then, when the failure frequency characteristic value is 1 or more, the parameter optimization unit 140 sets a predetermined period before and after each specified time period (time zone) as a target period (learning period) for generating the analysis model 108. ), The data cleansing (specifically, for example, the processing after the operating device extraction unit 125) is executed again.

Further, in the case of the characteristic change characteristic value, for example, the parameter optimization unit 140 calculates the characteristic change characteristic value for each of the stop periods generated for the device 5, and based on the calculated characteristic change characteristic value, the specific change characteristic value. Before and after the stop period of the apparatus 5 according to the above, data cleansing (specifically, for example, processing after the operating apparatus extraction unit 125) on the premise that different analysis models 108 are generated is executed again. Thereby, for example, the system analysis device 120 can generate an appropriate analysis model 108 even when the tendency of the operation of the device 5 changes before and after the device 5 is stopped or due to seasonal fluctuations.

Further, in the case of the system scale characteristic value, for example, when the calculated system scale characteristic value exceeds a predetermined threshold value, the parameter optimization unit 140 has a learning period in which the fluctuation of the state data is small or absent (for example,). , The period in which the fluctuation range of the state data is less than the predetermined value) is specified, and the data cleansing on the premise that the analysis model 108 is generated based on the state data of the sensor 6 in the predetermined period before the specified period is executed again. .. That is, when the scale of the analysis target system 10 is large, if the failure of the device 5 in the front stage of the process is resolved, there is a high possibility that the failure of the device 5 that performs the process in the latter stage will not be affected. Data cleansing (specifically, for example, processing after the device classification unit 130) that generates the analysis model 108 by performing machine learning only for 5 is executed again.

Note that FIG. 11 is a diagram illustrating an example of the content of the optimization process. As shown in the figure, the optimization process is performed on the system characteristics specified by the inter-device sequence characteristic value, the inter-device distance characteristic value, the system scale characteristic value, the failure frequency characteristic value, or the characteristic change characteristic value described above. Based on this, data cleansing and generation of analytical model 108 are performed.

The parameter optimization unit 140 repeats the above optimization process until the failure correlation degree cannot be expected to improve (s410), as described above.

In this way, the parameter optimization unit 140 estimates the system characteristics based on the processing results of the functional units of the operating device extraction unit 125, the device classification unit 130, and the predecessor-posterior relationship estimation unit 135, and at least one of the functional units. The analytical model 108 is optimized by giving constraints on the parameters of the crab. Since the repetition of such processing (optimization processing) has been performed by a system analysis engineer while discussing with a specialized engineer, it has been a heavy work, but the parameter optimization unit 140 has this. Such optimization processing can be performed automatically and accurately.

That is, the system analyzer 120 automatically discovers the characteristics of the equipment or system that have been conventionally given by engineers in specialized fields based on their knowledge and experience, and further performs optimization processing based on the results. It is possible to accurately grasp the state and system characteristics of the device 5 in the analysis system. Further, since the characteristic value calculated in this optimization process is managed in association with the history ID and the analysis result by the history table 140, the system analyzer 120 combines the past analysis result with the current analysis result. By making a comparison, further optimization can be performed and each parameter can be adjusted. As a result, the time required for optimizing the analysis model 108 can be reduced.

Next, the parameter optimization unit 140 manages historical data of various data and parameters registered in the work management table 550, which constitute the analysis result 106 (foundation data 105 and analysis model 108). Register in table 510 (s415).

Specifically, for example, the parameter optimization unit 140 sets each item of the record of the work management table 550 in which the value of a certain reference ID is stored in the reference ID 553, and each item of the new record of the history data management table 510. The value of the reference ID is stored in the reference ID 512 of the record of the history data management table 510.

Then, the parameter optimization unit 140 instructs the analysis model output unit 145 and the state prediction output unit 147 to output the analysis result 106 and present it to the user 110 (s420). Specifically, for example, the parameter optimization unit 140 instructs the analysis model output unit 145 to output the contents of the generated analysis model 108, and the analysis model output unit 145 outputs the contents. For example, the analysis model output unit 145 outputs information indicating the time when the analysis target device fails, which is estimated by the analysis model 108.

Further, the parameter optimization unit 140 instructs the state prediction output unit 147 to output the contents stored in the history data 150 in s415 as the basis data 105, and the state prediction output unit 147 outputs the contents.

The user 110 confirms the output analysis result 106 and confirms that an appropriate analysis model 108 has been created (s425).

The parameter optimization unit 140 receives an input from the user 110 as to whether or not an appropriate analysis model 108 has been created. When an input indicating that an appropriate analysis model 108 has been created is made (“OK”), the parameter optimization unit 140 stores a predetermined value in the adoption flag 515 of the history data management table 510.

On the other hand, when it is input that the appropriate analysis model 108 has not been created (“Retry”), the parameter optimization unit 140 generates a new record in the history data management table 510, and each of the generated records After copying the item to the information work management table 550 in the reference ID, the process from s330 is executed again (s430).

As described above, according to the system analyzer 120 of the present embodiment, state data is acquired for a plurality of devices 5, and each of the plurality of devices 5 is grouped according to the characteristics of changes in the state of the devices 5. Classify into (sensor group), estimate the temporal predecessor relationship of the processing performed by each device 5 in each group, generate an analysis model 108 that predicts the temporal change of the state of each device 5 in the group, and generate the content thereof. Since the indicated information is output, the user 110 and the like can estimate the temporal change of the state of the device 5 for each group of the devices 5 having a similar pattern of the state change based on this analysis model 108. As described above, according to the system analyzer 120 of the present embodiment, the worker or the like can accurately analyze the state of the system even if he / she does not have specialized knowledge about the system (system to be analyzed 10). Is possible. For example, in the past, when discovering a failure location or a sign of failure of a device or system, an engineer who is familiar with the characteristics of the device or system and an engineer who is familiar with analysis technology are indispensable. Needed to work closely together and create an analytical model through trial and error. Therefore, the creation of the analysis model depends on the knowledge and experience of the engineers in the field, and a long period of time is required for development. However, according to the system analyzer 120 of the present embodiment, such a burden is reduced. can do. In this way, by quickly grasping the failure of equipment and plants and their signs, these operators can reduce the cost related to personnel and replacement parts, and the impact on the system due to the failure and the end user. It is possible to reduce the risk of compensation from.

Further, the system analyzer 120 of the present embodiment receives input of information indicating a predetermined state in the past of the device 5, estimates and estimates the time for the device 5 to reach the predetermined state based on the generated analysis model 108. Since the information regarding the time is output, the user 110 or the like can predict the time when the device 5 will fail in the future by inputting the information of the past failure date and time of the device 5, for example. As a result, the user 110 and the like can accurately predict the state of the system even if they do not have specialized knowledge about the system (analysis target system 10).

Further, the system analysis device 120 of the present embodiment acquires operation data, acquires state data to be analyzed based on the acquired operation data and state data, and based on the acquired state data, each of the plurality of devices 5 Is grouped so that an analytical model 108 based only on the device 5 in operation or properly stopped can be generated. Thereby, for example, it is possible to generate an analysis model 108 in which an appropriate analysis is performed on the device 5.

Further, the system analysis device 120 of the present embodiment calculates a characteristic value which is a parameter that characterizes a predetermined system (analysis target system 10) based on the state data, and based on the calculated characteristic value, corrects the analysis model 108. Since it is generated, the optimum analysis model 108 can be generated according to the characteristics of the analysis target system 10. As a result, the user 110 and the like can accurately analyze the state of the system even if they do not have specialized knowledge about the system.

For example, the system analyzer 120 of the present embodiment calculates a parameter (order characteristic value between devices) indicating a temporal predecessor / posterior relationship between the devices 5 as a characteristic value, and based on the calculated parameter, the predecessor / posterior relationship is established. Since the satisfied and modified analysis model 108 is generated, it is possible to generate the analysis model 108 that correctly reflects the processing order of the apparatus 5 in the analysis target system 10.

Further, for example, the system analyzer 120 of the present embodiment calculates, as a characteristic value, a parameter (system scale characteristic value) indicating the time required to complete all the processes in the predetermined system (analysis target system 10). , When the calculated parameter exceeds a predetermined threshold, the analysis model 108 corrected based on the state data in a part of the time period is calculated, so that it is necessary even when the scale of the analysis target system 10 is large. The analytical model 108 can be quickly generated based only on the state data of a certain period.

Further, for example, the system analysis device 120 of the present embodiment calculates a parameter (failure frequency characteristic value) indicating the frequency with which the device 5 has reached a predetermined state as a characteristic value, and the analysis model 108 is based on the calculated parameter. Since the analysis period of the analysis model 108 is specified and the analysis model 108 in the specified analysis period is generated as a modified analysis model 108, an appropriate period (for example, a period in which the device 5 has failed) when the analysis model 108 is machine-learned is excluded. By generating the analysis model 108 based only on the section), a more appropriate analysis model 108 can be generated.

Further, for example, the system analyzer 120 of the present embodiment calculates a parameter (characteristic change characteristic value) indicating a difference in the characteristics of the state data in a predetermined period before and after a predetermined event as a characteristic value, and the calculated parameter is When the predetermined condition is satisfied, the analysis model 108 at the time before and after the predetermined event is calculated as the modified analysis model 108, respectively. Therefore, for example, the stop period of the device 5 or the period such as the predetermined season as the predetermined event. By specifying, it is possible to generate an appropriate type of analytical model 108 according to the difference in the characteristics of the period.

The characteristic value is information that generalizes the knowledge of engineers in each specialized field. In the conventional analysis method, in order to generalize or rule the knowledge of a specialist engineer, the specialist engineer in each field can interpret the technical terms and definitions each time and process them with mathematical formulas or other programs. It was necessary to convert it into a simple format, but such abstraction and extraction work was very difficult. However, since the system analyzer 120 of the present embodiment utilizes highly versatile characteristic values in consideration of such abstraction and extraction steps, these can be applied in various specialized fields.

<Other system configurations of analysis system 1>
By the way, in the analysis system 1 described above, it is assumed that the user 110 inputs the analysis request 100 including the information about the sensor 6 to the system analysis device 120, but the system analysis device 120 receives the analysis request 100. , It may be automatically acquired from the analysis target system 10.

FIG. 12 is a diagram showing an example of another configuration of the analysis system 1. As shown in the figure, in this analysis system 1, the sensor 6 acquires the state data of the device 5 together with the operation data at a predetermined timing or a predetermined time interval, and these acquired data are used as processing data 102 for system analysis. It is transmitted to the device 120. Further, the information processing device 8 different from the system analysis device 120 transmits the analysis target data 104 to the system analysis device 120 at a predetermined timing, a predetermined time interval, or the like. According to such a configuration of the analysis system 1, the user 110 and the like can quickly and accurately analyze the state of the system.

As described above, the present invention is not limited to the above-described embodiment, and includes various modifications. For example, the above-described embodiment has been described in detail in order to explain the present invention in an easy-to-understand manner, and is not necessarily limited to the one including all the described configurations. Further, it is possible to replace a part of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add the configuration of another embodiment to the configuration of one embodiment. Further, it is possible to add / delete / replace a part of the configuration of each embodiment with another configuration.

For example, in the present embodiment, a relational database (DB) realized by software is given as an example of implementation of the distributed history data 150, but the relational database is used to describe the relationship between the data and is actually used. The data may be stored or managed in a file server, a key-value store, a time-series database, or the like.

Further, the characteristic value described in the present embodiment is not limited to the content described in the present embodiment, and any characteristic value may be used as long as it is a characteristic value calculated from the state data based on a calculation formula or a program.

Further, it is preferable that the analysis system 1 of the present embodiment mechanizes the parameter adjustment work conventionally performed by engineers in each specialized field and executes various parameter setting patterns in parallel to generate the analysis model 108. Therefore, it is desirable to have a distributed execution environment consisting of multiple physical servers and virtual servers. For example, each function of the system analyzer 120 described in the present embodiment may be realized by one or a plurality of information processing devices using a physical server, a virtual server, or a container. Here, the physical server is a computer system composed of a CPU, a memory, a bus, a storage device, an OS (Operating System), a network interface, and the like, and the virtual server is the above CPU, a memory, a bus, a storage device, an OS, and a network. By logically dividing the interface or the like by software or circuit-dividing using hardware, one server is operated by pretending to be a plurality of servers. A container is a technology that separates the process space on an OS and makes the applications on the OS appear to be running on individual OSs.

1 analysis system, 5 devices, 6 sensors, 108 analysis model, 120 system analyzer, 130 device classification section, 135 forward-posterior relationship estimation section, 142 analysis model generation section, 145 analysis model output section

Claims (8)

A system analyzer with a processor and memory
In a predetermined system including a plurality of devices, state data which is data indicating the state of the devices in time series is acquired for each of the plurality of devices, and based on the acquired state data, the plurality of devices are obtained. A device classification unit that classifies each of the devices according to the characteristics of changes in the state of the device into at least one group.
Based on the acquired state data, a predecessor relationship estimation unit that estimates the temporal predecessor relationship of the processing performed by each device in the group, and a predecessor relationship estimation unit.
An analysis model generation unit that generates an analysis model that is a model that predicts a temporal change in the state of each device in the group based on the classified group and the estimated time-preceding relationship.
It is provided with an analysis model output unit that outputs information indicating the contents of the generated analysis model .
The analysis model generation unit calculates a characteristic value indicating the time required to complete all the processes in the predetermined system based on the estimated time-preceding relationship, and the calculated characteristic value sets a predetermined threshold value. If it exceeds, the modified analytical model based on the state data of each device in the period before the period in which the fluctuation range of the state data is less than the predetermined value in the period related to the analysis model is added to the modified analysis model at a predetermined date and time. It is provided with a parameter optimization unit generated by specifying the characteristic value such that an index indicating the high degree of involvement of each group in the failure of the device is optimized.
System analyzer. An information input unit that accepts input of information indicating a predetermined state in the past of the device, and
A state prediction output unit that estimates the time it takes for the device to reach the predetermined state based on the input state information of the device and the generated analysis model, and outputs information about the estimated time.
The system analyzer according to claim 1. An operating device acquisition unit that acquires operation data, which is data indicating that the device is operating or stopped, and acquires the state data to be analyzed based on the acquired operation data and the acquired state data. With
The system analysis device according to claim 1, wherein the device classification unit classifies each of the plurality of devices into at least one or more groups based on the acquired state data of the analysis target. The analysis model generation unit analyzes the time change characteristics of the state data of each device in each group based on the estimated time-preceding relationship, and arranges each group in the order in which the analyzed time change characteristics appear. Based on the characteristic value indicating the temporal predecessor relationship of the processing between the devices, the modified analysis model shows the high degree of involvement of each group in the failure of the device at a predetermined date and time. A parameter optimization unit generated by specifying the characteristic value such that the index is optimized is provided.
The system analyzer according to claim 1. The analysis model generation unit said when the characteristic value representing the number of periods in which the degree of correlation with the failure of the device at a predetermined date and time was equal to or greater than a predetermined threshold value among the respective groups was equal to or greater than a predetermined value. An index showing the degree of involvement of each group in the failure of the device at a predetermined date and time by modifying the analysis model for a period excluding a predetermined period before and after each of the periods related to the analysis model. It is provided with a parameter optimization unit generated by specifying the characteristic value such that
The system analyzer according to claim 1. The analysis model generation unit determines the modified analysis model that is different before and after each event, which is specified based on the characteristic value indicating the difference in the characteristics of the state data in the predetermined period before and after each event. It is provided with a parameter optimization unit generated by specifying the characteristic value such that an index indicating the high degree of involvement of each group with respect to the failure of the device at the date and time is optimized.
The system analyzer according to claim 1. An information input unit that accepts input of information indicating a predetermined state in the past of the device, and
A state prediction output unit that estimates the time it takes for the device to reach the predetermined state based on the input state information of the device and the generated analysis model, and outputs information about the estimated time.
An operating device acquisition unit that acquires operation data, which is data indicating that the device is operating or stopped, and acquires the state data to be analyzed based on the acquired operation data and the acquired state data. And with more
The device classification unit classifies each of the plurality of devices into at least one or more groups based on the acquired state data of the analysis target .
The analysis model generation unit
It is calculated by analyzing the time change characteristics of the state data of each device in each group based on the estimated time-preceding relationship, and arranging each group in the order in which the analyzed time change characteristics appear. Based on the characteristic value indicating the time-preceding relationship of the processing between the devices, the modified analysis model is optimized so that the index indicating the high degree of involvement of each group in the failure of the device at a predetermined date and time is optimized. Generated by specifying the above characteristic value
Of the period related to the analysis model, when the characteristic value representing the number of periods in which the correlation degree with respect to the failure of the device at the predetermined date and time was equal to or higher than the predetermined threshold value among the above groups was equal to or higher than the predetermined value. , The analysis model modified for the period excluding the predetermined period before and after each said period is optimized so that the index showing the high degree of involvement of each said group in the failure of the device at a predetermined date and time is optimized. Generated by specifying the characteristic value,
The modified analytical model, which is different before and after each event specified based on the characteristic value indicating the difference in the characteristics of the state data in the predetermined period before and after each event, is applied to the failure of the device at a predetermined date and time. It is generated by specifying the characteristic value such that the index indicating the high degree of involvement of each said group is optimized.
Equipped with a parameter optimization unit
The system analyzer according to claim 1. It is a system analysis method
An information processing device equipped with a processor and memory
In a predetermined system including a plurality of devices, state data which is data indicating the state of the devices in time series is acquired for each of the plurality of devices, and based on the acquired state data, the plurality of devices are obtained. A device classification process for classifying each of the devices of the above into at least one or more groups according to the characteristics of the change in the state of the device.
Based on the acquired state data, the pre-post relationship estimation process for estimating the temporal pre-post relationship of the processing performed by each of the devices in the group, and the pre-post relationship estimation process.
An analytical model generation process that generates an analytical model that is a model that predicts a temporal change in the state of each device in the group based on the classified group and the estimated temporal predecessor relationship.
An analysis model output process that outputs information indicating the contents of the generated analysis model, and
The execution,
In the analysis model generation process, a characteristic value indicating the time required for all the processes in the predetermined system to be completed is calculated based on the estimated time-preceding relationship, and the calculated characteristic value sets a predetermined threshold value. If it exceeds, the modified analytical model based on the state data of each device in the period before the period in which the fluctuation range of the state data is less than the predetermined value in the period related to the analysis model is applied at a predetermined date and time. A parameter optimization process generated by specifying the characteristic value is executed so that an index indicating the high degree of involvement of each group in the failure of the device is optimized.
System analysis method.

Images