Factor inference device, factor inference method, factor inference system, and terminal device

The factor inference device addresses the challenge of non-experts understanding complex machine learning results by inferring causal relationships using a knowledge model, enabling prompt and appropriate actions based on operational data.

JP7578199B2Active Publication Date: 2024-11-06JFE STEEL CORP
View PDF 13 Cites 0 Cited by

Patent Information

Application Number
JP2023536386
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Priority Date
2022-03-16
Filing Date
2023-03-02
Publication Date
2024-11-06
Estimated Expiration
2043-03-02

AI Technical Summary

Technical Problem

Non-experts and inexperienced users face challenges in understanding and taking appropriate actions based on machine learning results due to the lack of deep knowledge of the analysis target, especially when dealing with complex causal relationships and large numbers of variables, leading to difficulties in prompt problem-solving from operational data.

Method used

A factor inference device and method that infers the causes of phenomena in a process by utilizing a knowledge model expressed as a network of nodes, extracting relevant nodes, combining data, and presenting causal routes and relationships, even for users without deep knowledge, to facilitate prompt and appropriate actions.

Benefits of technology

Enables users to take timely and appropriate actions by providing useful auxiliary information through the factor inference device, even for those without deep knowledge, by presenting causal routes and relationships in a comprehensible manner.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007578199000004
    Figure 0007578199000004
  • Figure 0007578199000005
    Figure 0007578199000005
  • Figure 0007578199000006
    Figure 0007578199000006
Patent Text Reader

Abstract

Provided is a cause inference device comprising: a knowledge model acquisition means that acquires a knowledge model; an information creation means that creates information including at least a phenomenon abnormality indicator on the basis of data collected from a process; a node extraction means that uses the name of the data collected from the process to search node information of the knowledge model and extract a corresponding node; a data association means that associates the extracted node with corresponding information; and a cause inference means that infers and presents the cause of a phenomenon on the basis of the structure of the knowledge model and the information associated with the node.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to a factor inference device, a factor inference method, a factor inference system, and a terminal device. [Background technology]

[0002] Recently, big data analysis and the use of IoT have become popular, making it possible to obtain huge amounts of data. On the other hand, in order to analyze that data and solve problems, it is necessary to have deep knowledge of the subject of analysis, otherwise it is difficult to properly use the huge amount of data. For example, if sensor data from a production device shows unusual behavior, it is difficult to take measures to solve the problem unless there is knowledge about how such behavior affects the production device and the quality of the product.

[0003] Furthermore, advances in machine learning technology have made it possible to make highly accurate predictions and detect anomalies from huge amounts of data, and this technology is now being used in a variety of situations. For example, in the steel industry, deep learning technology is used to predict the occurrence of surface defects on products using manufacturing conditions as input data, and to detect equipment anomalies through a composite analysis of information from many monitoring items.

[0004] As the use of such data progresses, there are still challenges, as mentioned above, in taking action based on the results of machine learning. Specifically, in order to understand what mechanism and what kind of trouble may occur from the information on the monitoring items that the machine learning model detects as abnormal, deep knowledge of the subject of analysis is required. Without this knowledge, it is difficult for users to take action based on the results of machine learning. Since there are only a limited number of veterans with deep knowledge in the field, there is a demand for the output of auxiliary information that can be linked to appropriate actions. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] JP 2019-194849 A [Non-patent literature]

[0006] [Non-Patent Document 1] Marco, “Why should I trust you?:Explaining the predictions of any classifier”, Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, August 2016, pp.1135-1144 Summary of the Invention [Problem to be solved by the invention]

[0007] As a technique for deriving auxiliary information for linking to appropriate actions, for example, Non-Patent Document 1 proposes a method for calculating a numerical value of the relative influence of input variables that play an important role.

[0008] However, the technology proposed in Non-Patent Document 1 does not provide information on the mechanism of how each variable affects prediction or detection. Therefore, currently, these are inferred using the user's knowledge of the prediction or detection target. Therefore, non-experts or inexperienced users are unable to make accurate inferences and are unable to take appropriate actions.

[0009] In addition, when using machine learning with a large number of variables, even if you have knowledge of the subject of analysis, you may not be able to take quick action due to the time and effort required for inference.

[0010] For example, the technology proposed in Patent Document 1 attempts to present the influence of variables by automatically outputting the causal relationships between variables. However, it is difficult to say that non-experts or inexperienced users can adequately understand the mechanism just from the relationships between the variables used in the machine learning model.

[0011] The present invention has been made in view of the above, and has an object to provide a factor inference device, a factor inference method, a factor inference system, and a terminal device that can provide useful auxiliary information for problem solving from operational data such as sensor data and setting values, even when used by a user who does not have in-depth knowledge of the analysis target, and can lead to appropriate and prompt action. [Means for solving the problem]

[0012] In order to solve the above-mentioned problems and achieve the object, a factor inference device according to the present invention is a factor inference device that infers factors of a phenomenon in a process, and includes: knowledge model acquisition means for acquiring a knowledge model expressed in a network format connecting events occurring in the process as nodes for causal relationships of phenomena in the process; information creation means for creating information including at least an abnormality indicator related to the phenomenon based on data collected from the process; node extraction means for searching for information of the nodes in the knowledge model based on names of data collected from the process and extracting corresponding nodes; data linking means for linking the corresponding information to the extracted nodes; and factor inference means for inferring and presenting factors of the phenomenon based on a structure of the knowledge model and the information linked to the nodes.

[0013] In addition, in the factor inference device according to the present invention, in the above-mentioned invention, the node extraction means extracts the node by searching for character information that is the same as or similar to a name of the data, from character information of an event corresponding to a node of the knowledge model.

[0014] Furthermore, in the factor inference device of the present invention, in the above-mentioned invention, the factor inference means presents a causal route that connects the nodes in the knowledge model and indicates a causal relationship of the phenomenon, and when there are a plurality of causal routes, the causal routes are ranked and presented based on the information linked to each node of the causal route.

[0015] In addition, in the factor inference device of the present invention, in the above-mentioned invention, the factor inference means ranks and presents the causal route based on the number of nodes in the causal route to which anomaly indicators indicating an anomaly are linked, or the number of nodes that have been selected as a causal route in the past.

[0016] Further, in the factor inference device according to the present invention, in the above-mentioned invention, the factor inference means divides the knowledge model into blocks based on the information, and presents, for each block, a causal route that connects the nodes in the knowledge model and indicates a causal relationship of the phenomenon.

[0017] In addition, in the factor inference device according to the present invention, in the above-mentioned invention, the abnormality indicator includes at least one of an abnormality degree indicating the degree of abnormality and a category determined based on a deviation from a predetermined normal state.

[0018] In order to solve the above-mentioned problems and achieve the object, a factor inference method according to the present invention is a factor inference method that is executed by an apparatus constructed by a computer and infers factors of a phenomenon in a process, the factor inference method including: a knowledge model acquisition step in which a knowledge model acquisition means included in the computer acquires a knowledge model expressed in a network format that links events occurring in the process as nodes for causal relationships of phenomena in the process; an information creation step in which an information creation means included in the computer creates information including at least an abnormality indicator related to the phenomenon based on data collected from the process; a node extraction step in which a node extraction means included in the computer searches for information of the nodes in the knowledge model based on names of data collected from the process and extracts corresponding nodes; a data combination step in which a data combination means included in the computer combines the corresponding information with the extracted nodes; and a factor inference step in which a factor inference means included in the computer infers and presents factors of the phenomenon based on a structure of the knowledge model and the information combined with the nodes.

[0019] In order to solve the above-mentioned problems and achieve the object, a factor inference system according to the present invention is a factor inference system comprising a factor inference server device and a terminal device, wherein the factor inference server device comprises: knowledge model acquisition means for acquiring a knowledge model expressed in a network format connecting events occurring in a process as nodes for causal relationships of a phenomenon in the process, the nodes being nodes; information creation means for creating information including at least an abnormality indicator related to the phenomenon based on data collected from the process; node extraction means for searching for information of the nodes in the knowledge model based on names of data collected from the process and extracting corresponding nodes; data linking means for linking the extracted nodes with the corresponding information; factor inference means for inferring a factor of the phenomenon based on a structure of the knowledge model and the information linked to the nodes; and output means for outputting information including at least the factor of the phenomenon inferred by the factor inference means to the terminal device, and the terminal device comprises information acquisition means for acquiring information including at least the factor of the phenomenon from the factor inference server device, and display means for displaying the information acquired by the information acquisition means.

[0020] In order to solve the above-mentioned problems and achieve the object, a terminal device according to the present invention comprises an information acquiring means for acquiring information including at least a factor of a phenomenon in a process from a factor inference server device, and a display means for displaying the information acquired by the information acquiring means, and the factor of the phenomenon is inferred based on a knowledge model structure in which, regarding causal relationships of phenomena in a process, events occurring in the process are treated as nodes and expressed in a network format connecting the nodes, and information linked to nodes extracted in the knowledge model based on names of data collected from the process, the information including at least an abnormality indicator related to the phenomenon. Effect of the Invention

[0021] The factor inference device, factor inference method, factor inference system, and terminal device according to the present invention can provide useful auxiliary information for problem solving from operational data such as sensor data and set values, even when used by a user who does not have deep knowledge of the analysis target. This enables the user to take appropriate and prompt action to solve the problem by using this auxiliary information. [Brief description of the drawings]

[0022] [Figure 1] FIG. 1 is a diagram showing a schematic configuration of a factor inference device according to an embodiment of the present invention. [Diagram 2] FIG. 2 is a flowchart showing the procedure of a factor inference method executed by the factor inference device according to the embodiment of the present invention. [Diagram 3] FIG. 3 is a diagram showing an example of a knowledge model in the factor inference method according to the embodiment of the present invention. [Figure 4] FIG. 4 is a diagram showing an example in which some nodes of a knowledge model are blocked in the factor inference method according to the embodiment of the present invention. [Diagram 5] FIG. 5 is a diagram showing an example of an abnormality indicator (classification of abnormality / normality) acquired by a standalone monitoring system in the cause inference method according to the embodiment of the present invention. [Figure 6] FIG. 6 is a diagram showing an example of an anomaly index (degree of anomaly) acquired by the anomaly detection system in the factor inference method according to the embodiment of the present invention. [Figure 7] FIG. 7 is a diagram showing an example of monitoring items automatically assigned to a knowledge model in the factor inference method according to the embodiment of the present invention. [Figure 8] FIG. 8 is a diagram showing an example of a result of factor inference using abnormal / normal classification in the factor inference method according to the embodiment of the present invention. [Figure 9] FIG. 9 is a diagram showing an example of a result of factor inference using the degree of anomaly in the factor inference method according to the embodiment of the present invention. [Figure 10]FIG. 10 is a diagram showing a modified example in which the factor inference method according to the embodiment of the present invention is applied to prediction of the surface quality of a steel plate. [Figure 11] FIG. 11 is a diagram showing a schematic configuration of a factor inference system according to an embodiment of the present invention. [Figure 12] FIG. 12 is a diagram showing an example of a knowledge model in the example of the factor inference method according to the embodiment of the present invention. [Figure 13] FIG. 13 is a diagram showing an example of a result of factor inference in the example of the factor inference method according to the embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0023] A factor inference device, a factor inference method, a factor inference system, and a terminal device according to the embodiments of the present invention will be described with reference to the drawings. Note that the present invention is not limited to the following embodiments, and the components in the following embodiments include those that are replaceable and easy for a person skilled in the art, or those that are substantially the same.

[0024] [Factor inference device] A factor inference device according to an embodiment will be described with reference to Fig. 1. The factor inference device is for inferring factors of phenomena in a process. Examples of "processes" in this embodiment include manufacturing processes for steel products, power generation processes in power generation facilities, and transport processes in transport facilities.

[0025] In this embodiment, a description will be given assuming a case where the cause of an abnormality in a hot rolling process is inferred. The hot rolling process is a process in which a heated slab is rolled to a predetermined thickness. In the hot rolling process, processing is performed using a number of pieces of equipment such as a heating furnace, a rolling mill, and a quenching facility. If an abnormality occurs in the equipment, serious troubles such as breakage or holes in the plate during rolling may occur. Therefore, for stable production, it is important to infer the cause of the abnormality and present it to a user (an operator, etc.) to prevent such serious troubles from occurring.

[0026] 1 shows an example of the configuration of an information processing device 1 for realizing a factor inference device. The information processing device 1 is realized by a general-purpose computer such as a workstation or a personal computer. As shown in FIG. 1, the information processing device 1 includes an input unit 10, a storage unit 20, a calculation unit 30, and an output unit 40.

[0027] The input unit 10 is an input means for the calculation unit 30, and is realized by an input device such as a keyboard, a mouse pointer, a numeric keypad, etc. The input unit 10 inputs information necessary for various calculations in the calculation unit 30.

[0028] The storage unit 20 is realized by a recording medium such as an EPROM (Erasable Programmable ROM), a hard disk drive (HDD), and a removable medium. Examples of the removable medium include a disk recording medium such as a Universal Serial Bus (USB) memory, a Compact Disc (CD), a Digital Versatile Disc (DVD), and a Blu-ray (registered trademark) Disc (BD).

[0029] The storage unit 20 can store an operating system (OS), various programs, various tables, various databases, etc. The storage unit 20 stores a knowledge model 21. In addition to the knowledge model 21, the storage unit 20 may also store calculation results in the calculation unit 30, etc., as necessary.

[0030] The knowledge model 21 is a model that expresses the causal relationships of phenomena in a process in a network format that links the nodes, with events occurring in the process as nodes. This knowledge model 21 is, for example, created manually in advance and stored in the storage unit 20 in a computer-readable form. Only one knowledge model 21 may be created to handle multiple phenomena (problems), or multiple knowledge models 21 may be created to handle multiple phenomena. Details of the knowledge model 21 will be described later (see Figures 3 and 4).

[0031] The calculation unit 30 is realized by a processor such as a CPU (Central Processing Unit) and a memory (main storage unit) such as a RAM (Random Access Memory) or a ROM (Read Only Memory).

[0032] The calculation unit 30 loads a program into the working area of ​​the main storage unit, executes it, and controls each component unit through the execution of the program to realize a function that meets a predetermined purpose. Through the execution of the program, the calculation unit 30 functions as a knowledge model acquisition unit 31, an information creation unit 32, a node extraction unit 33, a data combination unit 34, and a factor inference unit 35. Note that, although FIG. 1 shows an example in which the functions of each unit of the calculation unit 30 are realized by one computer, the method of realizing the functions of each unit is not particularly limited, and the functions of each unit may be realized by multiple computers, for example.

[0033] The knowledge model acquisition unit 31 acquires the knowledge model 21 from the storage unit 20 .

[0034] The information creation unit 32 collects operation data from the process and creates information including at least an abnormality indicator related to the event based on the operation data. The information created by the information creation unit 32 is information that is linked to a node of the knowledge model 21 in a data linking unit 34 described later. This information may include, in addition to the abnormality indicators described above, for example, a data format, a type of equipment, and timing of the event.

[0035] The abnormality indicator includes, for example, an abnormality level (abnormality score (abnormal value)) indicating the degree of abnormality, a classification determined based on the degree of deviation from a predetermined normal state, etc. Note that the "classification determined based on the degree of deviation from a normal state" refers to, for example, a normal / abnormal determination result.

[0036] The data format includes, for example, an observed value, a set value, a category, etc. Note that the "observed value" refers to the sensor data of the equipment acquired by a sensor. Furthermore, the "set value" refers to the set value set in the equipment. Furthermore, the "category" refers to the type of data.

[0037] The timing of the event includes the time of the set calculation, during operation, and during non-operation. Note that "at the time of the set calculation" indicates that the event occurred during the set calculation. Also, "during operation" indicates that the event occurred during operation. Also, "non-operation" indicates that the event occurred during non-operation.

[0038] The node extraction unit 33 searches for node information in the knowledge model 21 based on the name of data collected from a process, and extracts the corresponding node. Specifically, the node extraction unit 33 extracts a node by searching character information of an event corresponding to the node in the knowledge model 21 for character information that is the same as or similar to the name of the data. Then, the node extraction unit 33 automatically assigns a monitoring item related to the content of the event of the node to the extracted node based on the similarity of the characters. Details of the processing by the node extraction unit 33 will be described later (see FIG. 7).

[0039] The data linking unit 34 links information (information created by the information creating unit 32) corresponding to the node of the knowledge model 21 extracted by the node extracting unit 33. Details of the processing by the data linking unit 34 will be described later (see Figs. 8 and 9).

[0040] The factor inference unit 35 infers factors that cause a phenomenon to occur based on the structure of the knowledge model 21 and information linked to the nodes of the knowledge model 21, and presents the inference to the user. Specifically, the factor inference unit 35 identifies a causal route that connects nodes in the knowledge model 21 and indicates the causal relationship of the phenomenon. The factor inference unit 35 then presents the causal route to the user via the output unit 40.

[0041] Furthermore, when there are multiple inferred causal routes, the factor inference unit 35 ranks and presents the causal routes based on information linked to each node of the causal routes. In this case, the factor inference unit 35 can rank and present the causal routes based on, for example, the number of nodes linked to an abnormality indicator indicating an abnormality in the inferred causal route or the number of nodes selected as a causal route in the past.

[0042] The factor inference unit 35 can also divide the knowledge model 21 into blocks based on information linked to each node, and present the causal route for each block in the knowledge model 21. Details of each process in the factor inference unit 35 will be described later (see Figs. 8 and 9).

[0043] The output unit 40 is an output means for outputting the calculation result by the calculation unit 30. The output unit 40 is realized by an input device such as a display, a printer, etc. The output unit 40 outputs, for example, the knowledge model 21 created in advance, the inference result of the cause of the phenomenon in the cause inference unit 35 (for example, a causal route), etc.

[0044] [Causal inference method] A factor inference method according to an embodiment will be described with reference to Fig. 2 to Fig. 9. In the factor inference method, as shown in Fig. 2, a knowledge model acquisition step (step S1), an operation data collection step (step S2), an information creation step (step S3), a node extraction step (step S4), a data combination step (step S5), a factor inference step (step S6), and a factor presentation step (step S7) are performed in this order. Each step will be described in detail below.

[0045] <Knowledge model acquisition step> In the knowledge model acquisition step, the knowledge model acquisition unit 31 acquires the knowledge model 21 pre-stored in the storage unit 20. Details of the knowledge model 21 will be described below.

[0046] (Knowledge Model) As described above, the knowledge model 21 is a model that expresses the causal relationships of phenomena in a process in the form of a network that connects the nodes, with events occurring in the process as nodes. In this embodiment, in order to predict on a computer how an event related to a monitoring item of a process will develop and what kind of trouble it will lead to, knowledge regarding the mechanism of the phenomenon that causes the trouble is prepared as a knowledge model 21 that can be processed by a computer.

[0047] As shown in Fig. 3, the knowledge model 21 expresses the mechanism of trouble occurrence using a network structure in the form of a causal chain that links events with causal relationships. The final result of this causal chain, that is, the end (top row) of each event, becomes the trouble of interest in the process. Specifically, in the knowledge model 21, events such as the setting conditions of the equipment related to the trouble, physical phenomena, and the state of the equipment are treated as nodes in the network structure, and events with causal relationships are connected by edges.

[0048] In the knowledge model 21 shown as an example in Figure 3, when an event occurs under an And condition, each event is connected by an edge via a circle. On the other hand, when an event occurs under an Or condition, each event is directly connected by an edge without a circle. For example, the knowledge model 21 shown in the figure indicates that "event 3 occurs when both events 5 and 6 occur," "event 1 occurs when either event 3 or event 4 occurs," and "trouble A occurs when either event 1 or event 2 occurs."

[0049] Specific troubles (trouble A in FIG. 3) described in the top row of the knowledge model 21 include, for example, holes, bending, and breakage in the hot rolling process. Examples of events represented by each node of the knowledge model 21 include, for example, "uneven heating of the slab in the heating furnace," "heat deviation occurs in the slab," "the crown becomes large," and so on.

[0050] 3, for convenience, "Trouble A" and "Events 1 to 7" are written, but in actual operation, specific contents corresponding to Trouble A and Events 1 to 7 are described. Thus, in this embodiment, the knowledge model 21 expressed in the network structure of causal relationships is created in a form that can be read by a computer.

[0051] (Blocking knowledge models) As described later, when attribute information is registered for the nodes of the knowledge model 21, it is also possible to block nodes corresponding to some events based on the attribute information, as shown in Fig. 4. For example, the block X shown in the figure contains another knowledge model 21 that has common attribute information. For example, events 3, 5, and 6 shown in the figure all have the attribute [timing] of "at the time of setting calculation," so they are expressed together as one block.

[0052] It is also possible to create blocks within blocks, enabling hierarchical representation. When the knowledge model 21 is presented to the user through the output unit 40, it is also possible to display details of the nodes and blocks contained within by clicking on a block. In this way, by dividing a part of the knowledge model 21 into blocks, the causal relationships between events can be presented in the same attribute units (for example, the same equipment units), making it possible to present the causes of troubles in a process in an easier-to-understand manner.

[0053] <Operational data collection step, information creation step> In the operation data collection step, the information creation unit 32 collects operation data from the process. Then, in the information creation step, the information creation unit 32 creates information including at least an abnormality indicator related to the event based on the collected operation data. In the operation data collection step and the information creation step, the operation data can be collected and information including an abnormality indicator can be created, for example, through an existing standalone monitoring system or an abnormality detection system.

[0054] (Standalone monitoring system) In a standalone monitoring system, a large amount of operational data is collected by various sensors installed in the equipment, and after determining the monitoring items, monitoring is performed based on whether or not there is a deviation from a predetermined threshold. Monitoring methods in a standalone monitoring system include, for example, a method in which an abnormality is determined when there is a deviation of 3σ or more from the average of the distribution under normal conditions, and a method in which an abnormality is determined when the value of a monitored item deviates from the upper or lower control limit range as shown in Figure 5.

[0055] Here, since the operational data is diverse, it is difficult to monitor all of it, and even if there is data showing abnormal behavior, in order to judge the impact on production from the data, deep knowledge of each facility and process is required. Therefore, it is difficult for anyone other than experienced experts to judge the impact on production from the data. Therefore, in this embodiment, the cause of the trouble is inferred using the abnormality index acquired by the standalone monitoring system and the knowledge model 21. Note that the abnormality index acquired by the standalone monitoring system may be a classification (determination result) of abnormal / normal as shown in FIG. 5, or may be an abnormality degree (anomaly score) indicating the degree of abnormality.

[0056] (Anomaly detection system) Anomaly detection systems use methods that use machine learning to determine whether equipment is exhibiting anomalies. Whereas individual monitoring systems look at individual monitored items to determine whether there is an anomaly, anomaly detection system looks at multiple monitored items in a combined manner to determine whether there is an anomaly.

[0057] Machine learning methods that can be used in an anomaly detection system include, for example, statistical-based methods such as the graphical Gaussian model and principal component analysis, and deep learning-based methods such as autoencoders. These methods learn data on monitored items under normal conditions and output whether the current data is abnormal or not. In an anomaly detection system, as shown in Figure 6, for example, the degree of anomaly (anomaly score) for each monitored item is output, and whether the monitored item is abnormal or not is determined based on the magnitude of the degree of anomaly. Examples of monitored items in the hot rolling process include "amount of meandering," "load difference," and "temperature difference of the slab."

[0058] Here, the number of monitoring items in the anomaly detection system is huge, amounting to several thousand items. Therefore, it is difficult to instantly determine what kind of trouble is connected to a monitoring item determined to be abnormal. In addition, it is difficult for an inexperienced user who does not have sufficient knowledge about troubles to guess what mechanism is involved in a phenomenon related to each monitoring item that may develop into a trouble in order to take action to deal with the trouble. Therefore, in this embodiment, the cause of the trouble is inferred by using the anomaly indicator acquired by the anomaly detection system and the knowledge model 21.

[0059] <Node extraction step> In the node extraction step, the node extraction unit 33 searches the character information of the nodes in the knowledge model 21 based on the names of the data collected from the processes, and extracts the corresponding nodes. Then, for the extracted nodes, the monitoring items related to the contents of the events of the nodes are automatically assigned based on the similarity of the characters.

[0060] Methods for searching for character information of nodes include searching for character strings based on a rule base, preparing a dictionary of synonyms that includes spelling variations, etc., and searching for similar words taking into account the effects of spelling variations, etc., and searching for similar words using a machine learning model.

[0061] For example, as shown in Figure 7, if the content of Event 2 is "uneven temperature of slab in heating furnace," the monitoring item "temperature difference of slab" is assigned. Note that some events cannot be monitored (observed) by sensors, etc. Therefore, it is not necessary to assign monitoring items to all nodes, and there may be nodes without monitoring items.

[0062] In addition, among the nodes of the knowledge model 21, attribute information may be assigned as necessary to the nodes in which the monitoring items are registered. This attribute information includes, for example, an abnormality indicator related to the event, a data format, a type of equipment, and a timing of the event.

[0063] The abnormality indicators include, for example, an abnormality degree (anomaly score) indicating the degree of abnormality, a category determined based on the degree of deviation from a predetermined normal state, etc. The data format includes, for example, an observed value, a set value, a category, etc. The timing of the event includes, for example, at the time of setting calculation, during operation, during non-operation, etc. As with the monitoring items, it is not necessary to assign attribute information to all nodes, and there may be nodes without attribute information.

[0064] <Data binding step> In the data combination step, the data combination unit 34 combines information (information created by the information creation unit 32) corresponding to a node (a node extracted by the node extraction unit 33) of the knowledge model 21. In the data combination step, the information to be combined with the knowledge model 21 includes at least an anomaly indicator obtained by the standalone monitoring system or the anomaly detection system.

[0065] Examples of information to be linked to the knowledge model 21 include an abnormality index, a data format, a type of equipment, and timing of an event. Examples of abnormality indexes include abnormal / normal classifications as shown in Table 1, and abnormality levels calculated for each monitoring item as shown in Table 2. Although Table 1 shows two classification patterns of "abnormal / normal" as an example, three classification patterns of "normal / minor abnormal / major abnormal" may also be used.

[0066] [Table 1] [Table 2]

[0067] <Factor inference step> In the causal inference step, the causal inference unit 35 infers the cause of the phenomenon based on the structure of the knowledge model 21 and the information linked to the nodes of the knowledge model 21. In the causal inference step, the knowledge model 21 expressed in the graph structure of the causal chain and the abnormality indexes shown in Tables 1 and 2 are used to infer what kind of trouble may occur in the process and how the trouble will occur. Below, inference using the abnormality indexes (classification of abnormal / normal) in Table 1 and inference using the abnormality indexes (degree of abnormality) in Table 2 will be explained separately.

[0068] (Inference using abnormal / normal classification) In the data merging step, the anomaly indicators in Table 1 are reflected (merged) in the knowledge model 21, and then the causal relationship is traced from the event of the monitoring item that was determined to be abnormal to the result direction, as shown in Figure 8. As a result, the trouble described in the final node reached (Trouble A in the figure) is identified as a trouble that may occur in the process. In addition, the causal route of the previously traced events is identified as the occurrence mechanism of the trouble, as shown by the thick line in the figure.

[0069] Here, in the cause inference step, when tracing the causal relationship in the result direction, if there is an And connection, tracing further in the result direction is done only if all the relevant events are abnormal. For example, in Figure 8, since both events 5 and 6 are abnormal, tracing is done to event 3. Also, if a monitoring item reaches a normal event while tracing in the result direction, the process of tracing further in the result direction is stopped.

[0070] (Inference using anomalies) In the data merging step, the anomaly indicators in Table 2 are reflected (merged) in the knowledge model 21, and then, as shown in Fig. 9, the causal relationship is traced from the event of the monitored item to the result direction according to the degree of anomaly. As a result, the trouble described in the final node reached (trouble A in the figure) is identified as a trouble that may occur in the process. In addition, the causal route of the previously traced events is identified as the occurrence mechanism of the trouble, as shown by the bold line in the figure.

[0071] Here, in the causal inference step, the causal route with the highest sum of anomalies is identified as the occurrence mechanism of the event described in the final node. For example, in Figure 9, the sum of anomalies of the causal route from Event 5 → Event 3 → Event 1 → Trouble A is the highest at "24.0", so the causal route from Event 5 → Event 3 → Event 1 → Trouble A is identified as shown by the thick line in the figure.

[0072] <Factor presentation step> In the factor presentation step, the factor inference unit 35 presents the factors of the phenomenon inferred in the factor inference step to the user through the output unit 40. Specifically, in the factor presentation step, the causal route shown by the thick line in Fig. 8 and Fig. 9 is presented. In the factor presentation step, the causal route may be highlighted, for example, as follows, in addition to the method of highlighting the causal route by the thick line as in the figures. For example, in the factor presentation step, the nodes and edges included in the causal route may be displayed in a color different from the other nodes and edges, or only the nodes and edges included in the causal route may be displayed, and the other nodes and edges may be hidden.

[0073] In addition, when a plurality of troubles or a plurality of causal routes are identified in the causal inference step, the plurality of troubles and the plurality of troubles may be presented to the user via the output unit 40. In this case, for example, the causal routes identified in the causal inference step may be presented ranked based on the number of nodes to which an abnormality indicator indicating an abnormality is linked or the number of nodes previously selected as causal routes.

[0074] [Modifications] In the above embodiment, it is assumed that the cause of an abnormality in the hot rolling process is inferred, but it is also possible to apply the present invention to predicting the surface quality of a steel sheet, for example. Problems may occur during casting or rolling, causing scratches on the surface of the steel sheet after or during production. Steel sheets with scratches cannot be shipped, which leads to a yield loss.

[0075] For this reason, a predictive model (machine learning model) has been developed that predicts whether or not defects will occur based on the manufacturing conditions of steel sheets. This technology makes it possible to output a numerical index (hereinafter referred to as "influence level") indicating which of the data items of the manufacturing conditions used (hereinafter referred to as "explanatory variables") influences the results predicted by the predictive model. When defects are predicted to occur on steel sheets, the influence level is used as a reference to find the cause, but without deep knowledge of the defect occurrence mechanism, it is difficult to determine how the phenomena of each explanatory variable are linked to cause the defect to occur.

[0076] Therefore, in this modification, a knowledge model 21 expressed in a network structure of causal relationships of events that cause defects is created in a computer-readable form as shown in Fig. 10. In addition, explanatory variables of a prediction model related to the events of the node (see explanatory variables 1 to 4 in the figure) are assigned to the nodes of this knowledge model 21.

[0077] In the data combining step of this modified example, information on whether each explanatory variable is important or not is linked to the nodes of the knowledge model 21. The information on whether each explanatory variable is important or not can be created based on, for example, whether the influence of the explanatory variable calculated by the prediction model exceeds a predetermined threshold value or not.

[0078] In this modification, in the factor inference step, a causal route is identified as shown by the thick line in Fig. 10. Then, in the factor presentation step, the identified causal route is presented.

[0079] [Factor inference system] 11, a factor inference system 2 according to the embodiment includes a factor inference server device 3 and a terminal device 4. The factor inference server device 3 and the terminal device 4 are configured to be able to communicate with each other through a network N such as the Internet. In the factor inference system 2, the terminal device 4 is disposed in, for example, a steelworks.

[0080] The factor inference server device 3 is realized by, for example, a server arranged on a cloud. This factor inference server device 3 is a network-compatible server device of the factor inference device realized by the information processing device 1 shown in Fig. 1, and has the same configuration as that of the factor inference device. That is, the factor inference server device 3 has at least an output unit and a calculation unit that functions as a knowledge model acquisition unit, an information creation unit, a node extraction unit, a data combination unit, and a factor inference unit among the components of the information processing device 1.

[0081] The knowledge model acquisition unit of the factor inference server device 3 acquires a knowledge model expressed in a network format that links events occurring in a process as nodes, regarding the causal relationship of a phenomenon in the process. The information creation unit of the factor inference server device 3 creates information including at least an abnormality indicator related to the phenomenon based on data collected from the process. The node extraction unit of the factor inference server device 3 searches for information on the nodes of the knowledge model based on the name of the data collected from the process, and extracts the corresponding node. The data link unit of the factor inference server device 3 links the corresponding information to the extracted node. The factor inference unit of the factor inference server device 3 infers the cause of the phenomenon based on the structure of the knowledge model and the information linked to the node. The output unit of the factor inference server device 3 outputs information including at least the cause of the phenomenon inferred by the factor inference unit to the terminal device 4 via the network N.

[0082] The terminal device 4 acquires various information from the factor inference server device 3 through the network N and displays it. This terminal device 4 is realized by, for example, a general-purpose computer such as a personal computer, or a mobile information processing device such as a tablet computer. The terminal device 4 also includes a calculation unit that functions as an information acquisition unit and a display unit. The terminal device 4 is also connected to an external display such as, for example, a liquid crystal display (LCD), an organic light-emitting display (OLED), or a touch panel display.

[0083] The information acquisition unit of the terminal device 4 acquires, from the factor inference server device 3, information including at least the factor of the phenomenon inferred by the factor inference server device 3. The factor of the phenomenon is inferred based on the structure of a knowledge model expressed in a network format connecting events occurring in a process as nodes and connecting the nodes, regarding the causal relationship of the phenomenon in the process, and on information linked to the nodes of the knowledge model. The information linked to the nodes of the knowledge model is information linked to nodes extracted in the knowledge model based on the names of data collected from the process, and includes at least an abnormality indicator related to the phenomenon.

[0084] In addition, the causes of the above-mentioned phenomenon may include, for example, the entire causal route that associates the phenomenon in the process with its causes (see Figs. 8 and 9), a highlighted display of the causal route of the cause of the inferred phenomenon, the ranking results of multiple causal routes, etc. The display unit of the terminal device 4 then displays this information acquired by the information acquisition unit on the above-mentioned external display.

[0085] [Example] An embodiment in which the factor inference method according to the present invention is applied to a rolling process of a steel plate will be described with reference to Figures 12 and 13. Figure 12 shows a knowledge model used in this embodiment.

[0086] The knowledge model shown in Figure 12 shows how multiple events affect a certain trouble C. Each node in the knowledge model is given an explanation of the events that show the causal influence. Note that some nodes (phenomena) cannot be observed directly or no data exists, so the causes are inferred using nodes of observable phenomena. In the figure, the shaded nodes indicate nodes of unobservable phenomena, and the other nodes indicate nodes of observable phenomena.

[0087] Table 3 shows the anomaly levels calculated by the anomaly diagnosis system for the monitoring items of the observable nodes in FIG.

[0088] [Table 3]

[0089] In this embodiment, a character search was performed on the monitoring items shown in Table 3, and nodes that were determined to match were extracted (node ​​extraction step). In this node extraction step, if the same word is simply used, the monitoring item is assigned as is. On the other hand, if the word is not the same but is similar, the similar words are linked in advance (for example, "pressure" and "hydraulic pressure", "operator" and "op"), and the monitoring item is assigned to the appropriate node.

[0090] In this embodiment, as shown in Fig. 13, a monitoring item "equipment A pressure" is assigned to a node of an event "hydraulic pressure of equipment A drops," and a monitoring item "operator operation amount" is assigned to a node of an event "operator operation amount exceeds appropriate amount." In addition, a monitoring item "steel plate thickness" is assigned to a node of an event "steel plate thickness differs from setting," and a monitoring item "steel plate temperature" is assigned to a node of an event "steel plate temperature differs from setting." In addition, a monitoring item "equipment B speed" is assigned to a node of an event "speed of equipment B is fast."

[0091] Next, the extracted nodes were assigned a corresponding degree of anomaly (data combination step), and the cause was inferred (causal inference step) using the knowledge model (causal route) as shown in Fig. 13. Note that, depending on the node (phenomenon), there may be cases where direct observation is not possible or where no data exists, so the cause is inferred using observable nodes.

[0092] In Fig. 13, there are four causal routes leading to trouble C. As shown by the bold line in the figure, by tracing the node with the highest total anomaly degree (hereinafter referred to as "total route anomaly degree") among the nodes leading to trouble C, it is found that the root cause is "speed anomaly (slow speed) in equipment status B." In this way, the causal inference method according to the present invention makes it possible to easily grasp the root cause of a trouble by using a knowledge model.

[0093] According to the above-described factor inference device, factor inference method, factor inference system, and terminal device of the embodiments, even when used by a user (operator, etc.) who does not have deep knowledge of the analysis target, it is possible to provide auxiliary information (causal route) useful for problem solving from operational data such as sensor data and set values. This enables the user to take appropriate and prompt action to solve the problem by using this auxiliary information.

[0094] Although the factor inference device, factor inference method, factor inference system, and terminal device according to the present invention have been specifically described above using the mode and examples for carrying out the invention, the gist of the present invention is not limited to these descriptions and must be broadly interpreted based on the claims. Furthermore, it goes without saying that various changes, modifications, etc. based on these descriptions are also included in the gist of the present invention. [Explanation of symbols]

[0095] 1. Information processing device 10 Input section (input means) 20 Storage unit (storage means) 21 Knowledge Model 30 Arithmetic unit (arithmetic means) 31 Knowledge model acquisition unit (knowledge model acquisition means) 32 Information Creation Department (Information Creation Means) 33 Node extraction unit (node ​​extraction means) 34 Data linking unit (data linking means) 35 Factor inference section (factor inference means) 40 Output section (output means) Two-factor inference system 3. Factor inference server device 4 Terminal Equipment N Network

Claims

1. A factor inference device for inferring factors of a phenomenon in a process, comprising: a knowledge model acquisition means for acquiring a knowledge model expressed in a network format connecting events occurring in the process with nodes, the nodes being related to causal relationships of phenomena in the process; an information generating means for generating information including at least an abnormality indicator related to the event based on data collected from the process; a node extraction means for searching for node information of the knowledge model based on the name of data collected from the process, and extracting a corresponding node; a data linking means for linking the extracted nodes with the corresponding information; a factor inference means for inferring and presenting a factor of the phenomenon based on a structure of the knowledge model and the information linked to the node; Equipped with the abnormality index is a classification of abnormality or normality determined based on a deviation from a predetermined normal state, or an abnormality degree indicating a degree of abnormality, The factor inference means includes: In the knowledge model, a node corresponding to an event determined to be abnormal by the anomaly indicator is traced in a direction of a result of a causal relationship, a phenomenon described in the final node reached is identified as a phenomenon that may occur in the process, and a causal route to the identified phenomenon is inferred as a cause of the phenomenon; When identifying the causal route, a causal route consisting of nodes in which all of the events indicate an abnormality is identified, or a causal route is identified based on a sum of the abnormality degrees of the events from among a plurality of causal routes from causes to effects. Factor inference device.

2. the node extraction means extracts the node by searching for character information that is the same as or similar to the name of the data from character information of an event corresponding to the node of the knowledge model; The factorial reasoning device according to claim 1 .

3. The factor inference means includes: presenting a causal route in the knowledge model that connects the nodes and indicates a causal relationship of the phenomenon; When there are a plurality of causal routes, the causal routes are ranked and presented based on the information associated with each node of the causal routes. The factor inference device according to claim 1 or 2.

4. the factor inference means ranks and presents the causal route based on the number of nodes to which an anomaly indicator indicating an anomaly is linked in the causal route or the number of nodes that have been previously selected as a causal route; The factorial inference device according to claim 3.

5. the factor inference means divides the knowledge model into blocks based on the information, and presents the causal route that connects the nodes in the knowledge model and indicates a causal relationship of the phenomenon for each block; The factorial reasoning device according to claim 1 .

6. A causal inference method for inferring a cause of a phenomenon in a process, the method comprising: a knowledge model acquisition step in which a knowledge model acquisition means included in the computer acquires a knowledge model expressed in a network format connecting events occurring in the process as nodes, regarding causal relationships of phenomena in the process; an information creation step in which information creation means included in the computer creates information including at least an abnormality indicator related to the event based on data collected from the process; a node extraction step in which a node extraction means included in the computer searches for node information of the knowledge model based on the name of the data collected from the process, and extracts a corresponding node; a data linking step in which a data linking means included in the computer links the extracted nodes with the corresponding information; a factor inference step in which a factor inference means included in the computer infers and presents factors of the phenomenon based on a structure of the knowledge model and the information linked to the nodes; Including, the abnormality index is a classification of abnormality or normality determined based on a deviation from a predetermined normal state, or an abnormality degree indicating a degree of abnormality, The factor inference step includes: In the knowledge model, a node corresponding to an event determined to be abnormal by the anomaly indicator is traced in a direction of a result of a causal relationship, a phenomenon described in the final node reached is identified as a phenomenon that may occur in the process, and a causal route to the identified phenomenon is inferred as a cause of the phenomenon; When identifying the causal route, a causal route consisting of nodes in which all of the events indicate an abnormality is identified, or a causal route is identified based on a sum of the abnormality degrees of the events from among a plurality of causal routes from causes to effects. Factorial inference methods.

7. A factor inference system including a factor inference server device and a terminal device, The factor inference server device a knowledge model acquisition means for acquiring a knowledge model expressed in a network format connecting events occurring in a process with respect to causal relationships of phenomena in the process, the nodes being events occurring in the process; an information generating means for generating information including at least an abnormality indicator related to the event based on data collected from the process; a node extraction means for searching for node information of the knowledge model based on the name of data collected from the process, and extracting a corresponding node; a data linking means for linking the extracted nodes with the corresponding information; a factor inference means for inferring a factor of the phenomenon based on a structure of the knowledge model and the information linked to the node; an output means for outputting information including at least the cause of the phenomenon inferred by the cause inference means to the terminal device; Equipped with the abnormality index is a classification of abnormality or normality determined based on a deviation from a predetermined normal state, or an abnormality degree indicating a degree of abnormality, The factor inference means includes: In the knowledge model, a node corresponding to an event determined to be abnormal by the anomaly indicator is traced in a direction of a result of a causal relationship, a phenomenon described in the final node reached is identified as a phenomenon that may occur in the process, and a causal route to the identified phenomenon is inferred as a cause of the phenomenon; When identifying the causal route, a causal route consisting of nodes in which all of the events indicate an abnormality is identified, or a causal route is identified based on a sum of the abnormality degrees of the events from among a plurality of causal routes from causes to effects; The terminal device an information acquisition means for acquiring information including at least a cause of a phenomenon from the cause inference server device; A display means for displaying the information acquired by the information acquisition means; A factor inference system comprising:

8. an information acquisition means for acquiring information including at least a cause of a phenomenon in a process from a cause inference server device; A display means for displaying the information acquired by the information acquisition means; Equipped with The display means, based on the information acquired by the information acquisition means, Displaying the entire causal route that links the phenomenon in the process and its factors; Highlighting possible phenomena that may occur in the process and the causal pathways leading to the phenomena; When multiple causal routes exist, the results are ranked based on the anomaly indicators and displayed together. the abnormality index is a classification of abnormality or normality determined based on a deviation from a predetermined normal state, or an abnormality degree indicating a degree of abnormality, The causal route is a causal route consisting of nodes whose events all indicate an abnormality, or a causal route identified based on a sum of the abnormality degrees of the events from among a plurality of causal routes from causes to effects. Terminal device.

Citation Information

Patent Citations

  • Process system of plant faulty signal

    JP1982123409A

  • Method of analyzing abnormality diagnosis

    JP1987014210A

  • Operational information providing system for plant

    JP1994066607A

  • Supervisory system for plant

    JP1994137908A

  • Method and device for monitoring plant

    JP1995219624A