Abnormality diagnosis method and abnormality diagnosis device

Abstract

This abnormality diagnosis method includes: an abnormality determination step for determining the presence or absence of an abnormality in a manufacturing facility by using a set of first prediction models which are constructed on the basis of a plurality of variables indicating a normal state of the manufacturing facility and in which said variables are complementarily used as target and explanatory variables; an operation time influence coefficient calculation step for constructing, on the basis of the plurality of variables, a set of second prediction models in which said variables are complementarily used as target and explanatory variables; an abnormality cause estimation matrix calculation step for calculating an abnormality cause estimation matrix; and an abnormality cause variable identification step for using the abnormality cause estimation matrix to identify a variable causing an abnormality.

Description

Abnormality diagnosis method and abnormality diagnosis device

[0001] The present invention relates to an abnormality diagnosis method and an abnormality diagnosis device.

[0002] There are two methods for diagnosing the manufacturing state of a manufacturing process, especially abnormal states: the model-based approach and the database approach. The model-based approach is an approach in which a model that mathematically represents the physical or chemical phenomena in the manufacturing process is constructed, and the constructed model is used to diagnose the manufacturing state of the manufacturing process. On the other hand, the database approach is an approach in which a statistical analysis model is constructed from operational data obtained in the manufacturing process, and the constructed model is used to diagnose the manufacturing state of the manufacturing process.

[0003] In manufacturing processes such as steelmaking, many different types and sizes of products are manufactured on a single production line, resulting in numerous operating patterns. Furthermore, in manufacturing processes such as blast furnaces, natural materials such as iron ore and coke are used as raw materials, resulting in significant variability in the manufacturing process. Therefore, when diagnosing the manufacturing status of manufacturing processes such as steelmaking, a model-based approach alone has limitations.

[0004] The database approach involves a diagnostic method that extracts characteristics from past operational data under normal conditions and determines the differences between them and the current operational data. One such method is to diagnose the degree of deviation from the normal state using a multiple regression model.

[0005] An example is described in Patent Document 1. Specifically, Patent Document 1 describes a method for estimating the cause of an abnormal state using influence coefficients of a multiple regression model created using operational data during normal times. State variables indicating the state of equipment and quality and cause variables that are candidates for the cause of the abnormality are used as input variables, and the method compares the influence coefficients of a multiple regression model, in which the state variables are response variables and the cause variables are explanatory variables, between normal and abnormal times, and estimates one of the cause variables with the greatest change as the cause of the abnormality.

[0006] Japanese Patent Application Publication No. 7-159289

[0007] In steel manufacturing processes and the like, while measurements from a large number of sensors can be handled, most of these measurements are taken for the purpose of operation management and equipment control. Therefore, as in Patent Document 1, it is not always possible to obtain sufficient measurement values ​​that directly indicate the equipment status or the cause of an abnormality. Furthermore, even if there are measurement values ​​that correspond one-to-one with the cause of such an abnormality, it is still impossible to cover all abnormalities with just those measurement values.

[0008] Meanwhile, recent advances in data collection and analysis technologies have created an environment for handling large amounts of data, commonly referred to as "big data." In light of the aforementioned situation, maintaining stable operations requires comprehensive and accurate detection of abnormalities from large amounts of data, presentation of data related to the abnormalities, and prompt maintenance action. In this case, the large amount of data does not necessarily correspond one-to-one to the cause of the abnormality, as in Patent Document 1, and therefore the content of the abnormality presented may not be limited to one.

[0009] The present invention has been made in consideration of the above, and aims to provide an abnormality diagnosis method and an abnormality diagnosis device that can detect abnormal conditions from large amounts of measurement data acquired in a manufacturing process and quickly and clearly present measurement values ​​related to the abnormality.

[0010] In order to solve the above-mentioned problems and achieve the object, an abnormality diagnosis method according to the present invention includes: an operating variable acquisition step of acquiring a plurality of variables indicating a state of manufacturing equipment during operation; an abnormality determination step of determining whether or not an abnormality exists in the manufacturing equipment using a set of first prediction models constructed based on the plurality of variables acquired in the operating variable acquisition step and a plurality of variables indicating a state of the manufacturing equipment during normal operation, the first prediction models being used complementarily as a dependent variable and an explanatory variable; an operating influence coefficient calculation step of, if the presence of an abnormality is determined, constructing a set of second prediction models based on the plurality of variables acquired in the operating variable acquisition step, the second prediction models being used complementarily as a dependent variable and an explanatory variable; an abnormality cause estimation matrix calculation step of calculating an abnormality cause estimation matrix by comparing a first influence coefficient, which is an influence coefficient of an explanatory variable on a dependent variable for the set of first prediction models, with a second influence coefficient, which is an influence coefficient of an explanatory variable on a dependent variable for the set of second prediction models; and an abnormality cause variable identification step of identifying a variable that causes the abnormality using the abnormality cause estimation matrix.

[0011] Furthermore, in the abnormality diagnosis method according to the present invention, in the above invention, the abnormality cause estimation matrix has a tabular structure in which the objective variable and the explanatory variable are key items, and the comparison result between the first influence coefficient and the second influence coefficient is used as an element.

[0012] In the anomaly diagnosis method according to the present invention, in the above invention, the anomaly cause variable identifying step uses the anomaly cause estimation matrix to focus on explanatory variables or objective variables related to the anomaly, and identifies the variable causing the anomaly based on changes in the influence coefficient related to the variable in question.

[0013] In the anomaly diagnosis method according to the present invention, in the above-described invention, the anomaly cause variable identifying step identifies, as the variable causing the anomaly, a variable for which the change in the influence coefficient is greater than a predetermined threshold in both the column of the explanatory variable and the row of the objective variable of the anomaly cause estimation matrix.

[0014] In addition, in the abnormality diagnosis method according to the present invention, in the above invention, the abnormality determination step calculates the degree of abnormality based on the prediction error of each variable using the first prediction model, and determines whether or not there is an abnormality in the manufacturing equipment by referring to the maximum, average, or sum of the degrees of abnormality of all or some of the variables.

[0015] Moreover, in the above-described invention, the abnormality diagnosis method according to the present invention further includes an abnormality cause estimating step of estimating the cause of the abnormality based on the identified variables.

[0016] In addition, in the anomaly diagnosis method according to the present invention, in the above invention, the anomaly cause estimation step estimates the cause of the anomaly by referring to a table of candidate causes of the anomaly according to combinations of the dependent variable and the explanatory variables that are candidates for the anomaly.

[0017] In addition, in the anomaly diagnosis method according to the present invention, in the above invention, the anomaly cause estimation step estimates the cause of the anomaly by referring to a table of candidate causes of the anomaly and corresponding actions according to combinations of the objective variable and explanatory variables that are candidates for the anomaly, and determines an action corresponding to the estimated cause of the anomaly.

[0018] In the abnormality diagnosis method according to the present invention, in the above invention, the abnormality cause estimation step outputs a command or an alert to the control device corresponding to the determined action.

[0019] In order to solve the above-mentioned problems and achieve the object, the abnormality diagnosis device of the present invention includes an abnormality determination unit that determines the presence or absence of an abnormality in the manufacturing equipment using a plurality of variables that indicate a state of the manufacturing equipment during operation and a set of first prediction models that are constructed based on a plurality of variables that indicate a state of the manufacturing equipment under normal conditions, and in which the variables are used complementarily as a dependent variable and an explanatory variable; an operation influence coefficient calculation unit that, when it is determined that an abnormality is present, constructs a set of second prediction models that use the variables complementarily as a dependent variable and an explanatory variable, based on the plurality of variables that indicate the state of the manufacturing equipment during operation; an abnormality cause estimation matrix calculation unit that calculates an abnormality cause estimation matrix by comparing a first influence coefficient that is an influence coefficient of the explanatory variables on the dependent variable for the set of first prediction models with a second influence coefficient that is an influence coefficient of the explanatory variables on the dependent variable for the set of second prediction models; and an abnormality cause variable identification unit that identifies a variable that causes the abnormality using the abnormality cause estimation matrix.

[0020] Moreover, in the above-described invention, the abnormality diagnosis device according to the present invention further comprises an abnormality cause estimation unit that estimates the cause of the abnormality based on the identified variables.

[0021] According to the abnormality diagnosis method and abnormality diagnosis device of the present invention, it is possible to detect an abnormal state from a large amount of measurement data acquired in a manufacturing process and to quickly and clearly present measurement values ​​related to the abnormality.

[0022] Fig. 1 is a block diagram showing a schematic configuration of an abnormality diagnosis device according to an embodiment of the present invention. Fig. 2 is a flowchart showing the flow of an abnormality diagnosis method executed by the abnormality diagnosis device according to an embodiment of the present invention. Fig. 3 is a diagram showing an example of an abnormality cause estimation matrix calculated with respect to changes in influence coefficients of a multiple regression model derived by the least squares method in the abnormality cause estimation matrix calculation step of the abnormality diagnosis method according to an embodiment of the present invention. Fig. 4 is a diagram showing an example of an abnormality cause estimation matrix calculated with respect to changes in influence coefficients of a multiple regression model derived by a sparse structure learning method such as Lasso regression analysis in the abnormality cause estimation matrix calculation step of the abnormality diagnosis method according to an embodiment of the present invention. Fig. 5 is a diagram showing an example of a table in which a response variable, explanatory variables serving as abnormality candidates, and abnormality cause candidates are previously organized in the abnormality cause estimation step of the abnormality diagnosis method according to an embodiment of the present invention.

[0023] An abnormality diagnosis method and an abnormality diagnosis device according to an embodiment of the present invention will be described with reference to the drawings.

[0024] (Abnormality diagnosis device) The configuration of an abnormality diagnosis device according to an embodiment of the present invention will be described with reference to Fig. 1. The abnormality diagnosis device 1 is for diagnosing the presence or absence of an abnormality in manufacturing equipment in a manufacturing process such as an iron and steel process. The abnormality diagnosis device 1 is realized by an information processing device such as a general-purpose computer such as a workstation or a personal computer, or a server located on the cloud. The abnormality diagnosis device 1 also includes an input unit 10, an output unit 20, a storage unit 30, and a calculation unit 40.

[0025] The input unit 10 inputs various sensor data, various setting values, etc., which indicate the operating conditions of the manufacturing equipment and products, to the calculation unit 40 periodically or irregularly.

[0026] The output unit 20 displays various calculation results from the calculation unit 40. Examples of the calculation results output by the output unit 20 include the abnormality determination result by the abnormality determination unit 42, the abnormality cause estimation matrix calculated by the abnormality cause estimation matrix calculation unit 44, and the variables that cause the abnormality identified by the abnormality cause variable identification unit 45. Examples of the calculation results output by the output unit 20 include the cause of the abnormality estimated by the abnormality cause estimation unit 46. Examples of the calculation results output by the output unit 20 include commands and alert information to a control device that correspond to the cause of the abnormality determined by the abnormality cause estimation unit 46.

[0027] The storage unit 30 is configured by a storage device such as an erasable programmable ROM (EPROM), a hard disk drive (HDD), a solid state drive (SSD), etc. The storage unit 30 may also include a temporary storage device such as a random access memory (RAM) that is used by the abnormality detection device 1 in the process of performing the abnormality detection process.

[0028] The storage unit 30 stores an operation data DB (database) 31 and a normal operation data DB (database) 32. The operation data DB 31 stores data such as various sensor data indicating the state of the manufacturing equipment and products during operation, various set values, other prediction results, and influence coefficients described below. The normal operation data DB 32 stores data such as various sensor data indicating the state of the manufacturing equipment and products during normal operation (during past normal operation), various set values, and influence coefficients described below.

[0029] The calculation unit 40 causes a computer to process each function of abnormality diagnosis, causing the computer to function as the abnormality diagnosis device 1. The calculation unit 40 is configured by a processor (arithmetic processing device) such as a CPU (Central Processing Unit). The calculation unit 40 also functions as a normal state influence coefficient calculation unit 41, an abnormality determination unit 42, an operation state influence coefficient calculation unit 43, an abnormality cause estimation matrix calculation unit 44, and an abnormality cause variable identification unit 45. In addition to the above, the calculation unit 40 also functions as an abnormality cause estimation unit 46 as necessary.

[0030] The normal state influence coefficient calculation unit 41 constructs a plurality of first prediction models (multiple regression models) based on a plurality of variables (e.g., sensor data) acquired from the manufacturing equipment and products during normal operation, and calculates the influence coefficient of each explanatory variable. Details of the processing by the normal state influence coefficient calculation unit 41 will be described later (see FIG. 2 ). The explanatory variables used when constructing the first prediction models may be not only sensor data but also operational settings (e.g., set values ​​of operating conditions).

[0031] The abnormality determination unit 42 calculates a prediction error, which is the difference between the actual value and the predicted value, using a first prediction model constructed based on multiple variables acquired during normal operation, and determines whether or not an abnormality exists based on the prediction error. Details of the processing by the abnormality determination unit 42 will be described later (see FIG. 2).

[0032] The operation influence coefficient calculation unit 43 constructs a plurality of second prediction models (multiple regression models) based on a plurality of variables (e.g., sensor data) acquired from the manufacturing equipment and products during operation, and calculates the influence coefficient of each explanatory variable. Details of the processing by the operation influence coefficient calculation unit 43 will be described later (see FIG. 2). The explanatory variables used when constructing the second prediction models may be not only sensor data but also operational settings (e.g., set values ​​of operating conditions).

[0033] The abnormality cause estimation matrix calculation unit 44 calculates an abnormality cause estimation matrix (abnormality cause estimation matrix) that allows comparison between the influence coefficients during normal times calculated by the normal time influence coefficient calculation unit 41 and the influence coefficients during operation calculated by the operation time influence coefficient calculation unit 43, and stores the matrix in the operation data DB 31. Details of the processing by the abnormality cause estimation matrix calculation unit 44 will be described later (see FIGS. 2 to 4).

[0034] The abnormality cause variable identifying unit 45 identifies variables that cause the abnormality based on the abnormality cause estimation matrix created by the abnormality cause estimation matrix calculating unit 44. Details of the processing by the abnormality cause variable identifying unit 45 will be described later (see FIGS. 2 to 4).

[0035] The abnormality cause estimation unit 46 estimates the cause of the abnormality based on the variables identified by the abnormality cause variable identification unit 45 and a previously created table, etc. Details of the processing by the abnormality cause estimation unit 46 will be described later (see FIGS. 2 and 5).

[0036] (Abnormality Diagnosis Method) Details of the abnormality diagnosis method executed by the abnormality diagnosis device according to an embodiment of the present invention will be described with reference to Figures 2 to 5. The abnormality diagnosis method includes a normal-state influence coefficient calculation step (step S1), an operational variable acquisition step (step S2), an abnormality determination step (steps S3 to S5), and an operational influence coefficient calculation step (step S6). The abnormality diagnosis method also includes an abnormality cause estimation matrix calculation step (step S7) and an abnormality cause variable identification step (step S8). In addition to the above steps, the abnormality diagnosis method further includes an abnormality cause estimation step (step S9) as needed.

[0037] <Normal State Impact Coefficient Calculation Step> In the normal state impact coefficient calculation step, an impact coefficient is derived for a set of multiple regression models based on normal state sensor data (variables) measured from manufacturing equipment or products (hereinafter referred to as "manufacturing equipment, etc.") (step S1).

[0038] In the normal state influence coefficient calculation step, the normal state influence coefficient calculation unit 41 first acquires a plurality of variables (sensor data) indicating the normal state of the manufacturing equipment, etc., from the normal state operation data DB 32. Next, based on the acquired plurality of variables, the normal state influence coefficient calculation unit 41 constructs a set of multiple regression models in which the variables are used complementarily as a response variable and an explanatory variable, and derives an influence coefficient (first influence coefficient) for each explanatory variable of each multiple regression model. Note that, hereinafter, the multiple regression model constructed in the normal state influence coefficient calculation step is also referred to as a "first prediction model." Also, hereinafter, the sensor data under normal conditions is also referred to as "normal state data."

[0039] In the normal state influence coefficient calculation step, for example, the commonly used least squares method can be used as the multiple regression analysis method. Furthermore, when a large number of variables are handled, a sparse structure learning method such as Lasso regression analysis can be used as the multiple regression analysis method. In a model created using a sparse structure learning method such as Lasso regression analysis, variables with small influence coefficients (degree of importance) are excluded from the explanatory variables in advance, which limits the variables to be verified when identifying the variable causing the abnormality, making it easier to identify the cause of the abnormality. A sparse structure learning method such as Lasso regression analysis is particularly effective when it is necessary to estimate the cause of the abnormality from a large amount of sensor data.

[0040] In the normal state influence coefficient calculation step, a plurality of multiple regression models are constructed while switching between the response variable and the explanatory variable in accordance with the number of variables acquired from the normal state operation data DB 32. For example, if 10 variables are acquired from the normal state operation data DB 32, a total of 10 multiple regression models are constructed while switching between the response variable and the explanatory variable in the normal state influence coefficient calculation step.

[0041] <Operational Variable Acquisition Step> In the operational variable acquisition step, sensor data (variables) during operation (current) is acquired from the manufacturing equipment, etc. (step S2). In the operational variable acquisition step, the abnormality determination unit 42 acquires multiple variables (sensor data) indicating the state of the manufacturing equipment, etc. during operation from the operational data DB 31. Note that in the operational variable acquisition step, preprocessing such as shifting the sensor data in the chronological order or sampling may be performed as necessary. In the following, the sensor data during operation will also be referred to as "operational data."

[0042] <Abnormality Determination Step> In the abnormality determination step, a predicted value is first calculated using a multiple regression model (first prediction model) constructed from variables under normal conditions (step S3). The abnormality determination unit 42 calculates a predicted value (objective variable) using the operational data acquired in step S2 as an explanatory variable and the influence coefficient under normal conditions (first influence coefficient) derived in step S1. Note that the calculation of the predicted value is performed for each multiple regression model constructed in step S1.

[0043] Next, the abnormality determination unit 42 calculates a deviation index (degree of abnormality) based on the difference between the predicted value calculated in step S3 and the actual measurement value of the sensor data acquired in step S2 (step S4). As the deviation index, for example, a value obtained by dividing the square of the difference between the predicted value and the actual measurement value by the standard deviation of the distribution of the differences between the predicted value calculated from normal data and the actual value can be used. In this case, it becomes possible to compare the deviation index as a statistical deviation index regardless of the unit (physical interpretation) of the variable to be predicted, regardless of the variable.

[0044] Next, the abnormality determination unit 42 determines whether or not there is an abnormality in the manufacturing equipment, etc., based on the deviation index calculated in step S4 (step S5). The abnormality determination unit 42 determines whether or not there is an abnormality by, for example, comparing the deviation index calculated in step S4 with a preset determination criterion. If it is determined in step S5 that there is an abnormality, the process proceeds to step S6, and if it is determined that there is no abnormality, the process returns to step S2.

[0045] In this way, in the anomaly determination step (steps S3 to S5), the presence or absence of an anomaly in the manufacturing equipment, etc. is determined using a set of multiple variables acquired in the operation variable acquisition step and a multiple regression model (first prediction model) constructed based on normal state data. Also, in the anomaly determination step, the degree of anomaly (deviation index) may be calculated based on the prediction error of each variable using the multiple regression model (first prediction model), and the presence or absence of an anomaly in the manufacturing equipment, etc. may be determined by referring to the maximum, average, or sum of the degrees of anomaly of all or some of the variables.

[0046] <Operational Impact Coefficient Calculation Step> In the operational impact coefficient calculation step, an impact coefficient is derived for a set of multiple regression models based on sensor data (variables) measured from manufacturing equipment or the like during operation (step S6).

[0047] In the operation impact coefficient calculation step, the operation impact coefficient calculation unit 43 constructs a set of multiple regression models in which the variables acquired in step S2 are used complementarily as a response variable and an explanatory variable, and derives an impact coefficient (second impact coefficient) for each explanatory variable of each multiple regression model. Note that, hereinafter, the multiple regression model constructed in the operation impact coefficient calculation step will also be referred to as a "second prediction model."

[0048] In the operation impact coefficient calculation step, the commonly used least squares method can be used as the multiple regression analysis method. When a large number of variables are handled, a sparse structure learning method such as Lasso regression analysis can be used as the multiple regression analysis method.

[0049] In the operation influence coefficient calculation step, a plurality of multiple regression models are constructed in accordance with the number of variables acquired from the operation data DB 31 and while interchanging the response variables and explanatory variables. For example, if 10 variables are acquired from the operation data DB 31, a total of 10 multiple regression models are constructed in the operation influence coefficient calculation step while interchanging the response variables and explanatory variables. In the operation influence coefficient calculation step, a plurality of multiple regression models are constructed using the same number and types of variables as those used in the normal state influence coefficient calculation step.

[0050] <Anomaly Cause Estimation Matrix Calculation Step> In the anomaly cause estimation matrix calculation step, the influence coefficient during normal operation (first influence coefficient) derived in step S1 is compared with the influence coefficient during operation (second influence coefficient) derived in step S6 to calculate an anomaly cause estimation matrix (step S7).

[0051] In the anomaly cause estimation matrix calculation step, the anomaly cause estimation matrix calculation unit 44 compares the influence coefficients (first influence coefficients) of the explanatory variables on the dependent variable for the set of first prediction models with the influence coefficients (second influence coefficients) of the explanatory variables on the dependent variable for the set of second prediction models. At this time, examples of the comparison method include calculating the amount of change (difference) in the influence coefficients and calculating the rate of change in the influence coefficients.

[0052] More specifically, the anomaly cause estimation matrix calculation unit 44 creates a matrix whose elements are the influence coefficients (first influence coefficients) of the set of first prediction models, and a matrix whose elements are the influence coefficients (second influence coefficients) of the set of second prediction models, and then calculates the anomaly cause estimation matrix by comparing the elements (influence coefficients) of each matrix of both matrices (calculating the amount of change or rate of change).

[0053] 3 shows an example of an anomaly cause estimation matrix calculated from the multiple regression model constructed using the least squares method in steps S1 and S6. As shown in FIG. 3, the anomaly cause estimation matrix is ​​configured in a matrix with the dependent variable arranged vertically and the explanatory variables arranged horizontally. The anomaly cause estimation matrix has a tabular structure with the dependent variable and the explanatory variables as key items, and uses the comparison results of the influence coefficients (first influence coefficient and second influence coefficient) as elements. The comparison results of the influence coefficients shown in FIG. 3 indicate the absolute values ​​of the differences between the influence coefficients, but the average of the influence coefficients or the sum of the absolute values ​​may also be used.

[0054] 4 shows an example of an anomaly cause estimation matrix calculated from a multiple regression model constructed using a sparse structure learning method such as Lasso regression analysis in steps S1 and S6. The comparison results of influence coefficients shown in Fig. 4 indicate absolute values ​​of the differences between the influence coefficients, but the average of the influence coefficients or the sum of the absolute values ​​may also be used.

[0055] 3 and 4, the larger the numerical value of an element in the matrix, the larger the change in the influence coefficient during operation relative to normal operation. Also, in FIGS. 3 and 4, the change in the influence coefficient is larger in the columns of the explanatory variables Xi and Xj and the rows of the objective variables Xi and Xj. The anomaly cause estimation matrix calculation unit 44 may color-code the inside of the anomaly cause estimation matrix according to the magnitude of the change in the influence coefficient, and output the result from the output unit 20. By color-coding the inside of the anomaly cause estimation matrix and presenting it to an operator or the like in this way, the variable causing the anomaly can be easily identified.

[0056] <Anomaly Cause Variable Identification Step> In the anomaly cause variable identification step, the anomaly cause estimation matrix calculated in step S7 is used to identify the variables that cause the anomaly (step S8). In the anomaly cause variable identification step, the anomaly cause variable identification unit 45 uses the anomaly cause estimation matrix to focus on explanatory variables or objective variables related to the anomaly, and identifies the variables that cause the anomaly based on changes in the influence coefficients related to the variables in question.

[0057] The anomaly cause variable identifying unit 45 identifies, for example, an explanatory variable corresponding to a dependent variable determined to be abnormal in step S5 as a variable causing the anomaly. At this time, it is considered that the influence coefficient of the variable causing the anomaly in a second prediction model using the variable causing the anomaly as an explanatory variable has changed significantly from the influence coefficient of the variable in a first prediction model constructed using variables in normal conditions. Therefore, if the influence coefficient (second influence coefficient) of a second prediction model that predicts dependent variables other than the dependent variable determined to be abnormal in step S5 has changed significantly from the influence coefficient (first influence coefficient) of a first prediction model constructed using variables in normal conditions, the explanatory variable corresponding to the dependent variable is also identified as a variable causing the anomaly.

[0058] On the other hand, the prediction error may also deviate due to an abnormality in another explanatory variable. Therefore, the influence coefficient of each explanatory variable is focused on, and the influence coefficient of the explanatory variable that has changed significantly from normal is extracted, and the corresponding variable is identified as the cause of the abnormality. The influence coefficients of the above-mentioned abnormality candidates in the other second prediction model are then compared with the normal state, and if it is determined that the amount of change in the influence coefficient is greater than a predetermined range, it is determined that the variable is the cause of the abnormality.

[0059] 3 and 4, the anomaly cause variable identifying unit 45 identifies a variable as the cause of the anomaly when, for example, the change in the influence coefficient in both the column of the explanatory variable and the row of the dependent variable is greater than a predetermined threshold. For example, when the threshold is "0.1," the change in the influence coefficient in the column of the explanatory variables Xi and Xj and in the row of the dependent variables Xi and Xj is greater than the threshold. In this case, the anomaly cause variable identifying unit 45 identifies the explanatory variables Xi and Xj as the variables causing the anomaly.

[0060] In this way, the anomaly diagnosis method according to the embodiment can detect an abnormal state from a large amount of measurement data acquired in a manufacturing process and quickly and clearly present measurement values ​​related to the abnormality. Furthermore, by identifying explanatory variables that cause the abnormality, it is possible to identify an abnormality in a sensor corresponding to the explanatory variables. Furthermore, when the setting values ​​of operating conditions other than sensor data are used as explanatory variables, it is also possible to identify setting values ​​of operating conditions that are candidates for an abnormality.

[0061] <Anomaly Cause Estimation Step> In the anomaly cause estimation step, the cause of the anomaly is estimated based on the variables identified in the anomaly cause variable identification step.

[0062] Estimating the cause of an anomaly largely depends on knowledge gained at the time of the anomaly occurring in a specific process or piece of equipment. Therefore, the following explains the concept of estimating the cause of an anomaly using the steelmaking process as an example. The causes of anomalies in manufacturing processes can be mainly classified into equipment anomalies such as mechanical or electrical failures, process anomalies such as operational factors or instability in the manufacturing process, system factors such as abnormal sensor data or control software malfunctions, and human factors such as errors in operating procedures or settings.

[0063] For example, in the steel rolling process, differential load, which is the difference in load between the two ends of the mill, is monitored. If a process system abnormality occurs, such as the rolled material meandering and threading off-center, an imbalance in the load distribution on the rolling rolls occurs, ultimately resulting in a differential load. Also, differential loads can occur as a result of a discrepancy between the target and actual leveling values ​​due to equipment abnormalities such as a malfunction in the roll-down mechanism or mechanical backlash in the mill components. Other differential loads can also occur due to systemic factors, such as an abnormal output from one of the load cells (sensors) that detect the loads at both ends of the mill, or due to an operator's leveling error.

[0064] In the abnormality diagnosis method according to this embodiment, for example, in a multiple regression model using differential load as the objective variable, items such as the off-center amount and leveling amount can be explanatory variables with significant influence coefficients. Therefore, if the off-center amount is identified as a variable causing an abnormality based on the abnormality cause estimation matrix in step S8, it is considered that the variable is the differential load, an operational factor caused by an increase in the off-center amount due to strip meandering. In this case, strip meandering due to excessive operation by the operator or the like is assumed to be the cause of the strip meandering.

[0065] On the other hand, if the leveling amount is identified as the variable causing the abnormality in step S8, it is possible that a malfunction has occurred in the rolling mechanism of the rolling mill, such as the cylinder of the rolling device. Also, if the differential load itself is identified as the variable causing the abnormality in step S8, there is a concern that the differential load may be caused by the shape of the rolled material (excessive wedge shape) or temperature (temperature non-uniformity).

[0066] Therefore, in the anomaly cause estimation step, candidate causes of anomalies are organized in advance in a table according to combinations of the dependent variable and the explanatory variables that are candidate anomalies, and the table is stored in the storage unit 30, for example, as shown in Fig. 5. Then, the anomaly cause estimation unit 46 estimates the cause of the anomaly by referring to the table according to the observed anomaly, i.e., the dependent variable and the candidate anomaly.

[0067] Furthermore, in the anomaly cause estimation step, it is desirable to organize the causes of anomalies and corresponding candidate actions in a table in advance according to combinations of the objective variable and the explanatory variables that are candidates for anomaly, and store the table in the storage unit 30, etc. Then, the anomaly cause estimation unit 46 may refer to the table in accordance with the observed anomaly, i.e., the objective variable and the candidate anomaly, to estimate the cause of the anomaly and determine the corresponding action.

[0068] Furthermore, in the anomaly cause estimating step, an action command to the process and equipment control device corresponding to the determined action may be output from the output unit 20. Also, in the anomaly cause estimating step, alert information may be output from the output unit 20 to the operator's monitoring monitor.

[0069] For example, if a differential load occurs in the steel rolling process and there is an equipment system abnormality such as a malfunction in the roll down mechanism or mechanical backlash in a mill component, the output unit 20 outputs an emergency stop command for the rolling process to prevent serious trouble. Also, if there is an abnormality in the process system where the rolled material meanders and is threaded off-center, the output unit 20 outputs a leveling correction command according to the amount of off-center to the mill control device. Also, if there is a systemic factor such as the output of a load cell (sensor), the output unit 20 outputs a command to switch to an alternative sensor to the mill control device. Also, if there is an operator error in leveling operation, the output unit 20 outputs alert information to the operator's monitoring monitor.

[0070] In the anomaly diagnosis method and anomaly diagnosis device according to the above-described embodiments, a large number of variables (sensor data) obtained during operation are collected, influence coefficients are calculated in a set of multiple regression models in which the variables serve as complementary response variables and explanatory variables, and an anomaly cause estimation matrix is ​​calculated that compares the calculated influence coefficients with influence coefficients under normal conditions.The anomaly cause estimation matrix is ​​then used to identify the variables that cause the anomaly.

[0071] 3 and 4, in an anomaly cause estimation matrix, when the vertical direction (row direction) is the objective variable and the horizontal direction (column direction) is the explanatory variable and the amount of change in the influence coefficient is input, the rows and columns including variables related to the anomaly change more significantly than variables not related to the anomaly. Therefore, by displaying it using a color chart or the like, the variables related to the anomaly can be presented in an easy-to-understand manner.

[0072] In this way, the anomaly diagnosis method and anomaly diagnosis device according to the embodiment estimate an anomaly cause variable from a set of many multiple regression models, thereby improving estimation accuracy compared to cause diagnosis using a single model as proposed in Patent Document 1. Furthermore, the anomaly diagnosis method and anomaly diagnosis device according to the embodiment calculate an anomaly cause estimation matrix that comprehensively compares influence coefficients in normal and abnormal conditions for a set of multiple regression models, thereby making it possible to clearly and easily present the variable that causes the anomaly even when multiple variables are involved.

[0073] Furthermore, in the anomaly diagnosis method and anomaly diagnosis device according to the embodiment, by using an anomaly cause estimation matrix to identify the variables that cause the multiple regression model to cause the anomaly, it is possible to estimate the cause of the anomaly and speed up the decision on specific actions. As described above, the anomaly diagnosis method and anomaly diagnosis device according to the embodiment can detect an abnormal state from a large amount of measurement data acquired in a manufacturing process and quickly present measurement values ​​related to the anomaly in an easy-to-understand manner.

[0074] Although the abnormality diagnosis method and apparatus according to the present invention have been specifically described above using the preferred embodiment and examples, the gist of the present invention is not limited to these descriptions and should be broadly interpreted based on the claims. It goes without saying that various changes and modifications based on these descriptions are also included in the gist of the present invention.

[0075] REFERENCE SIGNS LIST 1 Abnormality diagnosis device 10 Input section 20 Output section 30 Storage section 31 Operation data DB 32 Normal operation data DB 40 Calculation section 41 Normal operation influence coefficient calculation section 42 Abnormality determination section 43 Operation influence coefficient calculation section 44 Abnormality cause estimation matrix calculation section 45 Abnormality cause variable identification section 46 Abnormality cause estimation section

Claims

1. An anomaly diagnosis method comprising: an operating variable acquisition step of acquiring a plurality of variables indicating a state of manufacturing equipment during operation; an anomaly determination step of determining whether or not an anomaly exists in the manufacturing equipment using a set of first prediction models constructed based on the plurality of variables acquired in the operating variable acquisition step and a plurality of variables indicating a state of the manufacturing equipment during normal operation, the first prediction models being used complementarily as a dependent variable and an explanatory variable; an operating influence coefficient calculation step of, if it is determined that an anomaly exists, constructing a set of second prediction models based on the plurality of variables acquired in the operating variable acquisition step, the second prediction models being used complementarily as a dependent variable and an explanatory variable; an anomaly cause estimation matrix calculation step of calculating an anomaly cause estimation matrix by comparing a first influence coefficient, which is an influence coefficient of an explanatory variable on a dependent variable for the set of first prediction models, with a second influence coefficient, which is an influence coefficient of an explanatory variable on a dependent variable for the set of second prediction models; and an anomaly cause variable identification step of identifying a variable that causes the anomaly using the anomaly cause estimation matrix.

2. The abnormality diagnosis method according to claim 1, wherein the abnormality cause estimation matrix has a tabular structure with the objective variable and the explanatory variable as key items, and has the comparison results of the first influence coefficient and the second influence coefficient as elements.

3. An anomaly diagnosis method according to claim 1 or 2, wherein the anomaly cause variable identification step uses the anomaly cause estimation matrix to focus on explanatory variables or target variables related to the anomaly, and identifies the variable that causes the anomaly based on changes in the influence coefficient related to the variable being focused on.

4. The abnormality diagnosis method according to claim 3, wherein in the abnormality cause variable identification step, a variable for which the change in the influence coefficient is greater than a predetermined threshold in both the column of the explanatory variable and the row of the target variable of the abnormality cause estimation matrix is ​​identified as the variable causing the abnormality.

5. An abnormality diagnosis method according to any one of claims 1 to 4, wherein the abnormality determination step calculates the degree of abnormality based on the prediction error of each variable using the first prediction model, and determines whether or not there is an abnormality in the manufacturing equipment by referring to the maximum, average, or sum of the degrees of abnormality of all or some of the variables.

6. An abnormality diagnosis method according to any one of claims 1 to 5, further comprising an abnormality cause estimation step of estimating the cause of the abnormality based on the identified variables.

7. The abnormality diagnosis method according to claim 6, wherein the abnormality cause estimation step estimates the cause of the abnormality by referring to a table of candidate causes of the abnormality according to combinations of the objective variable and explanatory variables that are candidates for the abnormality.

8. The anomaly diagnosis method according to claim 6, wherein the anomaly cause inference step infers the cause of the anomaly by referencing a table of candidate causes of the anomaly and corresponding actions according to combinations of the objective variable and explanatory variables that are candidates for the anomaly, and determines an action corresponding to the inferred cause of the anomaly.

9. The abnormality diagnosis method according to claim 8, wherein the abnormality cause estimation step outputs a command or an alert to a control device corresponding to the determined action.

10. An abnormality diagnosis device comprising: an abnormality determination unit that determines whether or not an abnormality exists in the manufacturing equipment using a plurality of variables that indicate the state of the manufacturing equipment during operation and a set of first prediction models that are constructed based on a plurality of variables that indicate the state of the manufacturing equipment under normal conditions, and in which the variables are used complementarily as dependent variables and explanatory variables; an operational influence coefficient calculation unit that, when an abnormality is determined to exist, constructs a set of second prediction models that use the variables complementarily as dependent variables and explanatory variables based on the plurality of variables that indicate the state of the manufacturing equipment during operation; an abnormality cause estimation matrix calculation unit that calculates an abnormality cause estimation matrix by comparing a first influence coefficient that is the influence coefficient of the explanatory variables on the dependent variable for the set of first prediction models with a second influence coefficient that is the influence coefficient of the explanatory variables on the dependent variable for the set of second prediction models; and an abnormality cause variable identification unit that identifies the variable that causes the abnormality using the abnormality cause estimation matrix.

11. The abnormality diagnosis device according to claim 10, further comprising an abnormality cause estimation unit that estimates the cause of the abnormality based on the identified variables.

Images