Abnormal monitoring method and abnormal monitoring device

Abstract

To provide an abnormality monitoring method and an abnormality monitoring device that indicate the basis when a model update is necessary and also enable automatic model updates.SOLUTION: The abnormality monitoring method is an abnormality monitoring method for monitoring an abnormality in a process using a model, including: an acquisition step (S2) that acquires first operational performance data including operational data and observed values for a first evaluation period during past normal operation and second operational performance data including operational data and observed values for a second evaluation period, including the current or most recent period; an influence degree calculation step (S7) in which an amount of drift indicating the difference in distribution between the first evaluation period and the second evaluation period regarding a predicted value or a degree of abnormality is calculated using the first operational performance data, the second operational performance data, and a model, and a degree of influence, which indicates a degree of influence, is calculated based on the amount of drift; and a factor identification step (S8) to identify cause of the prediction error based on the degree of influence.SELECTED DRAWING: Figure 2

Description

Technical Field

[0001] The present disclosure relates to an abnormality monitoring method and an abnormality monitoring device. In particular, the present disclosure relates to an abnormality monitoring method and an abnormality monitoring device that monitor the state of a manufacturing process, present factors when a predicted value becomes abnormal or when the degree of abnormality is high, and improve a prediction model or an abnormality detection model.

Background Art

[0002] Conventionally, methods for diagnosing abnormal states in manufacturing processes and the like have been proposed. Conventional process operation monitoring constructs a prediction model and evaluates the error between the prediction by the prediction model and the current state as the degree of deviation, as described in Patent Document 1 and Patent Document 2, for example.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0004] Here, when detecting abnormalities such as in a manufacturing process, as described above, it is generally performed to learn a normal state, create a prediction model, and diagnose an abnormality from the degree of deviation. However, when the state of a manufacturing process or the like is normal but there is a change in equipment (such as maintenance and repair), there is a problem that it is diagnosed as deviating from the learned normal state. When a change in the process state occurs due to equipment change, there may be a response to immediately update the prediction model. However, when a human manages the model, there is a problem that it is difficult to immediately respond to a change in the process state and the timing of updating is difficult. Also, although a method of automatically updating the model in a system based on the accuracy of the model has been proposed, there is also a problem that the reliability of the model decreases because the basis for model update becomes a black box.

[0005] An object of the present disclosure made to solve the above problems is to provide an abnormality monitoring method and an abnormality monitoring device that show a basis when model update is necessary and enable automatic model update.

Means for Solving the Problems

[0006] (1) An abnormality monitoring method according to an embodiment of the present disclosure is an abnormality monitoring method for monitoring an abnormality in a process using a model, an acquisition step of acquiring first operation performance data including operation data and observed values during a first evaluation period that is a past normal operation, and second operation performance data including the operation data and the observed values during a second evaluation period including the present or the most recent; an influence degree calculation step of calculating a drift amount indicating a difference in distribution between the first evaluation period and the second evaluation period for a predicted value or an abnormality degree using the first operation performance data, the second operation performance data, and the model, and calculating an influence degree indicating the degree of influence based on the drift amount; a factor identification step of identifying a factor of a prediction error based on the influence degree.

[0007] (2) As an embodiment of the present disclosure, in (1), When the degree of influence is equal to or greater than a predetermined threshold value, a model reconstruction step for reconstructing the model is included.

[0008] (3) As one embodiment of the present disclosure, in (2), The influence degree calculation step calculates the normalized influence degree, and the predetermined threshold value is one.

[0009] (4) As one embodiment of the present disclosure, in (2) or (3), The model reconstruction step When the second operation performance data includes data in a number that enables reconstruction of the model, the model is reconstructed. When the second operation performance data does not include data in a number that enables reconstruction of the model, the continuous use of the model is interrupted.

[0010] (5) As one embodiment of the present disclosure, in any one of (2) to (4), The model reconstruction step When the degree of influence is equal to or greater than a predetermined threshold value and information indicating that there is no abnormal operation is obtained, the model is reconstructed.

[0011] (6) As one embodiment of the present disclosure, in any one of (1) to (5), The model is a prediction model that uses operation data as an explanatory variable and an observed value of a state quantity indicating the state of a process as an objective variable.

[0012] (7) As one embodiment of the present disclosure, in (6), When the model is a multiple regression model, the degree of influence is calculated based on the importance that is the regression coefficient of the multiple regression model and the amount of drift.

[0013] (8) As one embodiment of the present disclosure, in any one of (1) to (5), The model is an abnormality detection model for performing abnormality detection.

[0014] (9) As one embodiment of the present disclosure, in any one of (1) to (8), including a step of acquiring event information, the influence degree calculation step calculates the influence degree based on the drift amount and the event information.

[0015] (10) As one embodiment of the present disclosure, in any one of (1) to (9), the first evaluation period and the second evaluation period are set by a change point detection method.

[0016] (11) An abnormality monitoring device according to an embodiment of the present disclosure is an abnormality monitoring device that monitors the abnormality of a process using a model, an acquisition unit that acquires first operation performance data including operation data and observed values in a first evaluation period that is a past normal operation time, and second operation performance data including the operation data and the observed values in a second evaluation period including the present or the most recent past; an influence degree calculation unit that calculates a drift amount indicating a difference in distribution between the first evaluation period and the second evaluation period for a predicted value or an abnormality degree using the first operation performance data, the second operation performance data, and the model, and calculates an influence degree indicating the degree of influence based on the drift amount; and a factor identification unit that identifies a factor of a prediction error based on the influence degree.

Advantages of the Invention

[0017] According to the present disclosure, it is possible to provide an abnormality monitoring method and an abnormality monitoring device that show a basis when model update is necessary and enable automatic model update.

Brief Description of the Drawings

[0018]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Mode for Carrying Out the Invention

[0019] Hereinafter, an anomaly monitoring method and an anomaly monitoring device according to an embodiment of the present disclosure will be described with reference to the drawings.

[0020] (Anomaly Monitoring Device) Figure 1 is a diagram showing a configuration example of the anomaly monitoring device 10 according to the present embodiment. The anomaly monitoring device 10 includes a communication unit 11, a storage unit 12, a control unit 13, and a display unit 14. The control unit 13 includes a period setting unit 21, an acquisition unit 22, an influence degree calculation unit 23, a factor identification unit 24, and a model reconstruction unit 25. Details of each component of the anomaly monitoring device 10 will be described later.

[0021] The anomaly monitoring device 10 monitors anomalies in a process. In the present embodiment, the process is a manufacturing process, specifically, a continuous casting process in the steel industry, but is not limited thereto. The anomaly monitoring device 10 can be used, for example, for anomaly monitoring of processes such as heating, cooling, and forming in the manufacture of various industrial products.

[0022] In the process in which the abnormality monitoring device 10 is used, the operation state is monitored using a model. The model is not limited as long as it monitors the operation state. For example, it includes a prediction model that outputs a predicted value and an abnormality detection model that performs abnormality detection. The abnormality detection model is a model that can calculate an abnormality degree (abnormality score) by, for example, the method of PCA (principal component analysis). Unless otherwise specified in the following description, the model is described as a prediction model that uses operation data as an explanatory variable and an observed value of a state quantity indicating the state of the process as an objective variable. However, the model may be an abnormality detection model, and by replacing the following "prediction model" with "abnormality detection model" and "predicted value" with "abnormality degree", the description of the embodiment in the abnormality detection model is obtained. The operation data is data measured in the operation of the process and is not limited to a specific one. Also, the observed value of the state quantity indicating the state of the process is data observed, for example, for the operator to confirm the normal operation of the process, and is not particularly limited as long as it is observable. Also, the state quantity is a physical quantity but is not limited to a specific one. The observed value of the state quantity indicating the state of the process may be, for example, a part of the operation data, may be a physical quantity of the same type as the operation data, or may be a measurement value completely different from the operation data. For example, when a slab is conveyed by a plurality of motors in a continuous casting process, the current value of a specific motor may be set as the above-mentioned observed value. Also, the current value of another motor that operates in cooperation with the specific motor may be measured as the above-mentioned operation data. And a prediction model that predicts the observed value from the operation data is generated, and in the operation of the continuous casting process, operation parameters (setting values of equipment) and the like may be adjusted based on the predicted value of the prediction model.

[0023] In the present embodiment, the abnormality monitoring device 10 communicates with a host system including a process computer that manages a continuous casting process, and acquires operation result data including operation data and observed values, a prediction model, and the like that are stored in a storage device by the host system. As another example, the abnormality monitoring device 10 may be a part of the process computer or the host system.

[0024] Details of each component of the abnormality monitoring device 10 will be described below. The communication unit 11 has a function of communicating by wire or wirelessly. The communication unit 11 transmits and receives necessary data and signals to and from a process computer or a higher-level system. For example, the abnormality monitoring device 10 may transmit a signal for requesting operation performance data during a specific period or receive operation performance data by the communication unit 11. The communication method performed by the communication unit 11 may be a wired communication standard or a wireless communication standard. For example, the wireless communication standard may include a communication standard for cellular phones such as 5G (5th Generation). Also, for example, the wireless communication standard may include IEEE802.11 or the like. The communication unit 11 can support one or more of these communication standards.

[0025] The storage unit 12 may have a function of storing various kinds of information. The storage unit 12 may store, for example, a program executed in the control unit 13, data used in the process executed in the control unit 13, and the result of the process. The storage unit 12 can be configured by, for example, a semiconductor memory or the like, but is not limited thereto and can be an arbitrary storage device. For example, the storage unit 12 may be an internal memory of a processor used as the control unit 13 or a hard disk drive (HDD) accessible from the control unit 13. In the present embodiment, the storage unit 12 may store operation performance data, a prediction model, etc. as data used in the process executed in the control unit 13.

[0026] The control unit 13 controls and manages each functional unit constituting the abnormality monitoring device 10 and the entirety of the abnormality monitoring device 10. The control unit 13 includes at least one processor such as a CPU (Central Processing Unit) in order to control and manage various functions. The control unit 13 may be composed of one processor or a plurality of processors. The processor constituting the control unit 13 may function as a period setting unit 21, an acquisition unit 22, an impact degree calculation unit 23, a factor identification unit 24, and a model reconstruction unit 25 by reading and executing a program from the storage unit 12.

[0027] The period setting unit 21 sets two evaluation periods. The period setting unit 21 sets a first evaluation period which is a past normal operation period, and a second evaluation period including the current or the most recent current time. Normal operation means that the operation of the process is carried out normally and the prediction model can calculate accurate prediction values with the required accuracy. The second evaluation period is a period after the first evaluation period and may partially overlap. The first evaluation period and the second evaluation period may be set before and after an event related to the equipment used in processes such as repair, replacement, maintenance, or start of actual operation. The event related to the equipment may include a blank (non-operating period) for a certain period for regular management. Also, the first evaluation period and the second evaluation period do not have to be of the same length. For example, the first evaluation period may be longer or shorter than the second evaluation period.

[0028] The acquisition unit 22 acquires first operation performance data including operation data and observed values in the first evaluation period, and second operation performance data including operation data and observed values in the second evaluation period. When the abnormality monitoring device 10 is not part of the process computer or the upper system, the acquisition unit 22 also acquires the prediction model from the upper system. Here, when the first operation performance data and the second operation performance data are not distinguished, they may be collectively referred to as operation performance data. As an example, the operation performance data may be acquired as time series data of the current value of the motor driving the pinch roll. The acquisition unit 22 may store the acquired operation performance data and the prediction model in the storage unit 12. Then, the operation performance data and the prediction model stored in the storage unit 12 may be read by the influence degree calculation unit 23, the factor identification unit 24, and the model reconstruction unit 25.

[0029] The influence degree calculation unit 23 calculates a drift amount indicating the difference in distribution between the first evaluation period and the second evaluation period for the explanatory variables, the objective variable, and the predicted value, using the first operation performance data, the second operation performance data, and the prediction model. Further, the influence degree calculation unit 23 calculates an influence degree indicating the degree of influence based on the drift amount for the explanatory variables and the objective variable.

[0030] Here, taking a prediction model that predicts the objective variable y using the explanatory variables x1, x2, and x3 as an example, the drift amount and the influence degree are specifically explained. However, the number of explanatory variables is not limited, and as an example, the case where there are three explanatory variables is explained. Table 2 shows the symbols used in the following explanation. For example, D yp is the predicted value y p or the drift amount of the prediction error. Here, the prediction error is the absolute value of the difference between the predicted value and the observed value (the actual measured value of the objective variable), which is calculated as |y p -y|.

[0031]

Table 1

[0032] The influence degree calculation unit 23 calculates the predicted value (y p ) for each of the first evaluation period and the second evaluation period, using the operation data (explanatory variables) of the first operation performance data, the operation data (explanatory variables) of the second operation performance data, and the prediction model. Hereinafter, the first evaluation period may be referred to as "T1" and the second evaluation period may be referred to as "T2". Further, in the present embodiment, the influence degree calculation unit 23 further calculates the prediction error (|y p -y|) for each of T1 and T2.

[0033] The influence degree calculation unit 23 calculates the explanatory variables (x1, x2, x3), the objective variable (y), and the prediction error (|y pFor each of (-y|), a drift amount indicating the difference in distribution between T1 and T2 is calculated. For calculating the drift amount, as a specific calculation method, Jensen-Shannon divergence or KL divergence may be used. Also, as another calculation method for calculating the drift amount, Histgram Intersection, L1 norm, L2 norm, Population Stability Index, Wasserstein distance, etc. may be used. The drift amount indicates the difference between T1 and T2 regarding the distributions of the explanatory variable, the objective variable, and the prediction error respectively.

[0034] In addition, the influence degree calculation unit 23 acquires the importance of the explanatory variables (x1, x2, x3) and the objective variable (y). The calculation method of the importance varies depending on the prediction model. For example, in the case of a multiple regression model created using values at the same level or standardized values, the regression coefficient can be treated as the importance of the corresponding explanatory variable. That is, if the prediction model is a multiple regression model represented by the following formula (1), the importance corresponding to the explanatory variables x1, x2, x3 is the regression coefficients a1, a2, a3 respectively. Also, in the case of such a multiple regression model, the importance of the objective variable can be set to "1".

[0035] [Number]

[0036] Here, the prediction model is not limited to the multiple regression model. For example, the prediction model may be a machine learning model. When the prediction model is a machine learning model, for example, by a method such as SHAP (SHapley Additive exPlanations), the contribution of each explanatory variable to the prediction result can be calculated. In this way, the influence degree calculation unit 23 acquires the importance of the explanatory variables (x1, x2, x3) and the objective variable (y) by a calculation method corresponding to the prediction model.

[0037] The influence degree calculation unit 23 calculates an influence degree indicating the degree of influence based on the drift amount for the explanatory variables (x1, x2, x3) and the target variable (y). Specifically, the influence degree is calculated, for example, by formulas (2) to (5) using the drift amount and the importance degree.

[0038]

Number

[0039] Here, the parameters of formulas (2) to (5) are as shown in Table 1. The influence degree is not limited to formulas (2) to (5) as long as it is an operation using the drift amount and the importance degree, and for example, it may be obtained by an operation including event information (importance degree of an event) described later. As another example of formulas (2) to (5), the calculation of the influence degree may use a formula in which the division by the drift amount (D yp ) of the prediction error is omitted. However, by dividing by the drift amount (D yp ) of the prediction error, it is possible to normalize so that the influence degrees can be compared regardless of the period settings of T1 and T2. As will be described later, when at least one influence degree is equal to or greater than a predetermined threshold, the prediction model is reconstructed. By normalization, only one predetermined threshold is set, and it is possible to determine whether or not to reconstruct the prediction model based on the same criterion. Therefore, it is preferable that the influence degree calculation unit 23 calculates a normalized influence degree. The influence degree calculation unit 23 may store the calculated drift amount and influence degree in the storage unit 12.

[0040] The cause identification unit 24 identifies the cause of the prediction error based on the influence degree. The cause identification unit 24 may directly obtain the influence degree from the influence degree calculation unit 23 or may obtain it via the storage unit 12. Specifically, the cause identification unit 24 may sort (rearrange) the influence degrees in descending order and identify the explanatory variables with high influence degrees as the cause of the prediction error. Also, when the target variable is included in those with high influence degrees, the cause identification unit 24 may identify that the prediction model itself has become unsuitable for the current operation of the process as the cause of the prediction error. The cause identification unit 24 may output the sorted influence degrees in descending order to the display unit 14 for display. The cause identification unit 24 may also output the identified cause of the prediction error to the display unit 14 for display.

[0041] The model reconstruction unit 25 reconstructs the prediction model when the influence degree is equal to or greater than a predetermined threshold. As described above, the influence degrees are calculated for a plurality of explanatory variables and the target variable, but the model reconstruction unit 25 reconstructs the prediction model when at least one influence degree is equal to or greater than a predetermined threshold. Also, when the influence degree calculated by the influence degree calculation unit 23 is a normalized influence degree, the predetermined threshold may be one. As another example, the model reconstruction unit 25 may set respective thresholds for a plurality of explanatory variables and the target variable. The model reconstruction unit 25 may directly obtain the influence degree from the influence degree calculation unit 23 or may obtain it via the storage unit 12.

[0042] Here, the model reconstruction unit 25 may reconstruct the prediction model when the influence degree is equal to or greater than a predetermined threshold and information indicating that the operation is normal is obtained. When the prediction error is caused by an abnormal operation and there is no problem with the prediction model itself, it is possible to avoid erroneously reconstructing the prediction model. The information indicating that the operation is normal may be, for example, an instruction to perform reconstruction from an operator in the manufacturing process. The operator may check the influence degree displayed on the display unit 14 and, when determining that there is no abnormality in the operation and it is okay to proceed with the reconstruction of the prediction model, may give an instruction to perform the reconstruction. The model reconstruction unit 25 may obtain the instruction to perform the reconstruction from the input device of the process computer via the communication unit 11.

[0043] When the model reconstruction unit 25 performs the reconstruction of the prediction model, it uses the second operation performance data, that is, the operation data of T2 including the current or the most recent current and the observed values. When the prediction model is a multiple regression model, the reconstruction may be an adjustment of the regression coefficients based on the operation data of T2 and the observed values. Further, when the prediction model is a learning model, the reconstruction may be re-learning using the operation data of T2 and the observed values as learning data. The model reconstruction unit 25 performs the reconstruction of the prediction model when the second operation performance data includes a number of data that enables the reconstruction of the prediction model. However, the model reconstruction unit 25 may interrupt the continuous use of the prediction model when the second operation performance data does not include a number of data that enables the reconstruction of the prediction model. That is, the model reconstruction unit 25 may stop the continued use of the unreconstructed prediction model in a state where the prediction error is large in the operation of the process. When the continuous use of the prediction model is interrupted, the model reconstruction unit 25 may wait until a number of data that enables the reconstruction of the prediction model is obtained, and perform the reconstruction of the prediction model after a sufficient number of data is obtained. Further, as another example, even when the second operation performance data does not include a number of data that enables the reconstruction of the prediction model, the model reconstruction unit 25 may correct the output (predicted value) of the prediction model so that the operation of the process using the prediction model can be continued. For example, when the prediction model is a linear regression model, the correction may be an operation such as multiplying the reciprocal of the influence degree for each explanatory variable.

[0044] When the model reconstruction unit 25 performs the reconstruction of the prediction model, it can output the reconstructed prediction model to the upper system to perform automatic model update. Even when the automatic model update is performed, the influence degree is displayed on the display unit 14 before the model update. Therefore, there is no problem that the basis of the model update becomes a black box and the reliability of the model decreases.

[0045] The display unit 14 displays the influence degrees sorted in descending order according to the control of the control unit 13. Further, the display unit 14 may also display the factors of the specified prediction error according to the control of the control unit 13. By displaying such information, the operator can grasp the factors affecting the prediction error. The display unit 14 may be a display device such as a liquid crystal display or an organic EL panel.

[0046] (Abnormal monitoring method) FIG. 2 is a flowchart illustrating the abnormal monitoring method according to the present embodiment. The abnormal monitoring device 10 executes processing according to the flowchart of FIG. 2.

[0047] The abnormal monitoring device 10 sets an evaluation period (step S1). Specifically, there are two evaluation periods. As described above, a first evaluation period that is the past normal operation time and a second evaluation period that includes the present or the most recent present are set.

[0048] The abnormal monitoring device 10 acquires operation performance data (step S2). Specifically, first operation performance data including operation data and observed values in the first evaluation period and second operation performance data including operation data and observed values in the second evaluation period are acquired. Here, step S2 corresponds to an acquisition step.

[0049] The abnormal monitoring device 10 acquires a prediction model (step S3). Further, for each of the first evaluation period and the second evaluation period, the abnormal monitoring device 10 calculates predicted values using the operation data (explanatory variables) of the first operation performance data, the operation data (explanatory variables) of the second operation performance data, and the prediction model (step S4).

[0050] The abnormal monitoring device 10 calculates a drift amount indicating the difference in distribution between the first evaluation period and the second evaluation period for each of the explanatory variable, the target variable, and the prediction error. The predicted value may be used instead of the prediction error.

[0051] The abnormality monitoring device 10 acquires the importance levels of the explanatory variables and the target variable (step S6). The abnormality monitoring device 10 calculates the influence degrees of the explanatory variables and the target variable by using the drift amount and the importance level (step S7). Here, step S7 corresponds to the influence degree calculation step.

[0052] The abnormality monitoring device 10 sorts the influence degrees in descending order and displays the sorted influence degrees on the display unit 14 (step S8). The abnormality monitoring device 10 may also display on the display unit 14 the factors of the identified prediction error. Here, step S8 corresponds to the factor identification step.

[0053] When at least one of the influence degrees is equal to or greater than a predetermined threshold (YES in step S9), the abnormality monitoring device 10 proceeds to the process of step S10. When all of the influence degrees are less than the predetermined threshold (NO in step S9), the abnormality monitoring device 10 ends the series of processes.

[0054] When the abnormality monitoring device 10 acquires information indicating that there is no operation abnormality (YES in step S10), it proceeds to the process of step S11. When the abnormality monitoring device 10 does not acquire information indicating that there is no operation abnormality (NO in step S10), it ends the series of processes.

[0055] When there is a sufficient number of data for reconstructing the prediction model (YES in step S11), the abnormality monitoring device 10 reconstructs the prediction model (step S12). When there is not a sufficient number of data for reconstructing the prediction model (NO in step S11), the abnormality monitoring device 10 interrupts the continued use of the prediction model (step S13). Here, step S12 corresponds to the model reconstruction step.

[0056] (Change point detection) Here, "T1", which is the first evaluation period, and "T2", which is the second evaluation period, may be automatically set by a change point detection method. The change point is the time that separates "T1" and "T2". The change point (τ) is calculated by Equation (A) and given as the time that maximizes the right side.

[0057] [Number]

[0058] Here, JS represents the Jensen-Shannon information measure. z is an anomaly index and is calculated by varying (drifting) τ from 1 to n. This term is not limited to the JS information measure as long as it is a change point detection method, and may be, for example, the likelihood ratio of distributions or the difference in likelihoods. f(x) represents the penalty for blank time. "T τ -T τ-1 " represents the difference between the time when the previous data was obtained and the current time in time series data, meaning the blank time without data. The penalty may be represented by a function proportional to the blank time, for example, but in this example, a logistic function is used. The logistic function is a time function suitable for the manufacturing process. In the manufacturing process, it is unlikely that a major event such as equipment replacement will occur during a blank time of, for example, about 1 to 3 hours. Therefore, for example, a is set to 1, K is set to 0.5, and x0 is set to 6. In this case, if the blank time is 3 hours or less, the penalty (the value of f(x)) becomes sufficiently small (for example, about 0.1). For example, if the blank time is 6 hours, the penalty increases to about 0.25. Also, if the blank time is 12 hours or more, the penalty exceeds 0.5.

[0059] Also, in Equation (A), regarding the penalty for blank time, the importance of the event (E r) is multiplied. For example, the acquisition unit 22 acquires event information from a higher-level system or the like. The impact degree calculation unit 23 may identify an event that occurred during the idle time from the acquired event information and further correct the penalty according to the importance of the event. That is, the impact degree may be calculated based on the drift amount and the event information. For example, for something with a large change such as a large-scale repair, the importance can be set high and a high penalty can be given. Also, for an event such as the suspension of an unimportant facility, the importance can be set low so that the penalty for the idle time is small or not considered. FIG. 6 shows an example of event information. For example, the importance may be defined for each event. Here, x1 to x3 in FIG. 6 are explanatory variables.

[0060] Here, the time of the calculated change point may be output to the display unit 14. Also, events and importance levels at timings close to the change point may be output to the display unit 14. By displaying this information, it becomes possible to present stronger grounds. For example, if a long stop occurred immediately before the time when the change point occurred, it is highly likely that the equipment was replaced or the settings were changed by repair. If data drift occurred at that timing and it is found that the cause was a change in settings due to repair, it becomes a useful basis for reconstructing the model.

[0061] (Example 1) Hereinafter, the effects of the present disclosure will be specifically described based on examples, but the present disclosure is not limited to the contents of the examples.

[0062] In this example, the abnormality monitoring device 10 monitors an abnormality in the current of a pinch roll motor for pulling out a slab in a continuous casting process (hereinafter, sometimes simply referred to as "motor current"). Also, in this example, the operation of the continuous casting process is performed using a prediction model that is a multiple regression model (multiple regression prediction formula) having three explanatory variables. The prediction model is as shown in the following formula (6).

[0063]

Equation

[0064] FIG. 3 is a diagram showing the distribution of operation performance data, in which the transitions of three explanatory variables and an objective variable in a first evaluation period (T1) and a second evaluation period (T2) are shown. The number of slabs on the horizontal axis is the number assigned in the production order for the slabs produced in each period. 500 slabs are produced in each of T1 and T2, and the values of the motor current at the time of production of each slab are plotted. The same applies to the number of slabs in FIGS. 4 and 5 described later, but in FIG. 5, all the slabs produced in a long period including T1, T2, and the period therebetween are numbered in order. Here, as the value of the motor current, the average current during the use of the target motor at the time of production of each slab may be used, but it is not limited thereto. The value of the motor current may be a representative value indicating the state during the use of the target motor, such as the maximum current, the instantaneous value at a point, the minimum current, the standard deviation, or the difference between the maximum current and the minimum current, as long as it can be such a value.

[0065] The above formula (6) is a multiple regression prediction formula constructed based on the operation performance data of T1. FIG. 4 shows the predicted values obtained using the formula (6) in each of T1 and T2, and the prediction errors are calculated and plotted. As shown in FIG. 4, the prediction error is large in T2. It is considered that there is data drift in either the objective variable or the explanatory variable, resulting in a decrease in prediction accuracy.

[0066] Table 2 shows the drift amount, importance, and influence degree calculated or acquired by the anomaly monitoring device 10 according to the above embodiment. In this embodiment, the drift amount is calculated using the KL information amount. Also, the importance is calculated using the coefficients of the multiple regression prediction formula. Further, the influence degree is calculated according to formulas (2) to (5).

[0067]

Table 2

[0068] From Table 2, it can be seen that the explanatory variable x3 mainly affects the factors that cause the prediction error to increase. Also, the influence on the target variable y is significant. Figure 5 shows the data fluctuations over a long period including T1, T2, and the period in between. It can be seen that the distribution of x3 changes significantly at the part with about 4500 slab numbers as the boundary. Also, the transition of y changes at the same timing. Therefore, it is found that the abnormal monitoring device 10 according to the above embodiment correctly indicates the factors of the prediction error. Here, it is also possible for the operator to identify the cause by visually checking the data fluctuations as shown in Figure 5. However, it is not realistic for the operator to continue monitoring over a long period, and the problem can be solved by using the abnormal monitoring device 10. Here, in this embodiment, a predetermined threshold is set to 1, and the reconstruction of the prediction model is performed at T2.

[0069] (Example 2) As Example 2, abnormal monitoring of adjacent motor currents in a continuous casting machine using principal component analysis was performed. In principal component analysis, there is no target variable, and monitoring is performed based on the abnormal score by the Q statistic. In this example, abnormal monitoring was performed using 11 variables. Figure 7 shows the transition of the abnormal score. In this example, "T1" and "T2" were provided before and after the change points automatically determined by change point detection. Table 3 shows the drift amount, importance, content of the event, importance of the event, and influence degree. In this example, the influence degree was calculated as "(drift amount)×(importance)×(importance of the event)". As shown in Table 3, the abnormality in the motor current monitoring is mainly caused by the variable x8, and it can be seen that data drift occurred due to component replacement by repair. By displaying the factors as shown in Figure 8, for example, it is possible to update the model while showing many bases.

[0070]

Table 3

[0071] As described above, the anomaly monitoring device 10 and the anomaly monitoring method according to the present embodiment can indicate the degree of influence as a basis when model update is necessary by the above-described configuration and steps. Further, since the basis for model update does not become a black box and the problem of deterioration of the reliability of the model does not occur, the anomaly monitoring device 10 and the anomaly monitoring method according to the present embodiment also enable automatic model update.

[0072] Although the embodiments of the present disclosure have been described based on the drawings and examples, it should be noted that those skilled in the art can easily make various modifications or corrections based on the present disclosure. Therefore, it should be noted that these modifications or corrections are included in the scope of the present disclosure. For example, the functions included in each component or each step can be rearranged so as not to be logically contradictory, and it is possible to combine or divide a plurality of components or steps into one. The embodiments according to the present disclosure can also be realized as a program executed by a processor included in the device or a storage medium recording the program. It should be understood that these are also included in the scope of the present disclosure.

Description of Reference Numerals

[0073] 10 Anomaly monitoring device 11 Communication unit 12 Storage unit 13 Control unit 14 Display unit 21 Period setting unit 22 Acquisition unit 23 Degree of influence calculation unit 24 Factor identification unit 25 Model reconstruction unit

Claims

An anomaly monitoring method for monitoring process anomalies using a prediction model that is a multiple regression model with operation data as explanatory variables and observed values of state quantities indicating the state of the process as target variables, comprising: an acquisition step of acquiring first operation performance data including operation data and observed values in a first evaluation period that is a past normal operation period, and second operation performance data including the operation data and the observed values in a second evaluation period including the present or the most recent past; an influence degree calculation step of calculating, using the first operation performance data, the second operation performance data, and the prediction model, a drift amount indicating a difference in distribution between the first evaluation period and the second evaluation period for predicted values, and multiplying the drift amount by the importance degree that is the regression coefficient of the prediction model to calculate the influence degree for the explanatory variables and the target variables indicating the magnitude of the influence on the prediction error that is the difference between the predicted value and the observed value; a factor identification step of identifying factors of the prediction error based on the influence degree; The drift amount is calculated using the JS information amount or the KL information amount for the explanatory variables and the target variables. An anomaly monitoring method.

2. The anomaly monitoring method according to claim 1, further comprising a model reconstruction step of reconstructing the prediction model when the influence degree is equal to or greater than a predetermined threshold value.

3. The anomaly monitoring method according to claim 2, wherein the influence degree calculation step calculates the normalized influence degree, and the predetermined threshold value is one.

4. The model reconstruction step performs reconstruction of the prediction model when the second operation performance data includes a number of data that enables reconstruction of the prediction model; interrupts continuous use of the prediction model when the second operation performance data does not include a number of data that enables reconstruction of the prediction model. The anomaly monitoring method according to claim 2 or 3.

5. The model reconstruction step performs reconstruction of the prediction model when the influence degree is equal to or greater than a predetermined threshold value and an instruction for reconstruction from an operator is obtained. The anomaly monitoring method according to claim 2 or 3.

6. including a step of acquiring event information; The anomaly monitoring method according to any one of claims 1 to 3, wherein the influence degree calculation step calculates the influence degree based on the drift amount and the event information.

7. The first evaluation period and the second evaluation period are set based on a change point τ which is the time when the right side of the following formula (A) is maximized. In formula (A), JS represents the Jensen-Shannon information amount, z represents an anomaly index, f represents a function indicating a penalty for blank time, Tτ − Tτ−1 represents the difference between the time when the previous data was obtained and the current time in time-series data, and Er represents the importance of an event. The anomaly monitoring method according to any one of claims 1 to 3. 【Number 1】

8. An anomaly monitoring device that monitors process anomalies using a prediction model that is a multiple regression model with operating data as an explanatory variable and an observed value of a state quantity indicating the state of the process as an objective variable, an acquisition unit that acquires first operation performance data including operating data and observed values in a first evaluation period which is a past normal operation time, and second operation performance data including the operating data and the observed values in a second evaluation period including the present or the most recent past; an influence degree calculation unit that calculates, using the first operation performance data, the second operation performance data, and the prediction model, a drift amount indicating a difference in distribution between the first evaluation period and the second evaluation period for a predicted value, and multiplies the drift amount by the importance which is the regression coefficient of the prediction model, to calculate an influence degree for the explanatory variable and the objective variable, indicating the magnitude of the influence on the prediction error which is the difference between the predicted value and the observed value; a factor identification unit that identifies the factor of the prediction error based on the influence degree; The anomaly monitoring device, wherein the drift amount is calculated using the JS information amount or the KL information amount for the explanatory variable and the objective variable.

Images