Abnormality diagnosis model construction method, abnormality diagnosis method, abnormality diagnosis model construction device, and abnormality diagnosis device

Abstract

The method for constructing an abnormality diagnosis model is a method for constructing an abnormality diagnosis model for a process in which a plurality of devices successively process a metal material, including: a first model creation step in which a first abnormality diagnosis model that learns a relationship between a measurement value at a time and an abnormality is created using measurement values measured at the same time at predetermined regular measurement cycles for the plurality of devices; and a second model creation step in which a second abnormality diagnosis model that learns a relationship between a measurement value at the same position of the metal material and an abnormality is created using measurement values at the same position of the metal material edited for measurement values measured for the plurality of devices for each position of the metal material.

Description

Technical Field

[0001] This invention relates to a method for constructing an abnormality diagnostic model, an abnormality diagnostic method, an apparatus for constructing an abnormality diagnostic model, and an abnormality diagnostic apparatus. Background Technology

[0002] As methods for diagnosing the manufacturing status, especially abnormal status, of a manufacturing process, there are model-based methods and data-based methods. Model-based methods construct models that mathematically express the physical or chemical phenomena in the manufacturing process and use these models to diagnose the manufacturing status. On the other hand, data-based methods construct statistical analytical models based on operational data obtained during the manufacturing process and use these models to diagnose the manufacturing status.

[0003] In manufacturing processes like steelmaking, a single production line produces a variety of products in many sizes, resulting in numerous operational modes. Furthermore, in processes like blast furnaces, which use natural materials such as iron ore and coke as raw materials, the manufacturing process is prone to significant variations. Therefore, model-based methods alone have limitations in diagnosing the manufacturing status of processes like steelmaking.

[0004] As a data-based approach, there are diagnostic methods that database the operational data from past anomalies to determine similarities with current operational data, or conversely, that database the operational data from normal times to determine differences with current operational data. However, in manufacturing processes such as steelmaking, in addition to the large number of machines used, especially in countries like Japan where many machines are aging rapidly, unprecedented failures are not uncommon. Therefore, diagnostic methods based on past failure examples have limitations in predicting abnormal states.

[0005] On the other hand, as a diagnostic method for the latter (a diagnostic method using normal operating data), there are methods described in Patent Documents 1 to 4. Specifically, Patent Documents 1 and 2 describe methods for predicting or detecting abnormal states in the manufacturing process based on predictions made using models created using normal operating data. Furthermore, Patent Documents 3 and 4 describe methods for extracting templates from normal operating data, creating program libraries, and determining the differences between the acquired operating data and the program-library templates, thereby detecting situations different from usual in advance.

[0006] Patent Document 1: International Publication No. 2013 / 011745

[0007] Patent Document 2: Japanese Patent No. 4922265

[0008] Patent Document 3: Japanese Patent No. 5651998

[0009] Patent Document 4: Japanese Patent No. 5499900

[0010] In steel manufacturing processes, measurements from multiple sensors are processed, and most of these measurements are taken for operational management and equipment control. Therefore, as in Patent Documents 1-4, it may not be possible to obtain sufficient measurements that directly indicate the equipment's condition or the cause of any malfunction. Furthermore, even if measurements exist that correspond one-to-one with the cause of such malfunctions, it is still impossible to encompass all malfunctions using only these measurements.

[0011] On the other hand, thanks to the development of data collection and analysis technologies in recent years, environments for processing massive amounts of data, known as big data, are becoming increasingly sophisticated. Given this situation, it is essential to comprehensively and accurately detect anomalies from large amounts of data, identify data related to these anomalies, and maintain rapid maintenance actions to ensure stable operation. In this case, the large amount of data does not necessarily correspond one-to-one with the causes of anomalies as described in Patent Documents 1-4; therefore, the identified anomaly may not be singular. Summary of the Invention

[0012] The present invention was made in view of the above, and its purpose is to provide a method for constructing an anomaly diagnosis model, an anomaly diagnosis method, an anomaly diagnosis model construction apparatus, and an anomaly diagnosis apparatus that can detect abnormal states based on a large number of measurement values ​​obtained in the manufacturing process and identify the cause of the anomaly in advance by classifying and screening multiple candidates of causes of anomalies that do not correspond one-to-one.

[0013] To address the aforementioned issues and achieve the objectives, the present invention relates to a method for constructing an anomaly diagnosis model for a process of sequentially processing metal materials using multiple devices. This method includes: a first model creation step, in which a first anomaly diagnosis model is created by using measurement values ​​measured simultaneously at predetermined measurement cycles for the multiple devices to learn the relationship between the simultaneous measurement values ​​and anomalies; and a second model creation step, in which a second anomaly diagnosis model is created by using measurement values ​​at each corresponding location of the metal material, obtained by editing the measurement values ​​measured at the multiple devices according to each location of the metal material, to learn the relationship between the measurement values ​​at the same location and anomalies.

[0014] Furthermore, the method for constructing the abnormal diagnosis model involved in this invention is based on the above invention, wherein the above-mentioned metal material is a rolled material and the above-mentioned equipment is a rolling mill.

[0015] Furthermore, the method for constructing the abnormal diagnosis model involved in this invention is based on the above invention. In this method, the position of the rolled material on the non-final mill's delivery side is calculated based on the ratio of the total length of the rolled material on the final mill's delivery side to the total length of the rolled material on the non-final mill's delivery side, and the measurement value at the same position is calculated accordingly.

[0016] Furthermore, the method for constructing the abnormal diagnosis model involved in this invention is based on the above invention, wherein the total length of the rolled material on the delivery side of the mill is calculated by calculating the through-plate speed of the rolled material based on the roll speed and the forward slip ratio of the mill and integrating the through-plate speed over time.

[0017] To address the aforementioned issues and achieve the objectives, the anomaly diagnosis method of this invention utilizes an anomaly diagnosis model constructed using the aforementioned anomaly diagnosis model construction method. The method includes: a first anomaly diagnosis step, in which anomaly diagnosis is performed by inputting data representing the relationship between measured values ​​at the same instant into the first anomaly diagnosis model; a second anomaly diagnosis step, in which anomaly diagnosis is performed by inputting data representing the relationship between measured values ​​at the same location into the second anomaly diagnosis model; and an anomaly determination step, in which the cause of the anomaly is determined based on the diagnosis results from the first and second anomaly diagnosis steps.

[0018] Furthermore, the abnormal diagnosis method involved in this invention is completed on the basis of the above invention. In this method, the abnormal determination step is based on an abnormal diagnosis table to determine the cause of the abnormality. The abnormal diagnosis table establishes a correlation between the first diagnosis result and the second diagnosis result to indicate the cause of the abnormality.

[0019] To address the aforementioned issues and achieve the objectives, the anomaly diagnosis model construction apparatus of the present invention is an anomaly diagnosis model construction apparatus for a process of sequentially processing metal materials using multiple devices. It comprises: a first model creation unit that creates a first anomaly diagnosis model by using measurement values ​​measured simultaneously at the same time for the multiple devices at a predetermined measurement cycle, which learns the relationship between the simultaneous measurement values ​​and anomalies; and a second model creation unit that creates a second anomaly diagnosis model by using measurement values ​​at each same location of the metal material, obtained by editing the measurement values ​​measured at the multiple devices according to each location of the metal material, which learns the relationship between the measurement values ​​at the same location and anomalies.

[0020] To address the aforementioned issues and achieve the objectives, the anomaly diagnosis device of the present invention is an anomaly diagnosis device that uses an anomaly diagnosis model constructed by the aforementioned anomaly diagnosis model construction device. It comprises: a first anomaly diagnosis unit that performs anomaly diagnosis by inputting data representing the relationship between measured values ​​at the same moment into the first anomaly diagnosis model; a second anomaly diagnosis unit that performs anomaly diagnosis by inputting data representing the relationship between measured values ​​at the same location into the second anomaly diagnosis model; and an anomaly determination unit that determines the cause of the anomaly based on the diagnosis results from the first anomaly diagnosis unit and the diagnosis results from the second anomaly diagnosis unit.

[0021] The present invention relates to a method for constructing an anomaly diagnosis model, an anomaly diagnosis method, an anomaly diagnosis model construction device, and an anomaly diagnosis device that can construct two anomaly diagnosis models, thereby distinguishing between anomalies caused by the mechanical control system of equipment and anomalies caused by product quality and shape for diagnosis. Therefore, the causes of anomalies can be identified in advance, equipment downtime can be reduced, and efficient and effective anomaly countermeasures can be maintained. Attached Figure Description

[0022] Figure 1 This is a diagram illustrating a simplified structure of the anomaly diagnosis device and model building device involved in the embodiments of the present invention.

[0023] Figure 2 This is a flowchart illustrating the steps of creating and processing real-time relational data in the anomaly diagnosis device and model building device according to embodiments of the present invention.

[0024] Figure 3 This is a flowchart illustrating the steps of creating and processing location-related data in the anomaly diagnosis device and model building device according to embodiments of the present invention.

[0025] Figure 4 This is a flowchart illustrating the steps of an anomaly diagnosis method performed by an anomaly diagnosis device according to an embodiment of the present invention. Detailed Implementation

[0026] The method for constructing an anomaly diagnosis model, the anomaly diagnosis method, the apparatus for constructing an anomaly diagnosis model, and the anomaly diagnosis apparatus according to embodiments of the present invention will be described with reference to the accompanying drawings.

[0027] [Anomaly Diagnosis Device and Model Building Device]

[0028] First, refer to Figure 1The structure of the anomaly diagnosis device and the anomaly diagnosis model construction device (hereinafter referred to as the "model construction device") according to embodiments of the present invention will be described. The anomaly diagnosis device and the model construction device are applied in a process of sequentially processing metallic materials using multiple devices. In this embodiment, an example of application to a rolling process in which a rolled material such as a steel plate is rolled sequentially using multiple rolling mills will be described. Furthermore, in the following description, the rolling mill will also be referred to as a "stand".

[0029] The anomaly diagnosis device 1 includes an input unit 10, a storage unit 20, a calculation unit 30, and a display unit 40. Furthermore, the "model building device" is realized by the components of the anomaly diagnosis device 1, excluding the anomaly determination table 25, the simultaneous time relationship diagnosis unit 34, the same position relationship diagnosis unit 35, and the anomaly determination unit 36.

[0030] The input unit 10 is an input unit for the arithmetic unit 30. It receives the actual operating data (timing data) of the diagnostic target device via the information control system network and inputs it to the arithmetic unit 30 in a prescribed format.

[0031] The storage unit 20 consists of recording media such as EPROM (Erasable Programmable ROM), hard disk drive (HDD), and removable media. Examples of removable media include disc-type recording media such as USB (Universal Serial Bus) memory, CD (Compact Disc), DVD (Digital Versatile Disc), and BD (Blu-ray Disc). The storage unit 20 can store operating systems (OS), various programs, various tables, various databases, etc.

[0032] The storage unit 20 stores the following data: Simultaneous moment relationship data definition table 21, same position relationship data definition table 22, simultaneous moment relationship diagnostic model (first anomaly diagnostic model) 23, same position relationship diagnostic model (second anomaly diagnostic model) 24, and anomaly judgment table 25.

[0033] Simultaneous time relationship data definition table 21 is a table that records the settings required by the simultaneous time relationship data creation unit 31 to create simultaneous time relationship data. Simultaneous time relationship data definition table 21 is, for example, a table as shown in Table 1 below.

[0034] [Table 1]

[0035] (Table 1)

[0036]

[0037] As shown in Table 1, in the Simultaneous Relationship Data Definition Table 21, the signals indicating the start and end of the cut are specified according to the cut object signal of each time-series data related to actual operation.

[0038] The same position relationship data definition table 22 is a table that describes the settings required by the same position relationship data creation unit 32 to create same position relationship data. The same position relationship data definition table 22 is, for example, a table as shown in Table 2 below.

[0039] [Table 2]

[0040] (Table 2)

[0041]

[0042] As shown in Table 2, in the same position relationship data definition table 22, the following signals are specified according to the cutting object signal of each time sequence data related to actual operation.

[0043] (1) Signals indicating the start and end of the cut.

[0044] (2) The signal indicating the roll speed of the rolling mill to which the signal belongs.

[0045] (3) The signal indicating the forward slip ratio of the rolling mill to which the signal belongs.

[0046] Simultaneous moment relationship diagnostic model 23 is the model referenced in the anomaly diagnosis processing of simultaneous moment relationship data performed by simultaneous moment relationship diagnostic unit 34. Simultaneous moment relationship diagnostic model 23 is a learned model that has learned the relationship between simultaneous moment measurements and anomalies, and is created by model creation unit 33, which will be described later.

[0047] The co-location relationship diagnostic model 24 is the model referenced in the anomaly diagnosis processing of co-location relationship data performed by the co-location relationship diagnostic unit 35. The co-location relationship diagnostic model 24 is a learned model that has learned the relationship between co-location measurements and anomalies, and is created by the model creation unit 33, which will be described later.

[0048] The anomaly determination table 25 is a table referenced in the anomaly determination process performed by the anomaly determination unit 36. The anomaly determination table 25 is created based on past operational examples, and for example, as shown in Table 3 below, it consists of a classification table of the causes of anomalies.

[0049] The anomaly determination table 25 shown in Table 3 contains the following related information sequentially from the left column.

[0050] (1) The reasons for anomalies observed as anomalies, and in cases where anomalies are determined based on contemporaneous relationship data.

[0051] (2) Reasons for anomalies when anomaly detection is performed based on data with the same positional relationship.

[0052] (3) Reasons for anomalies when anomaly detection is performed based on simultaneous temporal relationship data and co-location relationship data.

[0053] [Table 3]

[0054] (Table 3)

[0055]

[0056] For example, if the results of the anomaly diagnosis processing of the synchronous relationship data performed by the synchronous relationship diagnosis unit 34 and the anomaly diagnosis processing of the same position relationship data performed by the same position relationship diagnosis unit 35 are as follows, the anomaly determination unit 36 ​​determines the cause of the anomaly of the differential load, rolling load and inter-stand tension as follows.

[0057] The results of the anomaly diagnosis and processing of the real-time relationship data are as follows: "Differential load: Anomaly present, Rolling load: Anomaly present, Inter-stand tension: No anomaly present."

[0058] The results of the anomaly diagnosis and processing of the positional relationship data are as follows: "Differential load: Anomaly present; Rolling load: No anomaly present; Inter-stand tension: Anomaly present."

[0059] Cause of abnormal differential load: "Faulty pressure gauge"

[0060] The cause of abnormal rolling load: "Excessive or insufficient reduction compensation control".

[0061] The cause of abnormal tension between frames: "Excessive variation in plate thickness"

[0062] In the anomaly diagnosis device 1 according to the embodiment, an anomaly determination table 25 is prepared in advance for the cases where anomalies are detected based on simultaneous relationship data, the cases where anomalies are detected based on same position relationship data, and the cases where anomalies are detected based on data from both parties. Furthermore, in the anomaly determination process, the causes of anomalies are classified by referring to the anomaly determination table 25.

[0063] The arithmetic unit 30 is implemented, for example, by a processor consisting of a CPU (Central Processing Unit) and a memory consisting of RAM (Random Access Memory) and ROM (Read Only Memory) (main storage unit).

[0064] The arithmetic unit 30 loads the program into the working area of ​​the main storage unit for execution. Through program execution, it controls various structural units, thereby achieving the intended function. The arithmetic unit 30 functions as the simultaneous-moment relationship data creation unit 31, the same-location relationship data creation unit 32, and the model creation unit (first and second model creation units) 33 through the execution of the aforementioned program. Furthermore, the arithmetic unit 30 functions as the simultaneous-moment relationship diagnosis unit (first anomaly diagnosis unit) 34, the same-location relationship diagnosis unit (second anomaly diagnosis unit) 35, and the anomaly determination unit (anomaly determination unit) 36 through the execution of the aforementioned program. In addition, in... Figure 1 For example, an example is shown in which the functions of each part are implemented by a single computer, but the unit implementing the functions of each part is not particularly limited. For example, the functions of each part can also be implemented by multiple computers respectively.

[0065] Simultaneous time-series data creation unit 31 processes the time-series data input from input unit 10 to create simultaneous time-series relationship data representing the relationship between measurements at simultaneous times. Specifically, simultaneous time-series relationship data creation unit 31 creates simultaneous time-series relationship data with reference to simultaneous time-series relationship data definition table 21. Furthermore, simultaneous time-series relationship data creation unit 31 creates simultaneous time-series relationship data in two scenarios: when creating simultaneous time-series relationship diagnostic model 23 offline and when performing anomaly diagnosis online (while running) using simultaneous time-series relationship diagnostic model 23. Hereinafter, refer to... Figure 2 An example illustrating a method for creating real-time relational data is provided.

[0066] First, the simultaneous time-relationship data creation unit 31 retrieves the rolling start time of the start stand that determines the start of anomaly diagnosis and the rolling end time of the final stand that determines the end of anomaly diagnosis from the time-series data related to actual operation (steps S1 and S2). In steps S1 and S2, for example in the case of a finishing continuous rolling mill consisting of 7 stands, stand F7 is defined as the start stand and stand F1 is defined as the end stand, thereby enabling the extraction of simultaneous time-relationship data for the constant portion of the rolled material (coil) in all stands.

[0067] Next, the simultaneous time relationship data creation unit 31 extracts the rolling data (sensor data) of each stand from the retrieved start time to the end time (step S3), thereby creating simultaneous time relationship data (step S4).

[0068] The location relationship data creation unit 32 processes the time-series data input from the input unit 10 to create location relationship data representing the relationship between measurements at the same location. Specifically, the location relationship data creation unit 32 creates location relationship data with reference to the location relationship data definition table 22. Furthermore, the location relationship data creation unit 32 creates location relationship data in two scenarios: when creating the location relationship diagnostic model 24 offline and when performing anomaly diagnosis online (while running) using the location relationship diagnostic model 24. Hereinafter, refer to... Figure 3 An example illustrating the method for creating data with the same positional relationship is provided.

[0069] First, the co-positional relationship data creation unit 32 retrieves the rolling start time and rolling end time of each stand as the diagnostic target from the time-series data related to actual operation (steps S11 and S12). Next, the co-positional relationship data creation unit 32 extracts the rolling data (sensor data), rolling speed (mill speed), and forward slip rate data of the target stand during the period from the retrieved start time to the end time (steps S13 and S14).

[0070] Next, the positional relationship data creation unit 32 calculates the total length of the rolled material in each stand (the total length of the rolled material on the delivery side of each mill) based on the extracted data (step S15). For example, if the rolling start time of the i-th stand is set to t... i0 Set the forward slip ratio at time t to f. i (t) Set the rolling speed (hereinafter referred to as "rolling speed") of the rolled material to v. i If (t), then the position Li(t) of the rolled material from the front end, obtained at time t, can be represented by the following equation (1). Furthermore, the through-plate speed of the rolled material can be calculated based on the mill's roll speed and forward slip ratio.

[0071] [Formula 1]

[0072]

[0073] In equation (1) above, t is set as the end time of rolling, thereby enabling the calculation of the total length of the rolled material in the mill stand. Furthermore, by focusing on the final mill stand, the total length of the final product can be calculated.

[0074] Here, when data collected from different stands is merged into co-positional relationship data, the total length of the rolled material in each stand is different. Therefore, the co-positional relationship data creation unit 32 sets the length of the rolled material as the reference for the final product, i.e., the final coil length. Furthermore, the position of the rolled material on the feed side of the target mill is calculated based on the ratio of the total length of the rolled material on the feed side of the final stand to the total length of the rolled material on the feed side of the target stand (stands other than the final stand) (step S16). Thus, co-positional relationship data is created (step S17).

[0075] Furthermore, for example, when rolling data is collected through fixed-period sampling, the resulting data is not evenly spaced over the entire length of the rolled material. Therefore, in this case, interpolation processing is required to obtain equally spaced co-positional relationship data. As interpolation methods in this situation, linear interpolation can be used when considering two adjacent points, and spline interpolation can be used when considering three or more points.

[0076] In this way, during the creation of positional relationship data in the positional relationship data creation unit 32, the total length of the rolled material is calculated by time integration of the material's through-plate speed, calculated based on the mill's roll speed and forward slip ratio. Furthermore, based on the ratio of the total length of the rolled material calculated similarly from the final stand to the total length of the rolled material in the target stand, positional data corresponding to the delivery side of the final stand and related to the rolling direction of the target stand is created. Thus, data measured in different stands can be edited into positional data related to the rolling direction from the front end of the final product.

[0077] The model creation unit 33 uses measurement values ​​measured simultaneously at the same time for multiple rolling mills with a predetermined measurement cycle to create a simultaneous relationship diagnostic model 23 that learns the relationship between simultaneous measurement values ​​and anomalies. Furthermore, the aforementioned "measurement value" refers to the simultaneous relationship data created by the simultaneous relationship data creation unit 31.

[0078] Furthermore, the model creation unit 33 creates a co-location relationship diagnostic model 24 by editing the measurement values ​​obtained from multiple rolling mills for each position of the rolled material, which learns the relationship between the measurement values ​​at the same position and the anomalies. Additionally, the "measurement value" mentioned above refers to the co-location relationship data created by the co-location relationship data creation unit 32.

[0079] Furthermore, the model creation method in the model creation unit 33 is not particularly limited. As a model creation method, methods such as anomaly detection based on the magnitude of the deviation between the predicted quantity derived from the regression model and the actual quantity, or anomaly detection based on the recovery error of a generative model based on an autoencoder, etc., can be used. Examples of methods for the former include linear regression, local regression, Lasso regression, Ridge regression, principal component regression, PLS regression, neural networks, regression trees, random forests, XGBoost, etc. Additionally, the model creation unit 33 stores the created simultaneous relationship diagnostic model 23 and co-location relationship diagnostic model 24 in the storage unit 20, respectively.

[0080] Simultaneous time-relationship diagnosis unit 34 performs anomaly diagnosis using simultaneous time-relationship diagnosis model 23. Simultaneous time-relationship diagnosis unit 34 performs anomaly diagnosis by inputting simultaneous time-relationship data created by simultaneous time-relationship data creation unit 31 into simultaneous time-relationship diagnosis model 23. The anomaly diagnosis results of simultaneous time-relationship diagnosis unit 34 are, for example, combinations of the observed items as anomalies (refer to the left column of Table 3) and the presence or absence of anomalies for those items, such as "Differential load: anomaly present / absent, rolling load: anomaly present / absent, inter-stand tension: anomaly present / absent". Furthermore, in the anomaly diagnosis performed by simultaneous time-relationship diagnosis unit 34, anomalies in the mechanical control system of the equipment can be primarily detected.

[0081] The positional relationship diagnosis unit 35 performs anomaly diagnosis using the positional relationship diagnosis model 24. The positional relationship diagnosis unit 35 performs anomaly diagnosis by inputting positional relationship data created by the positional relationship data creation unit 32 into the positional relationship diagnosis model 24. The anomaly diagnosis results of the positional relationship diagnosis unit 35 are, for example, combinations of the observed items as anomalies (see Table 3) and the presence or absence of anomalies for those items, such as "Differential load: Anomaly present / absent, Rolling load: Anomaly present / absent, Inter-stand tension: Anomaly present / absent". Furthermore, in the anomaly diagnosis performed by the positional relationship diagnosis unit 35, anomalies caused primarily by product quality and shape can be detected.

[0082] The anomaly determination unit 36 ​​determines the cause of the anomaly based on the diagnostic results (first diagnostic results) in the simultaneous relationship diagnosis unit 34 and the diagnostic results (second diagnostic results) in the same position relationship diagnosis unit 35. The anomaly determination unit 36 ​​determines the cause of the anomaly based on the anomaly determination table 25, which establishes a correlation between the diagnostic results in the simultaneous relationship diagnosis unit 34 and the diagnostic results in the same position relationship diagnosis unit 35 to indicate the cause of the anomaly.

[0083] Specifically, the anomaly determination unit 36 ​​determines the presence or absence of an anomaly based on the diagnostic results from the simultaneous relationship diagnosis unit 34 and the same position relationship diagnosis unit 35. Next, the anomaly determination unit 36 ​​compares the diagnostic results from the simultaneous relationship diagnosis unit 34 and the same position relationship diagnosis unit 35 with the anomaly determination table 25 (refer to Table 3), thereby classifying the causes of the anomaly. Furthermore, the anomaly determination unit 36 ​​outputs these determination results to the display unit 40.

[0084] The display unit 40 is implemented by a display device such as an LCD display or a CRT display. The display unit 40 displays, for example, the diagnostic results in the simultaneous relationship diagnostic unit 34, the diagnostic results in the same position relationship diagnostic unit 35, and the determination results in the anomaly determination unit 36 ​​based on the display signals input from the arithmetic unit 30, thereby providing guidance to the operator.

[0085] [Abnormal Diagnostic Methods]

[0086] Reference Figure 4 An anomaly diagnosis method using the anomaly diagnosis device 1 according to an embodiment of the present invention will be described. Furthermore, in the rolling process, the anomaly diagnosis method is performed each time the rolling of a single piece of material is completed.

[0087] First, the simultaneous time-relationship data creation unit 31 and the same position relationship data creation unit 32 determine whether the rolling of the rolled material has ended (step S21). In step S21, whether the rolling of the rolled material has ended can be determined, for example, based on the winding completion signal of the equipment for winding the rolled material. If it is determined that the rolling of the rolled material has not ended (no in step S21), the simultaneous time-relationship data creation unit 31 and the same position relationship data creation unit 32 return to step S1. On the other hand, if it is determined that the rolling of the rolled material has ended (yes in step S21), the simultaneous time-relationship data creation unit 31 and the same position relationship data creation unit 32 transfer from the event waiting state to the diagnostic process and collect timing data related to actual operation (step S22).

[0088] Next, the simultaneous moment relation data creation unit 31 creates simultaneous moment relation data by referring to the simultaneous moment relation data definition table 21 (step S23). The simultaneous moment relation data creation steps in step S23 are the same as those in step S23. Figure 2 The same. Next, the same positional relationship data creation unit 32 creates same positional relationship data with reference to the same positional relationship data definition table 22 (step S24). The same positional relationship data creation steps in step S24 are the same as... Figure 3 Same. In addition, either step S23 or S24 can be performed first, or both can be performed simultaneously.

[0089] Next, an anomaly diagnosis model is used for diagnosis (step S25). In step S25, anomaly diagnosis processing based on the simultaneous moment relationship data of the simultaneous moment relationship diagnosis unit 34 and anomaly diagnosis processing based on the same position relationship data of the same position relationship diagnosis unit 35 are performed.

[0090] Next, the anomaly determination unit 36 ​​determines whether an anomaly exists based on the two diagnostic results in step S25 (step S26). In step S26, if either of the two diagnostic results in step S25 contains an item marked "abnormal", the anomaly determination unit 36 ​​determines that an anomaly exists.

[0091] If an anomaly is determined ("Yes" in step S26), the anomaly determination unit 36 ​​classifies the cause of the anomaly by referring to the anomaly determination table 25 (refer to Table 3) (step S27). Next, the anomaly determination unit 36 ​​displays the anomaly classification result, i.e., the candidate causes of the anomaly, on the display unit 40, thereby providing guidance to the operator (step S28). Furthermore, the anomaly determination unit 36 ​​ends this process and transitions to the initial state of event waiting. Additionally, if no anomaly is determined in step S26 ("No" in step S26), the anomaly determination unit 36 ​​ends this process and transitions to the initial state of event waiting.

[0092] In the anomaly diagnosis model construction method, anomaly diagnosis method, anomaly diagnosis model construction apparatus, and anomaly diagnosis apparatus 1 described in the implementation method, the following processing is performed in the iron and steel rolling process: First, information related to the position of the rolling direction from the leading edge of the rolled material is assigned to the time-series data measured from the operating rolling mill by sensors or the like. Then, two sets of data are created: one representing the relationship between measurements at the same position, and the other representing the relationship between measurements at the same moment, thus constructing two anomaly diagnosis models.

[0093] In this way, by constructing two anomaly diagnosis models—for example, those capable of distinguishing between anomalies caused by the mechanical control system of equipment and anomalies caused by product quality and shape—diagnosis can be achieved. Therefore, the causes of anomalies can be identified in advance, equipment downtime can be reduced, and efficient and effective anomaly response strategies can be maintained.

[0094] The present invention has been specifically described above through methods and embodiments for implementing the invention, including the method for constructing an anomaly diagnosis model, the anomaly diagnosis method, the apparatus for constructing an anomaly diagnosis model, and the anomaly diagnosis apparatus 1. However, the scope of the present invention is not limited to these descriptions and must be broadly interpreted based on the descriptions within the scope of protection claimed by the present invention. Furthermore, various modifications and alterations made based on these descriptions are naturally included within the scope of the present invention.

[0095] For example, in the anomaly diagnosis device 1 according to the embodiment, two anomaly diagnosis models (simultaneous time relationship diagnosis model 23 and co-positional relationship diagnosis model 24) are used for anomaly diagnosis, but it is also possible to use only one of the anomaly diagnosis models for the first time. Then, anomaly diagnosis based on the other anomaly diagnosis model can be performed as needed.

[0096] Furthermore, in the abnormality diagnosis device 1 according to the embodiment, an example of its application to the rolling process has been described, but in addition to the rolling process, it can also be applied to the surface treatment process, etc.

[0097] Explanation of reference numerals in the attached figures

[0098] 1… Anomaly diagnosis device; 10… Input unit; 20… Storage unit; 21… Simultaneous moment relationship data definition table; 22… Same position relationship data definition table; 23… Simultaneous moment relationship diagnosis model; 24… Same position relationship diagnosis model; 30… Calculation unit; 31… Simultaneous moment relationship data creation unit; 32… Same position relationship data creation unit; 33… Model creation unit; 34… Simultaneous moment relationship diagnosis unit; 35… Same position relationship diagnosis unit; 36… Anomaly determination unit; 40… Display unit.

Claims

1. A method for constructing an anomaly diagnosis model, characterized in that: comprises: a first model creation step of creating a first abnormality diagnosis model that learns a relationship between a simultaneous-time measurement value and an abnormality, using measurement values measured at a simultaneous time with a predetermined prescribed measurement cycle for the plurality of devices; and a second model creation step of creating a second abnormality diagnosis model that learns a relationship between a same-position measurement value and an abnormality, using the measurement value of each same position of the metal material that is edited from the measurement values measured for the plurality of devices for each position of the metal material, the metal material being a rolled material, the device being a rolling mill, the same-position measurement value being calculated based on a ratio of a full length of the rolled material on a delivery side of a final rolling mill to a full length of the rolled material on a delivery side of a non-final rolling mill, and the position of the rolled material on the delivery side of the non-final rolling mill being converted by the ratio, whereby the same-position measurement value is calculated.

2. The abnormality diagnosis model construction method according to claim 1, wherein the full length of the rolled material on the delivery side of the rolling mill is calculated by calculating a strip speed of the rolled material from a roll speed and a forward slip of the rolling mill and time-integrating the strip speed. comprises:

3. An abnormality diagnosis method using an abnormality diagnosis model constructed by the construction method of an abnormality diagnosis model according to claim 1 or 2, characterized by, a first abnormality diagnosis step of performing abnormality diagnosis by inputting data indicating a relationship between simultaneous-time measurement values to the first abnormality diagnosis model; a second abnormality diagnosis step of performing abnormality diagnosis by inputting data indicating a relationship between same-position measurement values to the second abnormality diagnosis model; and an abnormality determination step of determining a cause of an abnormality based on a first diagnosis result in the first abnormality diagnosis step and a second diagnosis result in the second abnormality diagnosis step.

4. The abnormality diagnosis method according to claim 3, wherein the abnormality determination step determines a cause of an abnormality based on an abnormality diagnosis table that indicates a cause of an abnormality by associating the first diagnosis result with the second diagnosis result. comprises: a first model creation unit that creates a first abnormality diagnosis model that learns a relationship between a simultaneous-time measurement value and an abnormality, using measurement values measured at a simultaneous time with a predetermined prescribed measurement cycle for the plurality of devices; 5. An abnormality diagnosis model construction device for an abnormality diagnosis model for a process in which a metal material is sequentially processed by a plurality of devices, characterized by comprising: and a second model creation unit that creates a second abnormality diagnosis model that learns a relationship between a same-position measurement value and an abnormality, using the measurement value of each same position of the metal material that is edited from the measurement values measured for the plurality of devices for each position of the metal material, the metal material being a rolled material, the device being a rolling mill, the same-position measurement value being calculated based on a ratio of a full length of the rolled material on a delivery side of a final rolling mill to a full length of the rolled material on a delivery side of a non-final rolling mill, and the position of the rolled material on the delivery side of the non-final rolling mill being converted by the ratio, whereby the same-position measurement value is calculated. comprises: a first abnormality diagnosis unit that performs abnormality diagnosis by inputting data indicating a relationship between simultaneous-time measurement values to the first abnormality diagnosis model; 6.An abnormality diagnosis device that uses an abnormality diagnosis model constructed by the abnormality diagnosis model construction device according to claim 5, characterized by ​ ​ The second abnormality diagnosing unit performs abnormality diagnosis by inputting data indicating a relationship between measurement values at the same position to the second abnormality diagnosis model. And The abnormality determining unit determines a cause of abnormality based on the diagnosis result in the first abnormality diagnosing unit and the diagnosis result in the second abnormality diagnosing unit.

Images