WO2022091639A1 - 異常診断モデルの構築方法、異常診断方法、異常診断モデルの構築装置および異常診断装置 - Google Patents
異常診断モデルの構築方法、異常診断方法、異常診断モデルの構築装置および異常診断装置 Download PDFInfo
- Publication number
- WO2022091639A1 WO2022091639A1 PCT/JP2021/034596 JP2021034596W WO2022091639A1 WO 2022091639 A1 WO2022091639 A1 WO 2022091639A1 JP 2021034596 W JP2021034596 W JP 2021034596W WO 2022091639 A1 WO2022091639 A1 WO 2022091639A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- model
- abnormality
- regression
- regression model
- abnormality diagnosis
- Prior art date
Links
- 230000005856 abnormality Effects 0.000 title claims abstract description 179
- 238000010276 construction Methods 0.000 title claims abstract description 35
- 238000000034 method Methods 0.000 title claims description 55
- 238000003745 diagnosis Methods 0.000 claims description 117
- 238000012937 correction Methods 0.000 claims description 37
- 230000035945 sensitivity Effects 0.000 claims description 27
- 238000012549 training Methods 0.000 claims description 18
- 238000004364 calculation method Methods 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 13
- 230000002159 abnormal effect Effects 0.000 claims description 6
- 230000002547 anomalous effect Effects 0.000 claims 1
- 230000001419 dependent effect Effects 0.000 abstract 2
- 238000013145 classification model Methods 0.000 description 42
- 238000004519 manufacturing process Methods 0.000 description 23
- 239000000047 product Substances 0.000 description 12
- 238000003860 storage Methods 0.000 description 12
- 239000011159 matrix material Substances 0.000 description 10
- 238000013459 approach Methods 0.000 description 9
- 229910000831 Steel Inorganic materials 0.000 description 6
- 230000006870 function Effects 0.000 description 6
- 239000010959 steel Substances 0.000 description 6
- 238000002405 diagnostic procedure Methods 0.000 description 5
- 238000009826 distribution Methods 0.000 description 3
- 230000011218 segmentation Effects 0.000 description 3
- XEEYBQQBJWHFJM-UHFFFAOYSA-N Iron Chemical compound [Fe] XEEYBQQBJWHFJM-UHFFFAOYSA-N 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000007704 transition Effects 0.000 description 2
- 230000032683 aging Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 239000006227 byproduct Substances 0.000 description 1
- 239000000571 coke Substances 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 229910052742 iron Inorganic materials 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 239000005445 natural material Substances 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 239000002994 raw material Substances 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 239000002436 steel type Substances 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0259—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterized by the response to fault detection
- G05B23/0267—Fault communication, e.g. human machine interface [HMI]
- G05B23/027—Alarm generation, e.g. communication protocol; Forms of alarm
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
- G05B23/0243—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults model based detection method, e.g. first-principles knowledge model
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
- G05B23/0224—Process history based detection method, e.g. whereby history implies the availability of large amounts of data
- G05B23/024—Quantitative history assessment, e.g. mathematical relationships between available data; Functions therefor; Principal component analysis [PCA]; Partial least square [PLS]; Statistical classifiers, e.g. Bayesian networks, linear regression or correlation analysis; Neural networks
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0259—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterized by the response to fault detection
- G05B23/0262—Confirmation of fault detection, e.g. extra checks to confirm that a failure has indeed occurred
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06395—Quality analysis or management
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/04—Manufacturing
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/02—Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]
Definitions
- the present invention relates to an abnormality diagnosis model construction method, an abnormality diagnosis method, an abnormality diagnosis model construction device, and an abnormality diagnosis device.
- the model-based approach is an approach in which a model expressing a physical or chemical phenomenon in a manufacturing process is constructed by a mathematical formula, and the manufacturing state of the manufacturing process is diagnosed using the constructed model.
- the database approach is an approach in which a statistical analysis model is constructed from the operation data obtained in the manufacturing process, and the manufacturing state of the manufacturing process is diagnosed using the constructed model.
- Patent Documents 1 to 4 describe a method of predicting or detecting an abnormal state of a manufacturing process based on a prediction by a model created using normal operation data. Further, in Patent Documents 3 and 4, patterns are extracted from normal operation data and made into a library, and the difference between the acquired operation data and the library pattern is determined to detect a situation different from usual at an early stage. How to do it is described.
- the diagnostic accuracy can be improved by classification only when there is a sufficient number of training samples (number of training data), and it is difficult to construct an appropriate abnormality diagnosis model for rare varieties with a small production ratio. .. Further, even when the abnormality diagnosis model can be constructed, there is a problem that the sensitivity of the abnormality diagnosis model to the actual abnormality is greatly varied due to the small number of learning samples.
- the present invention has been made in view of the above, and is an appropriate classified abnormality diagnosis that ensures diagnostic accuracy even in a rare category with a small number of learning samples and has a small variation in sensitivity to an actual abnormality. It is an object of the present invention to provide an abnormality diagnosis model construction method, an abnormality diagnosis method, an abnormality diagnosis model construction device, and an abnormality diagnosis device capable of constructing a model.
- the method for constructing an abnormality diagnosis model is a method for constructing an abnormality diagnosis model for diagnosing an abnormality in a process, and all of them are collected in advance at normal times.
- a first regression model creation step for creating a first regression model in which the regression coefficient for an explanatory variable having a small influence on the objective variable is 0 using the operation data, and a plurality of predetermined operation data.
- the explanatory variable candidate determination step for determining the explanatory variable candidates within the range of the explanatory variables used in the first regression model, and the operation data included in the category, for each of the categories, It includes a second regression model creation step of creating a second regression model such that the regression coefficient for the explanatory variable candidate having a small influence on the objective variable is 0.
- the method for constructing the abnormality diagnosis model according to the present invention includes the variation index of the prediction error of the first regression model and the second for each category after the second regression model creation step.
- the correction coefficient calculation step for calculating the correction coefficient which is the ratio to the variation index of the prediction error of the regression model, and the prediction error of the second regression model for each category are corrected by the correction coefficient. Further includes a sensitivity correction step to be performed.
- the method for constructing the abnormality diagnosis model according to the present invention is the method for constructing the second regression model for each category constructed from the first training data set after the second regression model creation step in the above invention.
- Correction coefficient which is the ratio of the prediction error variation index to the prediction error variation index of the second regression model for each category constructed from the second training data set different from the first training data set. Further includes a correction coefficient calculation step for each category, and a sensitivity correction step for correcting the prediction error of the second regression model for each category by the correction coefficient.
- the plurality of categories include the product type, product size, operating conditions and operating pattern.
- the first regression model creation step creates the first regression model by Lasso regression
- the second regression model creation step creates the Lasso regression.
- the abnormality diagnosis method is an abnormality diagnosis method using the abnormality diagnosis model constructed by the above-mentioned abnormality diagnosis model construction method, and is a diagnosis target. It includes an abnormality diagnosis step for calculating an abnormality index using a second regression model according to the classification of the operation data to be used, and an abnormality determination step for determining the presence or absence of an abnormality based on the abnormality index.
- the abnormality diagnosis model construction device is an abnormality diagnosis model construction device for diagnosing process abnormalities, and all of them are collected in advance at normal times.
- a first regression model creation means for creating a first regression model in which the regression coefficient for an explanatory variable having a small influence on the objective variable is 0 using the operation data, and a plurality of predetermined operation data. It is divided into categories, and for each category, the explanatory variable candidate determination means for determining the explanatory variable candidates within the range of the explanatory variables used in the first regression model and the operation data included in the category are used.
- a second regression model creating means for creating a second regression model such that the regression coefficient for the explanatory variable candidate having a small influence on the objective variable becomes 0 is provided.
- the abnormality diagnosis device is an abnormality diagnosis device using the abnormality diagnosis model constructed by the above-mentioned abnormality diagnosis model construction device, and is a diagnosis target. It is provided with an abnormality diagnosing means for calculating an abnormality index by using a second regression model according to the classification of the operation data, and an abnormality determining means for determining the presence or absence of an abnormality based on the abnormality index.
- the explanatory variables within the range of the explanatory variables used in the regression model of all categories are used. Create a regression model for each segment.
- FIG. 1 is a diagram showing a schematic configuration of an abnormality diagnosis device and a model building device according to an embodiment of the present invention.
- FIG. 2 is a flowchart showing a procedure of a model building method executed by the model building apparatus according to the embodiment of the present invention.
- FIG. 3 is a flowchart showing a procedure of an abnormality diagnosis method executed by the abnormality diagnosis apparatus according to the embodiment of the present invention.
- model construction device the configuration of the abnormality diagnosis device and the abnormality diagnosis model construction device (hereinafter referred to as “model construction device”) according to the embodiment of the present invention will be described with reference to FIG.
- the abnormality diagnosis device is for diagnosing a process abnormality in a plant or the like
- the model construction device is for constructing a model for abnormality diagnosis.
- the abnormality diagnosis device 1 includes an input unit 10, a storage unit 20, a calculation unit 30, and a display unit 40.
- the "model construction device” is realized by the components excluding the diagnosis data history 24, the abnormality determination history 25, the abnormality diagnosis unit 37, and the abnormality determination unit 38 among the components of the abnormality diagnosis device 1.
- the input unit 10 is an input means for the calculation unit 30, receives operation data (for example, sensor data) of the equipment to be diagnosed via the information / control system network, and inputs the operation data to the calculation unit 30 in a predetermined format.
- operation data for example, sensor data
- the storage unit 20 is composed of a recording medium such as an EPROM (Erasable Programmable ROM), a hard disk drive (Hard Disk Drive: HDD), and a removable medium.
- a recording medium such as an EPROM (Erasable Programmable ROM), a hard disk drive (Hard Disk Drive: HDD), and a removable medium.
- removable media include disc recording media such as USB (Universal Serial Bus) memory, CD (Compact Disc), DVD (Digital Versatile Disc), and BD (Blu-ray (registered trademark) Disc).
- the storage unit 20 can store an operating system (Operating System: OS), various programs, various tables, various databases, and the like.
- OS Operating System
- the storage unit 20 stores the actual data history 21, the all-division model (first regression model) 22, the segmentation model (second regression model) 23, the diagnostic data history 24, and the abnormality determination history 25.
- the actual data history 21 is information related to the operation data (actual data) collected by the system.
- the all-division model 22 is a model created by the all-division model creation unit 32 based on the operation data.
- the classification model 23 is a model created by the classification model creation unit 35 based on actual data.
- the abnormality diagnosis in the abnormality diagnosis unit 37 is carried out using this classification model 23. Further, as the classification model 23, for example, a model that physically predicts the state quantity of the process and the like, a model that is statistically constructed, and the like can be mentioned.
- the diagnosis data history 24 is information regarding the abnormality diagnosis result by the abnormality diagnosis unit 37.
- the abnormality determination history 25 is information regarding the abnormality determination result by the abnormality determination unit 38.
- the storage unit 20 also stores various settings and the like necessary for operating the system as needed.
- the arithmetic unit 30 is realized by, for example, a processor including a CPU (Central Processing Unit) and a memory (main storage unit) including a RAM (Random Access Memory) and a ROM (Read Only Memory).
- a processor including a CPU (Central Processing Unit) and a memory (main storage unit) including a RAM (Random Access Memory) and a ROM (Read Only Memory).
- the arithmetic unit 30 loads the program into the work area of the main storage unit and executes it, and controls each component unit or the like through the execution of the program to realize a function that meets a predetermined purpose.
- the calculation unit 30 includes an all-division learning data creation unit 31, an all-division model creation unit (first regression model creation means) 32, a modeling matrix creation unit 33, and a division learning data creation unit 34. Functions as. Further, through the execution of the above-mentioned program, the calculation unit 30 includes a classification model creation unit (second regression model creation means) 35, a classification model sensitivity correction unit 36, an abnormality diagnosis unit (abnormality diagnosis means) 37, and an abnormality determination.
- FIG. 1 shows an example in which the functions of each part are realized by, for example, one computer, but the means for realizing the functions of each part is not particularly limited, and even if the functions of each part are realized by a plurality of computers, for example. good.
- the all-division learning data creation unit 31 creates learning data for constructing the all-division model 22.
- the all-division learning data creation unit 31 refers to the operation data necessary for model construction from the actual data history 21, normalizes it to a format suitable for learning data, and uses it as all-division learning data.
- the all-division model creation unit 32 creates the all-division model 22 based on the all-division learning data created by the all-division learning data creation unit 31.
- the all-division model creation unit 32 creates an all-division model 22 such that the regression coefficient for the explanatory variable having a small influence on the objective variable is 0 by using all the operation data in the normal state collected in advance. Then, the all-division model creation unit 32 stores the created all-division model 22 in the storage unit 20.
- ⁇ is a hyperparameter that specifies the weights of the first term and the second term in the weighted sum, and can be searched by changing it within a fixed value or a predetermined range.
- the demanding feature of the Lasso regression is derived by the second term of the above equation (1). That is, the regression coefficient ai , which has a small influence on the prediction error, has an effect of sparsification, which is positively set to 0. This effect makes it possible to extract the selection of explanatory variables and their degree of influence (impact coefficient) together.
- the modeling matrix creation unit 33 creates data (modeling matrix) for defining the structure of the division model 23 described later based on the structure of the all division model 22 created by the all division model creation unit 32. .. Specifically, the modeling matrix creation unit 33 creates data related to “ ⁇ i ” in the equation (3) described later.
- the segmented learning data creation unit 34 creates learning data for constructing the segmented model 23.
- the division learning data creation unit 34 refers to the operation data necessary for model construction from the actual data history 21, and divides the operation data into a plurality of predetermined divisions. Examples of the plurality of categories include the type of product to be manufactured (steel type in the case of steel products), the model of the product, the size of the product, the operating conditions, the operating pattern, and the like.
- the classification learning data creation unit 34 may perform classification based on product groups and the like that can be classified based on physical knowledge and the like.
- the division model creation unit 35 creates the division model 23 based on the division learning data created by the division learning data creation unit 34 and the modeling matrix created by the modeling matrix creation unit 33.
- the classification model creation unit 35 creates a classification model 23 such that the regression coefficient for the explanatory variable candidate having a small influence on the objective variable is 0 by using the operation data included in the above-mentioned classification. Then, the division model creation unit 35 stores the created division model 23 in the storage unit 20.
- the division model creation unit 35 In the creation of the division model 23 by the division model creation unit 35, important explanatory variables are converted into the structure of the all division model 22 created by all division learning data having a large N number (number of training samples (learning data number)). Limit based on. Then, among the limited explanatory variables, the explanatory variables are selected and the regression coefficients are determined by the segmented learning data. That is, the division model creation unit 35 determines the explanatory variable candidates used in the division model 23 within the range of the explanatory variables used in the all division model 22. As a result, the prediction accuracy of the model can be ensured by reducing the number of variables to be determined even in the divided learning data having a small number of N.
- the parameter search of the division model 23 by the division model creation unit 35 can be expressed by, for example, the following equation (2).
- y is the objective variable
- ⁇ i is a coefficient for setting the upper limit (explanatory variable candidate) of the explanatory variables obtained from the selection of the explanatory variables of the all-division model 22
- wi is the division model 23.
- the regression coefficient, xi is an explanatory variable.
- the ⁇ i in the above equation (2) is 1 for the variable selected as the explanatory variable in the all-partition model 22, and 0 otherwise. ..
- the regression model (second regression model) determined by the above equation (2) is further limited to the explanatory variables with the explanatory variables of the regression model (first regression model) determined by the above equation (1) as the upper limit. (In some cases, the same explanatory variable). Further, the regression coefficient (impact degree, impact coefficient) wi of the explanatory variable selected by the regression model determined by the above equation (2) is determined only from the segmented learning data.
- the classification model sensitivity correction unit 36 corrects the sensitivity of the classification model 23 created by the classification model creation unit 35.
- anomaly index a method of calculating the anomaly index using the distribution of the prediction error at the normal time is often used. ..
- the distribution of the prediction error deviates from the true distribution due to the shortage of the number of N. It has the potential to do. This means that the sensitivity of the calculated anomaly index fluctuates.
- the classification model sensitivity correction unit 36 corrects the sensitivity of the classification model 23 to unify the scale of the abnormality index among the classification models.
- the classification model sensitivity correction unit 36 specifically, as shown in the following equation (4), has a correction coefficient. ⁇ is calculated for each division of the division model 23. This correction coefficient ⁇ is a ratio between the variation index of the prediction error of the all-division model 22 (for example, variance, standard deviation) and the variation index of the prediction error of the classification model 23.
- the denominator is the variance of the prediction error of the classification model 22
- the numerator is the variance of the prediction error of the classification model 23.
- the same concept as in the above equation (4) can be applied even when the model sensitivity before and after the change is corrected.
- the correction coefficient ⁇ can be calculated from the ratio in which the variation index of the prediction error of the past classification model 23 is used as the denominator and the variation index of the prediction error of the newly introduced classification model 23 is used as the numerator.
- the first training data set and the second training data set are different training data sets.
- the first learning data set is, for example, a learning data set used when constructing the past segmentation model 23, and the second learning data set is, for example, when constructing the newly introduced segmentation model 23. It is a training data set used for.
- the classification model sensitivity correction unit 36 corrects the prediction error of the classification model 23 for each classification by the correction coefficient ⁇ .
- the variation in sensitivity between the divisions can be varied by matching the scale of the abnormality index calculated by the division model 23 with the scale of the all-division model 22. It can be removed.
- the abnormality diagnosis unit 37 calculates an abnormality index such as the degree of abnormality and the degree of deviation by using the classification model 23 according to the classification of the operation data to be diagnosed. For example, the abnormality diagnosis unit 37 inputs the operation data for diagnosis extracted from the actual data history 21 into the classification model 23, and the error (prediction error) between the predicted value by the classification model 23 and the corresponding measured value is large. Is calculated as an abnormality index. Then, the abnormality diagnosis unit 37 stores the calculated abnormality index in the storage unit 20 as the diagnosis data history 24.
- an abnormality index such as the degree of abnormality and the degree of deviation by using the classification model 23 according to the classification of the operation data to be diagnosed. For example, the abnormality diagnosis unit 37 inputs the operation data for diagnosis extracted from the actual data history 21 into the classification model 23, and the error (prediction error) between the predicted value by the classification model 23 and the corresponding measured value is large. Is calculated as an abnormality index. Then, the abnormality diagnosis unit 37 stores the calculated abnormality index in the storage unit 20 as the diagnosis
- the abnormality determination unit 38 determines the presence or absence of an abnormality based on the abnormality index calculated by the abnormality diagnosis unit 37. Then, the abnormality determination unit 38 stores the determination result in the storage unit 20 as the abnormality determination history 25. Further, the abnormality determination unit 38 outputs the determination result to the display unit 40.
- the display unit 40 is realized by a display device such as an LCD display or a CRT display.
- the display unit 40 gives guidance to the operator by displaying, for example, the diagnosis result in the abnormality diagnosis unit 37, the determination result in the abnormality determination unit 38, and the like based on the display signal input from the calculation unit 30.
- Model construction method A model construction method using the model construction apparatus according to the embodiment of the present invention will be described with reference to FIG. The model construction method is carried out offline at any time.
- the all-division learning data creation unit 31 creates all-division learning data based on the operation data stored in the actual data history 21 (step S1). Subsequently, the all-division model creation unit 32 creates the all-division model 22 by Lasso regression based on the all-division learning data (step S2).
- the modeling matrix creation unit 33 creates a modeling matrix based on the structure of the all-division model 22 (step S3).
- the segmented learning data creation unit 34 creates segmented learning data based on the operation data stored in the actual data history 21 (step S4).
- the division model creation unit 35 creates the division model 23 by Lasso regression based on the division learning data and the modeling matrix (step S5).
- the classification model sensitivity correction unit 36 calculates the correction coefficient ⁇ for correcting the prediction error of the classification model 23 by the above equation (4) (step S6). Subsequently, the classification model sensitivity correction unit 36 corrects the prediction error of the classification model 23 for each classification by the correction coefficient ⁇ (step S7), and ends this process.
- the abnormality diagnosis unit 37 determines whether or not an event has occurred (step S11).
- the occurrence of an event to be diagnosed is detected based on the operation data collected by the system. For example, when diagnosing the entire manufacturing process, the occurrence of an event can be detected by monitoring a flag signal or the like indicating the start or end of the manufacturing process.
- the abnormality diagnosis unit 37 If it is determined that no event has occurred (No in step S11), the abnormality diagnosis unit 37 returns to step S11. On the other hand, when it is determined that an event has occurred (Yes in step S11), the abnormality diagnosis unit 37 starts the abnormality diagnosis (step S12), and obtains the actual data (operation data) to be diagnosed from the actual data history 21. Extract only the required amount (step S13). Subsequently, the abnormality diagnosis unit 37 performs an abnormality diagnosis by the classification model 23 (step S14).
- the abnormality determination unit 38 determines the presence or absence of an abnormality based on the abnormality index calculated by the abnormality diagnosis unit 37 in step S14 (step S15).
- the abnormality determination unit 38 provides guidance to the operator by displaying the determination result on the display unit 40 (step S16). Then, the abnormality determination unit 38 ends this process and transitions to the initial state of waiting for an event. If it is determined in step S15 that there is no abnormality (No in step S15), the abnormality determination unit 38 ends this process and transitions to the initial state of waiting for an event.
- the classification model 23 is classified according to the type and physical characteristics of the target product.
- the all-division model 22 is constructed from the learning samples of all the divisions.
- the division model 23 is constructed based on the learning sample divided with the explanatory variable having a large influence as the upper limit.
- the division model 23 is created using the explanatory variables within the range of the explanatory variables used in the all division model 22.
- the abnormality diagnosis model construction method when the classification model 23 is constructed, the sensitivity is matched with that of the all division model 22. , The output of the classification model 23 is corrected by the correction coefficient ⁇ . In this way, by matching the scale of the anomaly index calculated by the classification model 23 with the scale of the all classification model 22, it is possible to eliminate the variation in sensitivity between the classifications.
- the abnormality diagnosis model construction method, the abnormality diagnosis method, the abnormality diagnosis model construction device, and the abnormality diagnosis device 1 according to the embodiment have the same sensitivity as the past classification model 23 when updating the classification model 23.
- the output of the newly mounted division model 23 is corrected by the correction coefficient ⁇ .
- the sensitivity does not change due to the model update.
- the method for constructing the abnormality diagnosis model, the method for diagnosing the abnormality, the apparatus for constructing the abnormality diagnosis model, and the abnormality diagnosis apparatus 1 according to the present invention have been specifically described with reference to the embodiments and examples for carrying out the invention.
- the gist of the present invention is not limited to these statements, and must be broadly interpreted based on the statements of the claims. Needless to say, various changes, modifications, etc. based on these descriptions are also included in the gist of the present invention.
- the calculation of the correction coefficient ⁇ and the sensitivity correction of the classification model 23 are performed by the classification model 23.
- the correction coefficient ⁇ can be calculated and the sensitivity of the classification model 23 can be corrected.
- Abnormality diagnosis device 10 Input unit 20 Storage unit 21 Actual data history 22 All-division model 23 Divisional model 24 Diagnosis data history 25 Abnormality judgment history 30 Calculation unit 31 All-division learning data creation unit 32 All-division model creation unit 33 Modeling matrix Creation unit 34 Divisional learning data creation unit 35 Divisional model creation unit 36 Divisional model sensitivity correction unit 37 Abnormality diagnosis unit 38 Abnormality judgment unit 40 Display unit
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Human Computer Interaction (AREA)
- Testing And Monitoring For Control Systems (AREA)
- Indicating And Signalling Devices For Elevators (AREA)
Abstract
Description
まず、本発明の実施形態に係る異常診断装置および異常診断モデルの構築装置(以下、「モデル構築装置」という)の構成について、図1を参照しながら説明する。異常診断装置は、プラント等におけるプロセスの異常を診断するためのものであり、モデル構築装置は、異常診断のためのモデルを構築するためのものである。
本発明の実施形態に係るモデル構築装置によるモデル構築方法について、図2を参照しながら説明する。モデル構築方法は、オフラインで任意のタイミングで実施される。
本発明の実施形態に係る異常診断装置1による異常診断方法について、図3を参照しながら説明する。異常診断方法は、操業中にオンラインで実施される。
10 入力部
20 記憶部
21 実績データ履歴
22 全区分モデル
23 区分化モデル
24 診断データ履歴
25 異常判定履歴
30 演算部
31 全区分学習データ作成部
32 全区分モデル作成部
33 モデル化マトリクス作成部
34 区分化学習データ作成部
35 区分化モデル作成部
36 区分化モデル感度補正部
37 異常診断部
38 異常判定部
40 表示部
Claims (8)
- プロセスの異常を診断する異常診断モデルの構築方法であって、
予め収集した正常時における全ての操業データを用いて、目的変数に対する影響度の小さい説明変数に関する回帰係数が0となるような第一の回帰モデルを作成する第一の回帰モデル作成ステップと、
前記操業データを予め定めた複数の区分に分け、前記区分ごとに、前記第一の回帰モデルで使用された説明変数の範囲内において、説明変数候補を決定する説明変数候補決定ステップと、
前記区分に含まれる操業データを用いて、目的変数に対する影響度の小さい説明変数候補に関する回帰係数が0となるような第二の回帰モデルを作成する第二の回帰モデル作成ステップと、
を含む異常診断モデルの構築方法。 - 前記第二の回帰モデル作成ステップの後に、
前記第一の回帰モデルの予測誤差のばらつき指標と、前記区分ごとの前記第二の回帰モデルの予測誤差のばらつき指標との比である補正係数を、前記区分ごとに算出する補正係数算出ステップと、
前記区分ごとの前記第二の回帰モデルの予測誤差を、前記補正係数によって補正する感度補正ステップと、
を更に含む請求項1に記載の異常診断モデルの構築方法。 - 前記第二の回帰モデル作成ステップの後に、
第一の学習データセットから構築された前記区分ごとの前記第二の回帰モデルの予測誤差のばらつき指標と、前記第一の学習データセットとは異なる第二の学習データセットから構築された前記区分ごとの前記第二の回帰モデルの予測誤差のばらつき指標との比である補正係数を、前記区分ごとに算出する補正係数算出ステップと、
前記区分ごとの前記第二の回帰モデルの予測誤差を、前記補正係数によって補正する感度補正ステップと、
を更に含む請求項1に記載の異常診断モデルの構築方法。 - 前記複数の区分は、製品の品種、製品のサイズ、操業条件および操業パターンを含む、
請求項1から請求項3のいずれか一項に記載の異常診断モデルの構築方法。 - 前記第一の回帰モデル作成ステップは、Lasso回帰により、前記第一の回帰モデルを作成し、
前記第二の回帰モデル作成ステップは、前記Lasso回帰により、前記第二の回帰モデルを作成する、
請求項1から請求項4のいずれか一項に記載の異常診断モデルの構築方法。 - 請求項1から請求項5のいずれか一項に記載の異常診断モデルの構築方法によって構築された異常診断モデルを用いた異常診断方法であって、
診断対象となる操業データの区分に応じた第二の回帰モデルを用いて、異常指標を算出する異常診断ステップと、
前記異常指標に基づいて、異常の有無を判定する異常判定ステップと、
を含む異常診断方法。 - プロセスの異常を診断する異常診断モデルの構築装置であって、
予め収集した正常時における全ての操業データを用いて、目的変数に対する影響度の小さい説明変数に関する回帰係数が0となるような第一の回帰モデルを作成する第一の回帰モデル作成手段と、
前記操業データを予め定めた複数の区分に分け、前記区分ごとに、前記第一の回帰モデルで使用された説明変数の範囲内において、説明変数候補を決定する説明変数候補決定手段と、
前記区分に含まれる操業データを用いて、目的変数に対する影響度の小さい説明変数候補に関する回帰係数が0となるような第二の回帰モデルを作成する第二の回帰モデル作成手段と、
を備える異常診断モデルの構築装置。 - 請求項7に記載の異常診断モデルの構築装置によって構築された異常診断モデルを用いた異常診断装置であって、
診断対象となる操業データの区分に応じた第二の回帰モデルを用いて、異常指標を算出する異常診断手段と、
前記異常指標に基づいて、異常の有無を判定する異常判定手段と、
を備える異常診断装置。
Priority Applications (5)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2022500007A JP7248188B2 (ja) | 2020-10-27 | 2021-09-21 | 異常診断モデルの構築方法、異常診断方法、異常診断モデルの構築装置および異常診断装置 |
EP21885758.9A EP4239427A4 (en) | 2020-10-27 | 2021-09-21 | METHOD FOR CONSTRUCTING AN ANOMALY DIAGNOSIS MODEL, METHOD FOR DIAGNOSIS OF ANOMALIES, DEVICE FOR CONSTRUCTING AN ANOMALY DIAGNOSIS MODEL AND DEVICE FOR DIAGNOSIS OF ANOMALIES |
KR1020237012293A KR20230066093A (ko) | 2020-10-27 | 2021-09-21 | 이상 진단 모델의 구축 방법, 이상 진단 방법, 이상 진단 모델의 구축 장치 및 이상 진단 장치 |
US18/031,656 US20230384780A1 (en) | 2020-10-27 | 2021-09-21 | Construction method of abnormality diagnosis model, abnormality diagnosis method, construction device of abnormality diagnosis model, and abnormality diagnosis device |
CN202180072111.1A CN116348829A (zh) | 2020-10-27 | 2021-09-21 | 异常诊断模型的构建方法、异常诊断方法、异常诊断模型的构建装置以及异常诊断装置 |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2020179341 | 2020-10-27 | ||
JP2020-179341 | 2020-10-27 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2022091639A1 true WO2022091639A1 (ja) | 2022-05-05 |
Family
ID=81382289
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2021/034596 WO2022091639A1 (ja) | 2020-10-27 | 2021-09-21 | 異常診断モデルの構築方法、異常診断方法、異常診断モデルの構築装置および異常診断装置 |
Country Status (7)
Country | Link |
---|---|
US (1) | US20230384780A1 (ja) |
EP (1) | EP4239427A4 (ja) |
JP (1) | JP7248188B2 (ja) |
KR (1) | KR20230066093A (ja) |
CN (1) | CN116348829A (ja) |
TW (1) | TWI788050B (ja) |
WO (1) | WO2022091639A1 (ja) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114999182A (zh) * | 2022-05-25 | 2022-09-02 | 中国人民解放军国防科技大学 | 基于lstm回馈机制的车流量预测方法、装置及设备 |
JP7504312B1 (ja) | 2023-05-15 | 2024-06-21 | 三菱電機株式会社 | 情報処理装置、生成方法、及び生成プログラム |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2009054843A (ja) * | 2007-08-28 | 2009-03-12 | Omron Corp | プロセス異常検出装置および方法並びにプログラム |
JP4922265B2 (ja) | 2008-08-19 | 2012-04-25 | 株式会社東芝 | プラント監視装置およびプラント監視方法 |
JP2012081518A (ja) * | 2010-09-16 | 2012-04-26 | Sumitomo Metal Ind Ltd | 厚鋼板の冷却制御方法、冷却制御装置及び厚鋼板の製造方法 |
WO2013011745A1 (ja) | 2011-07-15 | 2013-01-24 | 株式会社日立製作所 | 設備状態監視方法およびその装置 |
JP5499900B2 (ja) | 2010-05-25 | 2014-05-21 | Jfeスチール株式会社 | パターン自動抽出方法およびパターン自動抽出システム |
JP5651998B2 (ja) | 2010-05-25 | 2015-01-14 | Jfeスチール株式会社 | パターンライブラリを用いた異常診断方法および異常診断システム |
JP2019128904A (ja) * | 2018-01-26 | 2019-08-01 | Necソリューションイノベータ株式会社 | 予測システム、シミュレーションシステム、方法およびプログラム |
JP2020149282A (ja) * | 2019-03-13 | 2020-09-17 | 日本電信電話株式会社 | データ予測装置、データ予測方法、及びデータ予測プログラム |
Family Cites Families (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPS4922265B1 (ja) | 1969-08-08 | 1974-06-07 | ||
JPS5445065A (en) | 1977-09-12 | 1979-04-10 | Kito Corp | Chain block frame in light alloy casting in which core plate in steel is buried |
US20020069044A1 (en) * | 2000-05-04 | 2002-06-06 | The Procter & Gamble Company | Computer apparatuses and processes for analyzing a system having false start events |
WO2018083720A1 (ja) * | 2016-11-07 | 2018-05-11 | 日本電気株式会社 | 異常分析方法、プログラムおよびシステム |
US11112782B2 (en) * | 2017-06-19 | 2021-09-07 | Jfe Steel Corporation | Process anomalous state diagnostic device and process anomalous state diagnosis method |
CN108171422A (zh) * | 2017-12-28 | 2018-06-15 | 鞍钢集团自动化有限公司 | 一种钢铁智能工厂的平台构建方法 |
CN109459993B (zh) * | 2018-12-06 | 2021-06-25 | 湖南师范大学 | 一种流程工业过程在线自适应故障监测与诊断方法 |
-
2021
- 2021-09-21 US US18/031,656 patent/US20230384780A1/en active Pending
- 2021-09-21 WO PCT/JP2021/034596 patent/WO2022091639A1/ja active Application Filing
- 2021-09-21 JP JP2022500007A patent/JP7248188B2/ja active Active
- 2021-09-21 EP EP21885758.9A patent/EP4239427A4/en active Pending
- 2021-09-21 KR KR1020237012293A patent/KR20230066093A/ko unknown
- 2021-09-21 CN CN202180072111.1A patent/CN116348829A/zh active Pending
- 2021-10-13 TW TW110137935A patent/TWI788050B/zh active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2009054843A (ja) * | 2007-08-28 | 2009-03-12 | Omron Corp | プロセス異常検出装置および方法並びにプログラム |
JP4922265B2 (ja) | 2008-08-19 | 2012-04-25 | 株式会社東芝 | プラント監視装置およびプラント監視方法 |
JP5499900B2 (ja) | 2010-05-25 | 2014-05-21 | Jfeスチール株式会社 | パターン自動抽出方法およびパターン自動抽出システム |
JP5651998B2 (ja) | 2010-05-25 | 2015-01-14 | Jfeスチール株式会社 | パターンライブラリを用いた異常診断方法および異常診断システム |
JP2012081518A (ja) * | 2010-09-16 | 2012-04-26 | Sumitomo Metal Ind Ltd | 厚鋼板の冷却制御方法、冷却制御装置及び厚鋼板の製造方法 |
WO2013011745A1 (ja) | 2011-07-15 | 2013-01-24 | 株式会社日立製作所 | 設備状態監視方法およびその装置 |
JP2019128904A (ja) * | 2018-01-26 | 2019-08-01 | Necソリューションイノベータ株式会社 | 予測システム、シミュレーションシステム、方法およびプログラム |
JP2020149282A (ja) * | 2019-03-13 | 2020-09-17 | 日本電信電話株式会社 | データ予測装置、データ予測方法、及びデータ予測プログラム |
Non-Patent Citations (1)
Title |
---|
See also references of EP4239427A4 |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114999182A (zh) * | 2022-05-25 | 2022-09-02 | 中国人民解放军国防科技大学 | 基于lstm回馈机制的车流量预测方法、装置及设备 |
JP7504312B1 (ja) | 2023-05-15 | 2024-06-21 | 三菱電機株式会社 | 情報処理装置、生成方法、及び生成プログラム |
Also Published As
Publication number | Publication date |
---|---|
TW202230169A (zh) | 2022-08-01 |
TWI788050B (zh) | 2022-12-21 |
EP4239427A1 (en) | 2023-09-06 |
JPWO2022091639A1 (ja) | 2022-05-05 |
JP7248188B2 (ja) | 2023-03-29 |
US20230384780A1 (en) | 2023-11-30 |
KR20230066093A (ko) | 2023-05-12 |
CN116348829A (zh) | 2023-06-27 |
EP4239427A4 (en) | 2024-05-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US10809153B2 (en) | Detecting apparatus, detection method, and program | |
JP5539382B2 (ja) | 航空エンジン内の故障の識別 | |
JP4394728B2 (ja) | 影響要因特定装置 | |
WO2022091639A1 (ja) | 異常診断モデルの構築方法、異常診断方法、異常診断モデルの構築装置および異常診断装置 | |
WO2022038804A1 (ja) | 診断装置及びパラメータ調整方法 | |
US11740592B2 (en) | Control method, control apparatus, mechanical equipment, and recording medium | |
CN113420061B (zh) | 炼油和化工生产装置的稳态工况分析方法、优化调试方法及系统 | |
US20220188307A1 (en) | Data analysis apparatus, method and system | |
CN113834828A (zh) | 产品品质分析支援系统 | |
JP6885321B2 (ja) | プロセスの状態診断方法及び状態診断装置 | |
JP6312955B1 (ja) | 品質分析装置及び品質分析方法 | |
US20230393113A1 (en) | Construction method of abnormality diagnosis model, abnormality diagnosis method, construction device of abnormality diagnosis model, and abnormality diagnosis device | |
EP3623965A1 (en) | Fabricated data detection method | |
JP6579163B2 (ja) | プロセスの状態診断方法及び状態診断装置 | |
JPH05253797A (ja) | ライン異常判定方式 | |
JP7223947B2 (ja) | 製造条件計算装置、製造条件計算方法及び製造条件計算プログラム | |
JP2018005715A (ja) | 製造プロセスの異常状態診断方法及び異常状態診断装置 | |
JP6840294B1 (ja) | プログラマブル表示器、制御システムおよび解析方法 | |
JP7180676B2 (ja) | プロセス状態監視装置及びプロセス状態監視方法 | |
CN118482817B (zh) | 一种基于数据分析的数字化纺织设备管理方法及系统 | |
Becker et al. | Predictive Accuracy Index in evaluating the dataset shift (case study) | |
JP2024013399A (ja) | 設備の劣化診断装置及び方法 | |
JP2002006940A (ja) | 情報解析方法、情報解析装置及び情報解析プログラムを記録した記録媒体 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
ENP | Entry into the national phase |
Ref document number: 2022500007 Country of ref document: JP Kind code of ref document: A |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 21885758 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 20237012293 Country of ref document: KR Kind code of ref document: A |
|
WWE | Wipo information: entry into national phase |
Ref document number: 18031656 Country of ref document: US |
|
WWE | Wipo information: entry into national phase |
Ref document number: 202317028446 Country of ref document: IN |
|
REG | Reference to national code |
Ref country code: BR Ref legal event code: B01A Ref document number: 112023007208 Country of ref document: BR |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
ENP | Entry into the national phase |
Ref document number: 2021885758 Country of ref document: EP Effective date: 20230530 |
|
ENP | Entry into the national phase |
Ref document number: 112023007208 Country of ref document: BR Kind code of ref document: A2 Effective date: 20230418 |