Abnormal modulation cause determination device, abnormal modulation cause determination method, and abnormal modulation cause determination program

Abstract

To improve the accuracy of determining the cause of abnormal modulation in production equipment. An abnormal modulation cause determination device includes: a process data acquisition section that reads process data from a storage device that stores process data continuously output by a plurality of sensors included in production equipment; an abnormality determination section that calculates an abnormality degree that indicates the degree of modulation of the process data read by the process data acquisition section; and a cause diagnosis section that uses cause-and-effect relationship information that defines combinations of causes and modulations as hierarchical along a time series, determines whether the abnormality degree calculated by the abnormality determination section satisfies a prescribed reference with respect to process data output by the plurality of sensors, wherein the modulation is a modulation of the process data output by the plurality of sensors that occurs as an effect of the cause.

Description

Technical Field

[0001] This disclosure relates to an apparatus, a method, and a procedure for determining the cause of abnormal modulation. Background Technology

[0002] Previously, techniques have been proposed for inferring the cause of anomalies using operational data received from the unit. For example, a technique has been proposed that infers the cause of an anomaly by weighting the probability of a first anomaly event that has occurred in the past by weighting the probability of a second anomaly event that has not occurred (Patent Document 1). Furthermore, a technique has been proposed that, upon detecting an anomaly in a process being diagnosed, infers a contribution rate representing the proportion of a process variable contributing to the state of the process being diagnosed, and infers events that may be contributing factors to the anomaly from a predefined list of registered events based on this contribution rate (Patent Document 2).

[0003] In addition, a technique is proposed as follows: using multiple sub-models to predict the state of the process to calculate the deviation index of the process from the normal state, and inferring the cause of the abnormal state occurring in the process based on the deviation index pattern set composed of the deviation indexes calculated by each sub-model (Patent Document 3). Furthermore, a technique is proposed as follows: for each of the multiple operating data constituting the monitoring operating data, the monitoring contribution relative to the monitoring anomaly degree is calculated; a diagnostic target data group composed of the top N operating data with the largest monitoring contribution is extracted; a conformity index is calculated between the diagnostic target data group and a diagnostic reference data group composed of the top M operating data with the largest reference contribution included in the reference operating data; and an abnormality of the unit is predicted based on the reference operating data whose conformity index is above the conformity judgment threshold (Patent Document 4).

[0004] Existing technical documents

[0005] Patent documents

[0006] Patent Document 1: Japanese Patent Application Publication No. 2018-109851

[0007] Patent Document 2: Japanese Patent Application Publication No. 2018-120343

[0008] Patent Document 3: Japanese Patent Application Publication No. 2019-16039

[0009] Patent Document 4: Japanese Patent Application Publication No. 2019-57164 Summary of the Invention

[0010] The problem the invention aims to solve

[0011] Generally, in production equipment, the ideal is to prevent abnormal modulation and suppress its impact on safety, stability, product quality, and cost. The purpose of this technology is to improve the ability to determine the causes of abnormal modulation in production equipment.

[0012] Technical solution

[0013] The abnormal modulation cause determination device includes: a process data acquisition unit that reads process data from a storage device that stores process data continuously output by multiple sensors provided by a production equipment; an anomaly determination unit that calculates an anomaly degree representing the degree of modulation of the process data read by the process data acquisition unit; and a cause diagnosis unit that uses causal relationship information that defines the combination of cause and modulation as a hierarchical structure along a time sequence to determine whether the anomaly degree calculated by the anomaly determination unit meets a predetermined benchmark for the process data output by the multiple sensors, wherein the modulation is the modulation of the process data output by the multiple sensors that occurs as an effect caused by the cause.

[0014] By defining the combination of causal relationships between the modulation of process data output from multiple sensors and the effects arising from these causal factors as hierarchical causal relationship information along a time sequence, the number of process data output from multiple sensors that meet a predetermined benchmark for anomaly level increases. This allows for the detection of anomalies based on certain causes. In other words, even if the number of process data output from multiple sensors that meet a predetermined benchmark for anomaly level is small, signs of anomalies based on certain causes can still be detected. Therefore, the performance in determining the cause of anomaly modulation in production equipment can be improved.

[0015] Alternatively, the anomaly determination unit can be configured to calculate the anomaly degree corresponding to whether the deviation from the specified benchmark is positive or negative. By comprehensively considering whether the deviation is in a positive or negative direction when calculating the anomaly degree, false judgments can be reduced compared to calculating the anomaly degree based solely on the magnitude of the deviation.

[0016] Alternatively, the anomaly determination unit can be configured to calculate, for at least a portion of the maximum and minimum values ​​of the process data, the corresponding anomaly degree, whether the direction of deviation from a specified reference is positive or negative. By comprehensively considering the direction of deviation from the reference for the anomaly degree at characteristic points in the continuously acquired process data, the anomaly sensing accuracy can be efficiently improved.

[0017] Alternatively, the causal relationship information can be configured such that, for each piece of process data output from multiple sensors, the process data used for anomaly calculation is defined based on the timing, period, or interval of the process performed by the production equipment. The anomaly determination unit uses the values ​​extracted from the process data read by the process data acquisition unit, based on the timing, period, or interval defined by the causal relationship information, to calculate the anomaly degree. For example, the timing, period, or interval can be defined based on a so-called knowledge base. If the values ​​used for anomaly calculation in the process data can be defined in detail, the accuracy of anomaly determination can be improved, and the system load can be reduced by cutting unnecessary calculations.

[0018] Alternatively, the anomaly detection unit can be configured to use a neural network model that compresses and reconstructs the process data output from multiple sensors included in a pre-determined combination to calculate the anomaly degree corresponding to the difference between the input and output of the neural network model. By having the so-called self-encoder pre-learn the relationship between the output values ​​of multiple sensors under normal conditions, an anomaly can be detected when it becomes impossible to reconstruct the data properly.

[0019] Alternatively, the system may also include a preprocessing unit that calculates the average, maximum, minimum, slope, or standard deviation over a specified period for the process data read by the process data acquisition unit. The anomaly determination unit then uses the values ​​calculated by the preprocessing unit to calculate the degree of anomaly. By using these values ​​representing the characteristics of the process data instead of the process data itself obtained from the production equipment, unnecessary values ​​can be reduced, or statistical values ​​that easily indicate a tendency towards anomalies can be obtained.

[0020] Furthermore, causal information can also be created by analyzing the causal relationship between causes and modulation through HAZOP (Hazard and Operability Study), FMEA (Failure Mode and Effect Analysis), FTA (Fault Tree Analysis), or ETA (Event Tree Analysis), or based on any of these analytical methods. Creating causal information by extracting all hypothetical causes of modulation enables improved accuracy in anomaly detection and cause determination, allowing for earlier anomaly detection and cause determination based on values ​​that change before the final modulation occurs.

[0021] It should be noted that the content described in the technical solution can be combined as much as possible without departing from the scope of the problem and technical concept of this disclosure. Furthermore, the content of the technical solution can be provided as a system including a computer or multiple devices, a method executed by a computer, or a program that causes a computer to execute. It should also be noted that it can be provided as a recording medium for storing the program.

[0022] Invention Effects

[0023] Based on the publicly available technology, the accuracy of determining the cause of abnormal modulation in production equipment can be improved. Attached Figure Description

[0024] Figure 1 This is a diagram illustrating an example of the system described in this embodiment.

[0025] Figure 2 This is a schematic diagram illustrating an example of a process performed by the machinery of the unit.

[0026] Figure 3 This is a diagram illustrating an example of process data in a batch process.

[0027] Figure 4 This is a diagram representing an example of a pre-defined process line definition table.

[0028] Figure 5 This is a diagram representing an example of a pre-defined label definition table.

[0029] Figure 6 This is a diagram used to illustrate an example of process data in a continuous process.

[0030] Figure 7 This is a diagram representing an example of traceability information.

[0031] Figure 8 It is a diagram used to illustrate the correspondence between process data in continuous processes and production numbers in batch processes.

[0032] Figure 9 This is a diagram representing an example of information pre-registered in a knowledge base.

[0033] Figure 10 This is a diagram of an example of a logic tree that represents the relationship between modulation and its cause.

[0034] Figure 11 This is a diagram used to illustrate the synchronous processing of process data.

[0035] Figure 12 This is a diagram used to illustrate an example of calculating anomalies based on distance from a baseline for time series data.

[0036] Figure 13 This is a diagram illustrating an example of calculating anomalies based on distance from a baseline, taking into account both positive and negative directions, for time-series data.

[0037] Figure 14 This is a diagram used to illustrate anomaly sensing using an autoencoder.

[0038] Figure 15 This is a block diagram illustrating an example of the configuration of a device for determining the cause of abnormal modulation.

[0039] Figure 16 This is a flowchart illustrating an example of the learning process performed by the abnormal modulation cause determination device.

[0040] Figure 17 This is a diagram representing an example of an action table.

[0041] Figure 18 This is a flowchart illustrating an example of the abnormal sensing processing performed by the abnormal modulation cause determination device.

[0042] Figure 19 This is an example diagram showing a screen being output to an input / output device.

[0043] Figure 20 These are other examples of images that are output to an input / output device. Detailed Implementation

[0044] Hereinafter, the implementation of the abnormal modulation cause determination device will be described with reference to the accompanying drawings.

[0045] <Implementation Method>

[0046] Figure 1 This diagram illustrates an example of the system described in this embodiment. System 100 includes an anomaly modulation cause determination device 1, a control station 2, and a unit 3. System 100 is, for example, a distributed control system (DCS), including multiple control stations 2. That is, the control system of the unit 3 is divided into multiple zones, and each control zone is distributedly controlled by the control stations 2. The control stations 2 are existing devices in the DCS, receiving status signals output from sensors or other components of the unit 3 or outputting control signals to the unit 3. Then, based on the control signals, the actuators of valves or other machines of the unit 3 are controlled.

[0047] The abnormal modulation cause determination device 1 acquires the status signal (process data) of the unit 3 via the control station 2. The process data includes temperature, pressure, flow rate, etc., of the raw materials and intermediate products being processed, or setpoints that determine the operating conditions of the machines in the unit 3. Furthermore, the abnormal modulation cause determination device 1 creates an anomaly sensing model based on a knowledge base that stores the correspondence between hypothetical causes and, for example, the effects that occur as an anomaly. For example, a model created based on the knowledge base is used to determine abnormal modulation, its precursors, and its causes by sensing changes in process data that deviate from allowable ranges. Then, the abnormal modulation cause determination device 1 can use the model and process data to sense the occurrence of abnormal modulation or its precursors. Alternatively, the abnormal modulation cause determination device 1 may determine, for example, candidate operating conditions for suppressing abnormal modulation based on a table storing causes of abnormal modulation and actions to be taken in response, and the determined causes, and then suggest these conditions to the user.

[0048] Figure 2 This is a schematic diagram illustrating an example of a process performed by the machines within the unit. In this embodiment, the process may include batch operations 31 and continuous operations 32. In batch operations 31, the processed objects are processed sequentially according to each defined processing unit, for example, the processes of receiving, holding, and discharging raw materials from each machine are performed in sequence. In continuous operations 32, the continuously introduced processed objects are processed continuously, for example, the processes of receiving, holding, and discharging raw materials are performed in parallel. Furthermore, the process may include multiple series 33 performing the same processing in parallel.

[0049] The machines performing the various processes include reactors, distillation units, heat exchangers, compressors, pumps, and tanks, which are connected via piping. Furthermore, sensors, valves, etc., are installed at designated locations on the machines and piping. Sensors may include thermometers, flow meters, pressure gauges, level meters, and concentration meters. These sensors monitor the operating status of each machine and output status signals. Additionally, the sensors in Unit 3 are equipped with "tags" attached as identification information for each sensor. That is, the type of process data can be determined based on the tags. Then, the anomaly modulation cause determination device 1 and the control station 2 manage the input and output signals to each machine based on the tags.

[0050] <Batch Process>

[0051] Figure 3 This is a diagram illustrating an example of process data in a batch process. Figure 3 The columns on the left represent Figure 2This is part of the batching process 31 shown. Specifically, the process includes a pulverizer 301, a hydrocyclone 302, a pretreatment unit 303, a precooler 304, and a reactor 305. Furthermore, these processes are categorized as a pretreatment process, a precooling process, and a reaction process. Figure 3 The column on the right represents an example of process data acquired in each process. In the pretreatment process, timing data is acquired from sensors labeled 001 and 002. In the precooling process, timing data is acquired from sensors labeled 003 and 004. In the reaction process, timing data is acquired from sensors labeled 005, 006, and 007. Furthermore, in the batch processing process, processing objects are intermittently processed and associated with production numbers (also known as "product numbers," "batch numbers," or "management numbers"). That is, the production number is identification information used to identify processing objects processed centrally in the batch processing process. For example... Figure 3 As shown, over time, data regarding the timing of subsequent processing objects corresponding to production numbers is obtained. In this embodiment, the steps are defined as the control station 2 managing production numbers and representing the processing stages in the subdivided processes that constitute the batch process. It should be noted that it can also be configured such that, when the steps are reset via the PLC (Programmable Logic Controller) in the unit 3 connected to the control station 2, the production number of the process data output from the control station 2 is appropriately adopted based on the timing of the communication between the control station 2 and the unit 3 (e.g., after switching steps in the PLC, after a set time has elapsed). Furthermore, it can also be configured such that the set time can be set for each manufacturing line, or for each subdivided process.

[0052] Figure 4 This diagram illustrates an example of a pre-defined process line definition table. The process line definition table registers production numbers for each series and process, defines the steps of each processing stage within each process, and specifies the types of objects processed in each process. The process line definition table can be a table in a database or a predefined file like a CSV file. Furthermore, the process line definition table is also pre-created by the user and read out by the anomaly modulation cause determination device 1.

[0053] The process line definition table includes attributes such as series, process, product number, step, and variety. The series column records identification information used to determine the series of processes. The process column records identification information representing the subdivided processes within a batch process. The product number column records the production number used to identify the processing object being processed centrally in the batch process. The step column records the timing definitions of multiple steps representing the processing stage within that process. The variety column records the category of the processing object.

[0054] Figure 5 This diagram illustrates an example of a pre-defined tag definition table. The tag definition table defines the timing for acquiring process data from the sensors corresponding to each tag. It should be noted that the tag definition table can be either a table in a database or a predefined file such as CSV (Comma Separated Values). Furthermore, the tag definition table is pre-created by the user and read by the abnormal modulation cause determination device 1.

[0055] The label definition table includes attributes such as label, series, process, and collection interval. The label column records the label as identification information for the sensor. The series column records identification information for the series used to determine the process. The process column records identification information for the subdivided processes within a batch process. The collection interval column records information indicating the interval at which the sensor's output values ​​are acquired.

[0056] <Continuous Process>

[0057] Figure 6 This is a diagram used to illustrate an example of process data in a continuous process. Figure 6 The columns on the left represent Figure 2 This is part of the continuous process 32 shown. Specifically, the process includes a reservoir 311 and a pump 312. Figure 6 The column to the right represents an example of process data acquired in each process. In continuous process 32, time-series data corresponding to tags but not to production numbers is continuously acquired from sensors. In continuous process, time-series data is acquired from sensors labeled 102 and 103. In continuous process, the machine continuously receives and processes objects.

[0058] In the case of a continuous process following a batch process, in order to associate the processed objects in the batch process with the processed objects in the continuous process, this embodiment uses traceability information pre-set by the user. Figure 7 This is a diagram illustrating an example of traceability information. Traceability information includes attributes such as sampling interval and dwell time. The sampling interval column records the intervals at which samples are taken for process inspection during consecutive processes, for example, using a reduction method. The dwell time column records the time a processed object remains from the completion of a batch process until it reaches the processes included in the subsequent process.

[0059] Figure 8This diagram illustrates the correspondence between process data in continuous processes and production numbers in batch processes. For example, process data is acquired at intervals set in the traceability information. Furthermore, when a continuous process follows a batch process, products obtained from the batch process completed within a specified period are introduced into a storage container or similar device as the processing object for the continuous process. Therefore, regarding the process data in the continuous process, the dwell time of the processed object from the completion of the batch process to the time of sensor measurement can be traced, establishing a correspondence between the completion time of the batch process and the production number group included in the specified period. By establishing this association, the accuracy of determining the cause of anomalies using process data from the batch process can be improved when batch and continuous processes are performed consecutively.

[0060] As mentioned above, by establishing a correspondence between the production number in batch processing and the measurement timing in continuous processes, the accuracy of determining the cause of the anomaly can be improved.

[0061] Figure 9 This diagram represents an example of information pre-registered in a knowledge base. The knowledge base is assumed to be pre-stored in the storage device of the abnormal modulation cause determination device 1. Figure 9 The table includes columns for "Impact" corresponding to each sensor (tag) and rows for "Assumed Cause" of modulation. That is, the column corresponding to the sensor affected by the "Cause 1," "Cause 2," etc., shown in each row registers the direction of the value change. In the knowledge base, the direction of change is displayed using "Up" to indicate an increase (rising) in the sensor's output value or "Down" to indicate a decrease (falling) in the sensor's output value. It should be noted that, as... Figure 9 As shown, the combination of cause and effect is not limited to a one-to-one relationship. Furthermore, the calculation method, extraction timing, and threshold for anomaly detection are determined corresponding to each sensor. The row entry for calculation method indicates information about the calculations performed on the output values ​​of each sensor. It should be noted that, in this embodiment, for example, machine learning methods such as Hotelling's method, k-nearest neighbor algorithm, DTW Barycenter Averaging, autoencoder, and graphical lasso are used for the calculations. The row entry for extraction timing indicates information about the timing for extracting values ​​from the output values ​​of each sensor used for anomaly detection. Regarding timing, for example, in batch processing, it could be defined by steps representing the processing stage in each process, a defined period, a time point, etc. Furthermore, in continuous processing, it could be defined by... Figure 7The sampling interval, as shown, is defined. Thresholds are registered in the threshold row as the benchmark for determining anomalies in each anomaly determination method. Thresholds include, for example, both an upper and lower limit. As described above, the knowledge base defines the combination of causal relationships between the event that becomes the cause and the influence of the modulation of the process data generated by it. Furthermore, the combination of causal relationships can be represented in a tree form, with the modulation that occurs as an influence as the root and its assumed cause as the leaf, connecting events occurring from the cause to the modulation in a hierarchical, time-series manner.

[0062] The knowledge base is pre-created by the user, for example, based on HAZOP (Hazard and Operability Study). HAZOP is used, for example, for methods such as: establishing and comprehensively listing the following based on sensing units at monitoring points of the equipment constituting the unit, management ranges (thresholds for upper and lower limits and alarm setpoints), deviations from the management ranges (abnormalities, modulations), enumeration of hypothetical causes of deviations from the management ranges, logic (sensing units) for determining which hypothetical cause caused the deviation, the effects caused by the deviation, the measures taken in the event of the deviation, and actions taken in response to those measures. It should be noted that, not limited to HAZOP, the knowledge base can also be created based on FTA (Fault Tree Analysis), FMEA (Failure Mode and Effect Analysis), ETA (Event Tree Analysis), or methods applying them, similar methods, content extracted from operator feedback, or content extracted from operational standards and technical standards. In this embodiment, abnormality sensing is performed based on parameters that are set to have causal relationships in the knowledge base.

[0063] Based on the information set in the table above, the abnormal modulation cause determination device 1 extracts the data of the specified timing from the process data obtained from the unit 3, and performs abnormal determination by a predetermined method. Figure 10 This is a diagram illustrating an example of a logic tree representing the relationship between modulation and its causes. It can be based on... Figure 9 The knowledge base shown is used to create the logic tree. Furthermore, Figure 10The logic tree arranges upstream and earlier-timing events in the production process on the left and downstream and later-timing events on the right, connected by arrows in a layered manner from the hypothetical cause towards the modulation that occurs as an effect. Furthermore, in the logic tree, if multiple hypothetical causes exist for a modulation in the knowledge base table, they are branched together, and events that occur together from the hypothetical cause to the modulation are displayed in bundles. The thick solid-line rectangles at the upstream ends of each branch correspond to the hypothetical causes in the knowledge base table. Figure 9 and Figure 10 The numbers within the parentheses are corresponding. Furthermore, the rectangles with thin solid lines represent the effects in the knowledge base table, indicating events observable through process data. For each such effect, calculations are performed according to the methods determined in the knowledge base table. Additionally, for each hypothetical cause, a model can be defined including formulas for performing the above calculations, used to sense anomalies or their precursors, or to assist in determining their causes.

[0064] <Calculation Method>

[0065] The aforementioned calculations can include, for example, the following methods. Alternatively, the abnormal modulation cause determination device 1 can be configured to display the results of these calculations.

[0066] ·Hotlinfa (T 2 Law)

[0067] For example, assuming multiple process data points obtained from a single sensor follow a prescribed probability density function, the population mean and standard deviation are estimated based on the sample mean and standard deviation calculated using the process data. The prescribed probability density function is, for example, a normal distribution. Then, the outlier is calculated based on the distance from the population mean to the process data of the verification object. For example, the outlier is determined based on the square of the Mahalanobis distance. It should be noted that either the instantaneous values ​​of the process data themselves can be used, or the maximum, minimum, cumulative, standard deviation, or differential coefficients (slopes) of the process data over a specified period can be used to calculate the outlier based on Hotelling's method. According to Hotelling's method, outliers relative to a prescribed benchmark can be detected.

[0068] k-nearest neighbor algorithm

[0069] For example, time-series process data obtained from more than one sensor can be vectorized or matrixed, and the distances between the data points can be calculated. These distances can be Euclidean, Mahalanobis, or Manhattan distances. Then, anomaly severity is determined based on the distance to the k-th nearest neighbor data point to the data being verified. In the k-nearest neighbor algorithm, this determination is based on relationships with other data points. Therefore, in cases where normal values ​​can be classified into multiple clusters, outliers that are far from any of these clusters can be detected.

[0070] • DTW (Dynamic Time Wrapping) - Barycenter Averaging

[0071] The average time-series data is calculated based on multiple time-series data, such as process data from different batch processes. For example, the distance to the average time-series data can be calculated separately for process data of different production numbers within the corresponding intervals of the batch process. Figure 11 This diagram illustrates the synchronized processing of process data. For each value of an element included in the time-series data processed in batches with different production numbers, the shortest distance between values ​​in different time-series data is calculated iteratively. The time-series data is then aligned by sliding along the time axis in a manner that minimizes the cumulative value of the shortest distance. In other words, multiple time-series data are synchronized based on their similarity. This allows multiple process data to be overlaid in a way that makes the steps in the process implemented in Unit 3 correspond in time. Then, based on the cumulative value of the distance between the synchronized time-series data, anomaly is calculated using the k-nearest neighbor algorithm and Hotelling's theory. According to DTWBarycenter Averaging, anomalies can be sensed based on the similarity between time-series data.

[0072] Alternatively, it can be set to calculate the outlier degree with an added positive or negative sign for deviations from a benchmark such as the average. Figure 12 This is a diagram used to illustrate an example of calculating anomalies based on the distance from a baseline for time series data. Figure 13 This is a diagram used to illustrate an example of calculating anomalies based on distance from a baseline, taking into account both positive and negative directions, for the same time series data. Figure 12 and Figure 13 In the diagram, the vertical axis represents, for example, the degree of deviation from the average value. The portion shown by the dashed rectangle actually exhibits modulation, but only according to... Figure 12 The values ​​shown in the examples are difficult to perceive. On the other hand, in Figure 13 In the example, it is shown that the positive and negative directions tend to deviate in opposite directions, thus making the sensing of modulation easy.

[0073] For example, in the Hotling method described above, the degree of deviation from the reference is determined without squaring the distance, and this is used as the value for adding a positive or negative sign. Thus, the image is calculated. Figure 13 Such anomalies. In DTW BarycenterAveraging et al., for characteristic points such as maxima in time series data, the sign of positive or negative is determined by, for example, the following formula, and the calculated value is multiplied by the distance.

[0074] Sign determinant = (μ-x) / |μ-x|

[0075] It should be noted that μ is the average value of the training data (benchmark value), and x is the process data of the validation object. Thus, according to the sign determination formula, the sign representing the direction of deviation from the benchmark at the specified time point can be determined based on the relationship between the benchmark value at the specified time point of the time series data and the corresponding process data of the validation object at that time point. Furthermore, by using the value of an additional sign indicating the degree of deviation from the benchmark, it is possible to calculate... Figure 13 Such anomalies can suppress false sensing. Alternatively, in addition to using maxima, characteristic points in time-series data can also use minima, the difference between process data at one time point and process data at another time point, etc. • Autoencoder

[0076] Figure 14 This diagram illustrates anomaly sensing using an autoencoder. In this method, anomaly detection is based on the characteristics of relationships between process data from multiple sensors. Specifically, a neural network is used, with process data itself—for example, continuously processed or batch processed as input data—served as teaching values ​​to create a model capable of compressing (encoding) and restoring (decoding) the input data. In the neural network, for example, the number of nodes in the input and output layers corresponds to the number of sensors, while the number of nodes in the intermediate layers is less than the number of sensors. Information input to the input layer is compressed in the intermediate layers and restored in the output layer. It should be noted that multiple intermediate layers can exist, and the connection structure between layers is not limited to full associativity. Then, normal process data is used as training data for learning processing to create a model that adjusts parameters to minimize the difference between the values ​​of the input and output layers. Furthermore, in the anomaly detection processing, the process data of the verification object is input, and the anomaly degree corresponding to the difference between the values ​​of the input and output layers is calculated. That is, when abnormal process data is input, the information compressed in the intermediate layer cannot be properly recovered in the output layer, and the difference between the values ​​of the input layer and the output layer becomes larger. Therefore, anomaly detection can be performed based on this difference. According to the autoencoder, anomalies can be detected based on the characteristics of the relationship between the output values ​​of multiple sensors.

[0077] • Graphic Lasso

[0078] For example, the covariance matrix of process data from multiple sensors in continuous and batch processing can be used to quantify the dependencies between variables, representing them as a sparse curve that serves as a baseline. Under normal conditions, the dependencies between variables can be identified as relationships that have not deviated significantly from the baseline. Then, in the anomaly detection process, the process data of the verification object is used to determine the dependencies between variables, and the anomaly degree corresponding to the magnitude of the difference from the aforementioned baseline is calculated. Based on the graphical lasso, the correlation between process data can be quantified, and the anomaly degree can be sensed based on the deviation of the relationship.

[0079] In addition, it can also be configured to use general anomaly sensing methods, or methods that apply these methods. Furthermore, regarding the thresholds used for anomaly sensing in each method, it can be configured to use process data actually obtained during the operation of Unit 3 to explore values ​​such as minimizing false judgments during normal operation, rapidly sensing the occurrence of anomalies during anomalies, and their precursors, and pre-register these values. Figure 9 The knowledge base shown.

[0080] <Device Composition>

[0081] Figure 15 This is a block diagram illustrating an example of the configuration of an abnormal modulation cause determination device 1. The abnormal modulation cause determination device 1 is a general-purpose computer, equipped with a communication interface (I / F) 11, a storage device 12, an input / output device 13, and a processor 14. The communication I / F 11 can be, for example, a network interface card (NIC) or a communication module, communicating with other computers based on a defined protocol. The storage device 12 can be a main storage device such as RAM (Random Access Memory) or ROM (Read Only Memory), or an auxiliary storage device (secondary storage device) such as HDD (Hard-Disk Drive), SSD (Solid State Drive), or flash memory. The main storage device temporarily stores programs read by the processor 14, information transmitted and received between other computers, or ensures the operating area of ​​the processor 14. The auxiliary storage device stores programs executed by the processor 14, information transmitted and received between other computers, etc. The input / output device 13 can be, for example, a keyboard, a mouse, or other input device, a monitor, or an input / output device such as a touch panel—a user interface. The processor 14 is a computing device such as a CPU (Central Processing Unit), which performs various processes in this embodiment by executing programs. Figure 15In the example, functional blocks are shown within the processor 14. That is, the processor 14 functions as a process data acquisition unit 141, a preprocessing unit 142, a learning processing unit 143, an anomaly determination unit 144, a cause diagnosis unit 145, and an output control unit 146 by executing a prescribed program.

[0082] The process data acquisition unit 141 acquires process data from sensors on the unit 3 via, for example, communication I / F 11 and control station 2, and stores it in storage device 12. As described above, the process data is associated with the sensors via tags.

[0083] During the creation of the anomaly sensing model, the preprocessing unit 142 processes the process data. For example, the preprocessing unit 142 associates the process data with the production number. That is, based on the aforementioned traceability information stored in advance in the storage device 12, the process data corresponding to the specified tag, system, and production number in batch processing is associated with the process data corresponding to the specified tag and output at a specified time in continuous processing. Furthermore, based on the set values ​​of tables such as knowledge bases, data for a specified period used for anomaly determination is extracted, and feature quantities corresponding to each method are calculated. It should be noted that, alternatively, during the learning process, the preprocessing unit 142 may perform data cleaning, excluding data from non-steady-state operation periods, data when anomalies occur, noise, and other outliers to extract training data.

[0084] The learning processing unit 143 creates, for example, an anomaly sensing model including one or more operations based on a knowledge base and stores it in the storage device 12. At this time, the learning processing unit 143 determines the parameters after learning the features of the training data. It should be noted that, alternatively, when using the output values ​​of multiple sensors for learning processing, appropriate normalization can be performed.

[0085] The anomaly determination unit 144 uses process data and anomaly sensing model to calculate the anomaly degree. That is, in the learning process, the anomaly determination unit 144 uses test data for cross-validation and the anomaly sensing model to calculate the anomaly degree. In addition, in the anomaly determination process, process data obtained from unit 3 is used to calculate the anomaly degree.

[0086] The cause diagnosis unit 145 uses the calculated anomaly degree to calculate the validity (accuracy) of each of the multiple hypothetical causes. Regarding validity, for example, the anomaly degree calculated by the anomaly determination unit is based on the proportion and degree of influence appearing in the process data among the influences associated with each hypothetical cause in the knowledge base. Alternatively, it can be configured to pre-store actions indicating countermeasures to be taken against the cause in correspondence with each hypothetical cause in the storage device 12, thereby prompting the user with the appropriate actions.

[0087] The output control unit 146, for example via the input / output device 13, issues an alarm or outputs the validity of each hypothetical cause when an anomaly is detected. The output control unit 146 appropriately connects the components described above via the bus 15 according to the user's operation. It should be noted that, for convenience, Figure 15 The apparatus shown includes a process data acquisition unit 141, a preprocessing unit 142, a learning processing unit 143, an anomaly determination unit 144, a cause diagnosis unit 145, and an output control unit 146. However, it may also be configured such that at least some of the functions are distributed among different devices.

[0088] <Learning Processing>

[0089] Figure 16 This is a flowchart illustrating an example of the learning process performed by the abnormal modulation cause determination device 1. The processor 14 of the abnormal modulation cause determination device 1 executes a predefined program to perform tasks such as... Figure 16 The processing is as shown. Regarding the learning process, it uses process data obtained from the past operation of Unit 3 and executes it at arbitrary time intervals. Furthermore, the learning process mainly includes preprocessing (…). Figure 16 The processes are: S1), model building (S2), and validation (S3). Alternatively, a portion of the process data can be used as training data, and the remainder as test data for cross-validation. It should be noted that the tables mentioned above are user-created and pre-stored in storage device 12. For convenience, in... Figure 16 The process flow shown describes preprocessing, learning, and verification processes, but it can also be configured such that at least some of these processes, such as preprocessing and verification, are distributed to different devices for execution.

[0090] The process data acquisition unit 141 of the abnormal modulation cause determination device 1 acquires process data. Figure 16 (S11). In this step, the image is extracted. Figure 3 , Figure 6 The process data shown is used for the anomaly sensing model. The process data is stored on storage device 12 as OPC (Object Linking and Embedding for Process Control) data, database tables, or files in a specified format such as CSV. Furthermore, the process data includes attributes such as date and time, tags, and, especially in batch-processed process data, may also include attributes such as product number and step.

[0091] Furthermore, the preprocessing unit 142 of the abnormal modulation cause determination device 1 establishes a correlation between the continuously processed process data and the production number. Figure 16(S12). In this step, as... Figure 8 As shown, a correspondence will be established between the process data acquired in continuous processes and the production number groups of the process data acquired in batch processing, and a correspondence will also be established between the process data used in the calculation of anomalies. That is, in Figure 9 The knowledge base shown Figure 11 In the logic tree shown, when a certain reason affects both batch processing process data and continuous processing process data, the abnormality and validity are calculated based on the data that have been correlated in this step.

[0092] Then, the preprocessing unit 142 extracts and processes the data used in the anomaly detection model. Figure 16 (S13). In this step, the preprocessing unit 142 extracts data for a specified period for anomaly determination based on the set values ​​of tables such as knowledge bases, and calculates the feature quantities corresponding to each method.

[0093] For example, when calculating the anomaly degree obtained by the Hotelling method, the preprocessing unit 142 extracts process data for a specified timing and period, calculates the instantaneous value, maximum value, minimum value, cumulative value or difference of the process data itself, the cumulative value of the reaction rate, the differential coefficient at a specified time point, etc., and stores them in the storage device 12. Furthermore, when calculating the anomaly degree obtained by the k-approximation method, the time-series process data is vectorized or matrixed. Furthermore, when calculating the anomaly degree obtained by DTWBarycenter Averaging, multiple process data are processed synchronously to obtain the average time-series data. Furthermore, when calculating the anomaly degree obtained by an autoencoder or graphic lasso, multiple process data are processed synchronously.

[0094] It should be noted that, alternatively, the preprocessing unit 142 may perform prescribed data cleaning on the process data. Data cleaning is the process of removing outliers, and various methods can be employed. For example, the most recent data can be used to calculate a moving average. Furthermore, the difference between the moving average and the measured value is taken to determine the standard deviation σ, representing the unevenness of the difference. Then, for example, values ​​that do not fall within a prescribed reliable interval, such as the interval from the mean of the probability distribution -3σ to the mean of the probability distribution +3σ (also called the 3σ interval), may be excluded. Similarly, values ​​that do not fall within the 3σ interval may be excluded based on the difference between consecutive measured values.

[0095] Subsequently, the learning processing unit 143 of the abnormal modulation cause determination device 1 performs abnormal sensing model construction processing. Figure 16 :S2). In this step, based on Figure 9 The knowledge base shown is used to create an anomaly detection model that includes anomaly degree calculations. Specifically, for... Figure 9For each "hypothetical cause," one or more corresponding "influences" are established. Anomalies obtained from the methods registered in the "operation method" are calculated, and an anomaly sensing model is created, showing a combination of anomalies. Furthermore, the learning processing unit 143 adjusts the model parameters using anomaly sensing methods and training data. For example, when calculating anomalies obtained from an autoencoder, the inter-layer weight coefficients are adjusted in a way that allows the information from the input process data to be recovered after compression. When calculating anomalies obtained from graphical lasso, the dependencies between variables are numerically represented based on the covariance matrix of process data from multiple sensors. Then, the learning processing unit 143 stores the created anomaly sensing model in the storage device 12.

[0096] The anomaly determination unit 144 of the anomaly modulation cause determination device 1 uses the created anomaly sensing model and test data to calculate the anomaly degree. Figure 16 (S31). In this step, the anomaly determination unit 144 calculates the anomaly degree according to the anomaly degree calculation method. For example, when calculating the anomaly degree obtained by the Hotelling method, the sample mean and sample standard deviation of the population are estimated using process data, and the anomaly degree is calculated based on the distance from the population mean to the process data of the verification object. When calculating the anomaly degree obtained by the k-nearest neighbor algorithm, the distance between data is calculated, and the anomaly degree corresponding to the distance to the k-th nearest data to the data of the verification object is calculated. When calculating the anomaly degree obtained by DTW BarycenterAveraging, the anomaly degree is calculated based on the cumulative value of the distance between time-series data synchronized in preprocessing, using the k-nearest neighbor algorithm and Hotelling theory. When calculating the anomaly degree obtained by the autoencoder, the process data of the verification object is input into the autoencoder, and the anomaly degree corresponding to the difference between the input layer value and the output layer value is calculated. When calculating the anomaly degree obtained by the graphical lasso method, the process data of the verification object is used to determine the dependencies between variables, and the anomaly degree corresponding to the magnitude of the difference between the dependencies that serve as the baseline is calculated.

[0097] The cause diagnosis unit 145 of the abnormal modulation cause determination device 1 uses the calculated abnormality degree to determine the validity of the assumed cause. Figure 16 (S32). In this step, for the assumed causes of each knowledge base, the validity is calculated based on the proportion of corresponding modulation occurrences established as influences. For example, Figure 9 The reason (2) corresponds to the three effects of the increase in moisture content of label 002, the increase in temperature 1 of label 004, and the decrease in temperature 2 of label 005. Alternatively, it can be used in... Figure 16In section S31, the anomaly score calculated for each impact is used to determine the validity score, which is the proportion of impacts whose anomaly scores exceed a threshold. For example, if two of the three impacts have anomalies exceeding the threshold, the validity score could be set to 66.7%. Furthermore, the validity score can be weighted based on the type of impact (label) or the magnitude of the anomaly score. For instance, the validity score could also be calculated by summing the weights of each impact.

[0098] Furthermore, the output control unit 146 outputs the anomaly calculated in S31 and the validity calculated in S32 for user evaluation of the created model. Figure 16 (S33). In this step, cross-validation is performed using test data that differs from the training data used to build the model, based on process data collected during the past operation of Unit 3. Furthermore, in this step, process data from past anomaly points are also used to appropriately sense anomalies and verify alarms, and whether actions to respond to them are output. Additionally, the learning processing unit 143 determines whether the anomaly was sensed with sufficient accuracy. Figure 16 S4). If the accuracy is deemed insufficient (S4: No), the threshold registered in the knowledge base (in other words, the normal range of process data) is corrected in a way that appropriately detects anomalies, and the processing after S31 is repeated. If S4 determines that the anomaly can be detected with sufficient accuracy (S4: Yes), the anomaly detection model and threshold created in S2 are applied. It should be noted that, alternatively, at least part of the judgment in S4 can be performed by the user.

[0099] It should be noted that, regarding actions, for example, it is set that, corresponding to the hypothetical cause, the actions that the operators of unit 3 should perform in order to deal with the hypothetical cause are pre-stored in the storage device 12. Figure 17 This is a diagram representing an example of an action table. Figure 17 The table includes attributes such as cause, action 1, and action 2. The cause column records the cause corresponding to the hypothetical cause in the knowledge base. The action 1 and action 2 columns record information indicating the actions that the operators of unit 3 should take to eliminate the corresponding cause.

[0100] <Anomaly Detection Processing>

[0101] Figure 18 This is a flowchart illustrating an example of the abnormal sensing processing performed by the abnormal modulation cause determination device 1. The processor 14 of the abnormal modulation cause determination device 1 executes a predetermined program to perform actions such as... Figure 18 The processing is as shown. Regarding anomaly sensing processing, it is performed approximately in real-time using process data obtained from the operation of unit 3. Anomaly sensing processing mainly includes preprocessing ( Figure 18 The processes include: S10, model readout processing (S20), and anomaly detection processing (S30). Figure 18 In the middle, to and Figure 16 The steps corresponding to the learning process shown are given the same reference numerals. Hereinafter, the description will focus on the differences from the learning process. For convenience, the process will be described as being implemented by the same device as the learning process, but the device for anomaly detection processing may be a different device from the learning process. Furthermore, it is assumed that the tables of anomaly detection models, thresholds, knowledge bases, etc., created in the learning process are pre-stored in the storage device 12.

[0102] The process data acquisition unit 141 of the abnormal modulation cause determination device 1 acquires process data. Figure 18 (S11). Process data is stored in storage device 12 as OPC data, tables in a database, or files in a specified format such as CSV. This step is related to... Figure 16 S11 is roughly the same, but data related to the operating process is acquired in unit 3. Furthermore, the preprocessing unit 142 of the abnormal modulation cause determination device 1 associates the continuously processed process data with the production number. Figure 18 (S12). This step is related to... Figure 16 Similarly, in S12. Then, the preprocessing unit 142 extracts and processes the data used in the anomaly detection model. Figure 18 (S13). This step is related to... Figure 16 The S13 is roughly the same, but no data cleaning is required.

[0103] Subsequently, the anomaly determination unit 144 of the anomaly modulation cause determination device 1 reads the anomaly sensing model created in the learning process from the storage device 12. Figure 18 (S20). Furthermore, the anomaly determination unit 144 uses the created anomaly sensing model and process data obtained from the operation of unit 3 to calculate the degree of anomaly. Figure 18 (S31). This step is related to... Figure 16 Similarly, in S31. Furthermore, the cause diagnosis unit 145 of the abnormal modulation cause determination device 1 uses the calculated abnormality degree to determine the validity of the assumed cause. Figure 18 (S32). This step is related to... Figure 16 The same applies to the S32.

[0104] Furthermore, the output control unit 146 outputs the anomaly level calculated in S31 and the validity level calculated in S32, and issues an alarm if either anomaly level exceeds a predetermined threshold. Figure 18(S303). In this step, process data, abnormality, and the validity of assumed causes indicating the operating status of unit 3 are provided to the user through input / output device 13.

[0105] Figure 19 This is a diagram showing an example of a screen that is output to the input / output device 13. Figure 19 As an example of a main management chart, a line graph is used to illustrate the progression of individual process data. Area 131 of the input / output device 13 displays a combination of identification information and the latest values ​​for multiple process data acquired from unit 3. In the management chart of area 132, for specific process data, a line graph is used to illustrate the progression of values. It should be noted that the vertical axis represents the value of the process data, and the horizontal axis represents the time axis. Furthermore, in... Figure 19 In this example, the solid line represents the true value, and the dashed line represents the estimated value. It should be noted that the true value is the process data of the object being calculated for anomaly, while the estimated value can be obtained through regression analysis of the process data of the object being calculated for anomaly. The thin dashed line represents the upper and lower limits of the normal range (in other words, the threshold used for anomaly sensing). It should also be noted that it can be set like... Figure 19 As circled in the diagram, when the user operates the input / output device 13, such as a point-and-click device, to move the pointer on the graph, the numerical value of the process data at the time indicated by the pointer is displayed. In the cause-effect graph of region 133, the horizontal axis displays the cause of the modulation of the process data displayed in region 132, or the label that can identify it, and the vertical axis uses a bar chart to represent the validity of the cause. It indicates that the higher the validity, the more likely it is to be the cause of the modulation of the process data. In addition, regarding the validity, the cause diagnosis unit 145 calculates the anomaly degree based on the anomaly determination unit 144 for events that are set as the hypothetical cause of the modulation of the process data. The user can identify the candidate cause of the modulation and its accuracy based on the magnitude of the validity, and can easily determine the cause of the modulation. Furthermore, the cause-effect graph is set as a graph in which the anomaly determination unit 144 calculates the anomaly degree at a specified time or the current time and displays it through the output control unit 146 when the "Diagnosis" button in region 134 is pressed. Then, when the user operates the input / output device 13 and selects any bar in the bar chart of the cause-effect diagram, the reason for the modulation corresponding to the bar will be highlighted in the logic tree.

[0106] Figure 20 This is a diagram showing another example of a screen output from the output control unit 146 to the input / output device 13. Figure 20 Here is an example of a tree diagram, showing something like... Figure 10 A logic tree like the one shown. For example, in Figure 19When the bar chart for label 004 is selected, the logic tree will emphasize the impact corresponding to the process data for label 004. This emphasis is achieved, for example, through changes in display format such as color or line type. Figure 20 In this context, shading is applied to the corresponding rectangles. Furthermore, the thick-lined rectangles connected to the upstream side of the logic tree represent the assumed cause of the influence. This can be set up as shown in... Figure 20 The reasons are displayed as circled in the diagram, or it can be set to display the impact on process data other than the reasons. It should be noted that it can also display... Figure 18 The validity of each cause calculated in S32 can also be displayed, or the action can be shown. Alternatively, it can be set to display the cause when the user moves the pointer over each rectangle.

[0107] It should be noted that after pressing... Figure 19 , Figure 20 When the "Trend" button is displayed, the process trend of each label shown in the cause-effect diagram can be displayed, or the process trend of the label that can determine the cause of the modulation can be specifically displayed. The process trend is set to use the process data stored in the storage device 12 to calculate the value for each period, such as every specified time, every specified number of days, every specified number of months, or every season, and mark it on the graph.

[0108] Alternatively, the output control unit 146 can be configured to output an anomaly log when, for example, the anomaly level calculated by each calculation method exceeds a predetermined threshold. Furthermore, it can also be configured to output logs of assumed causes and their validity. Each log is output by associating date and time, production number, calculation method, anomaly sensing model, etc., facilitating the analysis of anomaly modulation.

[0109] <Variation Example>

[0110] The various components and combinations in each embodiment are merely examples, and appropriate additions, omissions, substitutions, and other modifications to the components can be made without departing from the spirit of the invention. This disclosure is not limited to the embodiments, but only to the claims. Furthermore, the various solutions disclosed in this specification can also be combined with any other features disclosed in this specification.

[0111] Furthermore, while the above embodiments have been described using a chemical unit as an example, they can be applied to manufacturing processes in general production equipment. For example, batch numbers could be used as processing units, and the batch processing steps described in the embodiments could be applied instead of the production numbers for the batch processes in the embodiments.

[0112] Alternatively, at least a portion of the function of the abnormal modulation cause determination device 1 can be implemented in a manner distributed among multiple devices, or multiple devices can provide the same function in parallel. Furthermore, at least a portion of the function of the abnormal modulation cause determination device 1 can be located in a so-called cloud.

[0113] Furthermore, this disclosure includes a method for performing the above-described processing, a computer program, and a computer-readable recording medium on which the program is recorded. The recording medium on which the program is recorded can perform the above-described processing by causing a computer to execute the program.

[0114] Here, computer-readable recording media refers to recording media that can store information such as data and programs and be read from a computer through electrical, magnetic, optical, mechanical, or chemical action. Examples of such recording media that can be removed from a computer include floppy disks, optical disks, optical discs, magnetic tapes, and memory cards. Furthermore, recording media that are fixed to a computer include HDDs, SSDs (Solid State Drives), and ROMs.

[0115] Explanation of reference numerals in the attached figures

[0116] 1: Device for determining the cause of abnormal modulation;

[0117] 11: Communication I / F;

[0118] 12: Storage device;

[0119] 13: Input / output devices;

[0120] 14: Processor;

[0121] 141: Process Data Acquisition Department;

[0122] 142: Pre-processing section;

[0123] 143: Learning Processing Department;

[0124] 144: Anomaly Detection Department;

[0125] 145: Cause Diagnosis Department;

[0126] 146: Output control unit;

[0127] 2: Control station;

[0128] 3: Generating unit.

Claims

1. An abnormality-modulation-cause determining apparatus, comprising: a process data acquisition section that reads out process data that is continuously output by a plurality of sensors provided in a production facility from a storage device that stores the process data; an abnormality determination section that calculates an abnormality degree that indicates a degree of modulation of the process data read out by the process data acquisition section; the modulation being modulation of the process data output by the plurality of sensors that occurs as an effect of a cause, the cause being a factor that produces an effect, the cause-and-effect relationship information connecting events that occur in a process from a cause in a process on an upstream side to modulation in a process on a downstream side in production processes performed by the production facility in a hierarchical manner along a time series, the process data used for calculation of the abnormality degree being defined in accordance with timing in the production processes.

2. The abnormality-modulation-cause determining apparatus according to claim 1, wherein the timing includes a period or an interval.

3. The abnormality-modulation-cause determining apparatus according to claim 1, wherein the abnormality degree determination section calculates an abnormality degree corresponding to which of positive or negative a direction of deviation from a prescribed reference. The cause diagnosing section judges whether the abnormality degree calculated by the abnormality judging section satisfies a prescribed reference with respect to the process data output from the plurality of sensors, using the cause-modulated causal relationship information defined as a stratified causal relationship information along a time series, wherein 4. The abnormality-modulation-cause determining apparatus according to claim 3, wherein the abnormality determination section calculates an abnormality degree corresponding to which of positive or negative a direction of deviation from a prescribed reference with respect to at least some of maximum values and minimum values of the process data.

5. The abnormality-modulation-cause determining apparatus according to any one of claims 1 to 4, wherein the abnormality determination section performs calculation of an abnormality degree using values extracted from the process data read out by the process data acquisition section based on timing defined based on the cause-and-effect relationship information.

6. The abnormality-modulation-cause determining apparatus according to any one of claims 1 to 4, wherein the abnormality determination section calculates an abnormality degree corresponding to a difference between input and output of a neural network model that compresses and restores values of process data output by a plurality of sensors included in a combination decided in advance, using the neural network model.

7. The abnormality-modulation-cause determining apparatus according to any one of claims 1 to 4, further comprising: a preprocessing section that calculates an average value, a maximum value, a minimum value, a slope, or a standard deviation for a prescribed period with respect to process data read out by the process data acquisition section, the abnormality determination section calculating the abnormality degree using values calculated by the preprocessing section.

8. The abnormality-modulation-cause determining apparatus according to any one of claims 1 to 4, wherein the cause-and-effect relationship information is created by analyzing the cause-and-effect relationship of the cause and the modulation by any one of a hazard and operability analysis (HAZOP), a failure mode and effects analysis (FMEA), a fault tree analysis (FTA), or an event tree analysis (ETA).

9. An abnormality-modulation-cause determining method, comprising: reading out process data that is continuously output by a plurality of sensors provided in a production facility from a storage device that stores the process data, calculating an abnormality degree that indicates a degree of modulation of the read-out process data, ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ using a combination of cause-and-modulation causal relationship defined as layered causal relationship information along a time sequence, determining whether the calculated abnormality degree satisfies a prescribed reference with respect to the process data output from the plurality of sensors, wherein the modulation is a modulation of the process data output by the plurality of sensors that occurs as an effect resulting from the cause, the causal relationship information connects events occurring in a process from a cause in an upstream process to a modulation in a downstream process in the production process performed by the production equipment in a stratified manner along a time series, the process data used for the calculation of the abnormality degree among the process data output continuously is defined according to a timing in the production process.

10. The abnormality modulation cause determination method according to claim 9, wherein the timing includes a period or an interval.

11. An abnormality modulation cause determination program product for causing a computer to execute the following processing: reading out process data output continuously by a plurality of sensors possessed by a production equipment from a storage device that stores the process data, calculating an abnormality degree that represents a degree of modulation of the read-out process data, using a combination of cause-and-modulation causal relationship defined as layered causal relationship information along a time sequence, determining whether the calculated abnormality degree satisfies a prescribed reference with respect to the process data output from the plurality of sensors, wherein the modulation is a modulation of the process data output by the plurality of sensors that occurs as an effect resulting from the cause, the causal relationship information connects events occurring in a process from a cause in an upstream process to a modulation in a downstream process in the production process performed by the production equipment in a stratified manner along a time series, the process data used for the calculation of the abnormality degree among the process data output continuously is defined according to a timing in the production process.

12. The abnormality modulation cause determination program product according to claim 11, wherein the timing includes a period or an interval.

Images