A method for tracing the source of longitudinal crack defects on the surface of steel products based on knowledge graph

Abstract

A method for tracing and analyzing the defects of longitudinal cracks on the surface of steel products based on knowledge graphs belongs to the technical field of quality traceability of steel products. A top-down knowledge graph construction route is adopted, and industry experts in the field of longitudinal crack defects on the surface of steel products are used to sort out knowledge, design the knowledge graph ontology of the traceability of longitudinal crack defects on the surface of steel products, and combine the actual data of the enterprise to create entities and relationships of the knowledge graph, and construct a cross-process steel product surface longitudinal crack defect traceability knowledge graph. The advantage is that the knowledge graph visualization technology is used to represent the quality defects of the entire process and the transmission relationship between quality events, breaking the process restrictions and realizing the traceability analysis of longitudinal crack defects on the surface of steel products across processes.

Description

Technical Field

[0001] The present invention belongs to the technical field of steel product quality traceability, and in particular relates to a steel product surface longitudinal crack defect traceability analysis method based on a knowledge graph. Technical Background

[0002] In the original traceability of the surface longitudinal crack quality defects of steel products, the entire production process of the steel products with surface longitudinal crack quality defects must first be obtained through the association of specific fields in the relational database or the product ID, and then the cause of the occurrence must be traced through the corresponding data analysis. Technical personnel need to clearly understand the complex correlation between the various data tables, the meaning and associated influence of each production parameter recorded in the data table. However, the causes of the surface longitudinal crack defects are complex, and relying solely on data for analysis will to a certain extent ignore the impact of other quality events without parameter records, as well as the transmission effect caused by the heredity and correlation of quality events between processes. Summary of the invention

[0003] The purpose of the present invention is to provide a method for tracing and analyzing the longitudinal crack defects on the surface of steel products based on a knowledge graph. It is more reasonable and flexible to use a graph to describe the relationship between the quality events that occur in the steel production process and the longitudinal crack defects on the surface of the final steel products rather than simply using a relational database table to describe it. The graph is used to replace the field association of the table structure in the original relational database, and a knowledge graph of the causes of quality defects of steel products is constructed from top to bottom with the help of expert knowledge. The ontology is used to standardize the expert knowledge and precipitate the expert knowledge. The knowledge graph visualization technology is used to represent the quality defects of the entire process and the transmission relationship between quality events, breaking the limitations of the process and realizing the cross-process tracing analysis of longitudinal crack defects on the surface of steel products.

[0004] The present invention adopts a top-down knowledge graph construction route, uses industry experts in the field of steel product surface longitudinal crack defects to sort out knowledge, designs the steel product surface longitudinal crack defect traceability knowledge graph ontology, and combines the actual data of the enterprise to create entities and relationships of the knowledge graph, and constructs a cross-process steel product surface longitudinal crack defect traceability knowledge graph. Specifically, it includes the following steps:

[0005] (1) Constructing the knowledge graph ontology of the traceability of longitudinal crack defects on the surface of steel products based on expert experience knowledge;

[0006] (2) Based on the ontology and actual production conditions, entity instances and relationship instances are constructed to form a knowledge graph for tracing the defects of longitudinal cracks on the surface of steel products: The instances corresponding to the models in the knowledge graph ontology are specific events that occur during the processing and production of steel products. For example, the entities corresponding to the quality event model are the process steps of steel products in the production process, such as "converter slag" and "converter re-blowing" in the converter process, and "LF argon gas static blowing time does not match" in the LF refining process; the entities corresponding to the abnormal event model are the abnormal situations that may occur in each process step (quality event entity), such as "converter steel temperature drop is too large" and "converter re-blowing twice" that may occur in the converter process. Construct entities, each entity will correspond to a certain ontology model, and then establish specific relationships between entities for all entities based on the relationships between the ontology models.

[0007] (3) Based on the constructed knowledge graph entity network, the possible cause paths of the surface longitudinal crack quality defects of steel products are searched: by inputting the quality defects of steel products, such as "surface longitudinal crack", all entity nodes and relationships pointing to the "surface longitudinal crack" node will be obtained, forming a graph network centered on the "surface longitudinal crack" node. Then, according to the actual situation, the screening conditions are input, and the entity types and relationship types appearing in the network are attribute filtered to obtain the cause paths that may cause the surface longitudinal crack defects. For example, by inputting the billet number of a steel product with a surface longitudinal crack defect, the processing steps for processing the billet number steel will be obtained, thereby filtering out all the cause paths that may cause the surface longitudinal cracks in the processing steps of the billet number steel product.

[0008] (4) Based on the possible paths searched out, combined with actual production data and events, analysis is performed to infer the real cause of the quality defects: According to the billet number of the steel product with quality defects, on the one hand, all processing steps of the billet number product and the abnormal events that may occur in the process are obtained, and then the production process parameter model and judgment rules that will cause the abnormal event are obtained; on the other hand, the actual production parameters of the steel product are obtained from the enterprise production process database, and matched with the judgment rules, so as to obtain the path that does not match the production rules and causes the quality defects. For example, the billet number of the steel product with surface longitudinal crack defects is 200900011. Surface longitudinal cracks are selected in quality defects, and the billet number is input as 20092300011. The system retrieves that during the production process of this product, in the converter process with furnace number 20092301, the temperature drop during the converter steelmaking process is too large, T 终点温度 -T 大包温度 =73℃, which is greater than the critical value of the judgment rule endpoint of 65℃, resulting in the appearance of surface longitudinal crack defects. Therefore, after analysis, the output will be (: judgment rule {rule description: "When the converter is tapped, T 终点温度 -T大包温度 ≥65℃", rule id: 10})->(:parameter{parameter name:["T 终点温度 ”,”T 大包温度 ”], metadata: {Converter monitoring sensor database}})->(: Abnormal event {Abnormal event name: "Converter steel temperature drop is too large", Occurrence location: "All ingots produced by this furnace", Equipment: "Converter equipment name", Process: "Converter"})->(: Quality event {Quality event name: "Converter steel temperature drop", Process: "Converter"})->(: Quality defect {Defect name: "Surface longitudinal crack", Process: "Continuous casting"}) This causal path;

[0009] Through attribute filtering, a cross-process critical path of causes is generated for the longitudinal crack defects on the surface of steel products, and a specific quality event chain is output; at the same time, a critical path flow diagram for tracing the quality defects of steel products is established based on empirical probability and data-driven construction, and the attribution probability of each possible traceability path is quantitatively analyzed to provide support for quality decisions.

[0010] The knowledge graph for tracing the source of longitudinal crack defects on the surface of steel products contains five knowledge graph ontology models and their relationships. The specific ontology models are as follows:

[0011] Quality defect model: refers to the quality defect events that may occur in steel products, such as the "surface longitudinal crack" defect. The quality defect ontology model contains two attributes: defect name and process to which it belongs. The outgoing relationship "produced" points to the failed product model, and the relationship is accompanied by a weight. A certain quality defect will cause the steel product to fail during the use of the product, and the weight is the degree of influence of the product defect on the product failure.

[0012] Quality event model: refers to the process events that the steel product processing goes through. The quality event ontology model contains two attributes: the quality event name and the process to which it belongs. The outgoing relationship "leads to" points to the quality defect model, and the relationship is accompanied by a weight. A quality event in the process of steel product processing will lead to the occurrence of a certain quality defect. The weight refers to the degree of influence on the quality defect; the outgoing relationship "conduction" points to itself, and the relationship is accompanied by a weight. Each process consists of multiple quality events, and the quality events between different processes also have mutual influence. The weight refers to the degree of influence between different quality events.

[0013] Abnormal event model: refers to the abnormal situation that may occur in a certain quality event during the processing of steel products. The abnormal event model contains four attributes: abnormal event name, occurrence location, belonging equipment, and belonging process. The outgoing relationship "belongs to" points to the quality event model.

[0014] Parameter model: refers to the influencing parameters for determining whether abnormal events will occur during the processing of steel products. The judgment rule model contains two attributes: parameter name and metadata, and the output relationship points to the abnormal event. The metadata attribute records the location of the parameter in the database table and the access method or interface. There are multiple parameters that jointly determine the occurrence of a certain abnormal event.

[0015] Decision rule model: refers to the decision rule corresponding to the parameter. The decision rule model contains two attributes: rule description and rule ID. The outgoing relationship points to the parameter model. The rule ID is the serial number of a rule in the rule base. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is the knowledge graph ontology model diagram for traceability analysis of quality defects on the surface of steel products such as longitudinal cracks.

[0017] Figure 2 It is a knowledge graph entity node and relationship diagram built based on the knowledge graph ontology model. DETAILED DESCRIPTION

[0018] The technical solution of the present invention will be described in more detail below in conjunction with the accompanying drawings:

[0019] The method of the present invention is based on the knowledge graph. The knowledge graph is essentially a semantic network, whose nodes represent entities or concepts, and edges represent various semantic relationships between entities / concepts. Each model in the knowledge graph ontology is a concept in an abstract sense, and the entity node corresponding to the ontology model is the concrete representation of the ontology concept in reality.

[0020] The present invention adopts a top-down approach to construct the knowledge graph for tracing and analyzing the surface longitudinal crack defects of steel products. Therefore, it is necessary to first design and define the ontology model of the knowledge graph and the relationship between the models. With the help of experts in related fields, the cross-process causes of surface longitudinal crack defects of steel products and related quality events are summarized, and the ontology is created and the relationship between the ontologies is defined by manual modeling.

[0021] The knowledge graph ontology of the traceability of longitudinal crack defects on the surface of steel products includes quality defect model, quality event model, abnormal event model, parameter model, judgment rule model, failure product model and solution model as well as the relationship between each model. The specific information of the model has been described in detail in the content of the invention.

[0022] After the knowledge graph ontology model is designed, a cross-process steel product surface longitudinal crack defect traceability knowledge graph is constructed based on the ontology model and combined with the data source in the steel enterprise production database. Figure 2 shown.

[0023] The implementation scheme is described below with the help of a specific example:

[0024] The quality defect model refers to the quality defect events that may occur in steel products. A corresponding entity node can be established (a: quality defect {defect name: "surface longitudinal crack", process: "continuous casting"});

[0025] In the production of steel products, steel products will go through a series of processing procedures and steps. The quality event model refers to the process events during the processing. The corresponding quality event entity node can be established (b: {quality event name: "Crystallizer liquid level does not match", belonging to the process: "Converter"}), and according to the knowledge graph ontology model, there is a relationship (b)->(a), and the influence degree relationship weight is 0.6. Because there are different steps in the same process, the quality events they represent will also have mutual influence. Therefore, a quality event entity node can be established (c: {quality event name: "Single block casting speed fluctuation", belonging to the process: "Continuous casting"}). According to the ontology model relationship, there is (c)->(a), and the relationship weight is 0.4; at the same time, according to the actual production sequence of steel products, there is (c)->(b), and the relationship weight is 0.4.

[0026] During the processing of steel products, abnormal situations may occur in a certain quality event. Therefore, an abnormal event entity node (d: abnormal event {abnormal event name: "0<pulling speed fluctuation< 0.1m / min", occurrence location: "all production ingots of this furnace", equipment: "continuous casting machine", process: "continuous casting"}) is established. According to the ontology model relationship, there is a (d)->(c) relationship, and the impact weight is 0.2.

[0027] The parameters that affect whether this abnormal event will occur are the actual maximum and minimum pulling speeds, so (e: parameter {parameter name: Metadata: {Continuous casting machine monitoring sensor database}}), according to the ontology model relationship, there is (e)->(d).

[0028] The rule judgment model refers to the judgment rules of the parameters corresponding to the occurrence of abnormal events. Therefore, for the parameters involved in the entity node (e), there will be corresponding rule judgment entity nodes (f: judgment rules {rule description: "During continuous casting, for a single billet, the pulling fluctuation of the continuous casting machine is between 0.1m / min and 0.25m / min, Rule id: 88}), (f)->(e) exists at the same time.

[0029] The data involved in all entity nodes are derived from the actual situation in the actual production and processing of steel enterprises, and the data source is mainly semi-structured data. Based on the knowledge graph ontology model and relationship rules, the steel enterprise data is established as a semi-structured data table in CSV format. By writing scripts, knowledge extraction is performed, and the fields in the CSV table are selectively extracted as attributes and input and output relationships of entity nodes, and the entity attributes after information extraction are mapped with the defined ontology attributes to complete the construction of the knowledge graph for tracing the longitudinal crack defects on the surface of iron products. Among them, there is 1 quality defect entity node, 59 quality event entity nodes, 119 abnormal event entity nodes, 60 parameter entity nodes, and 93 judgment rule instances.

[0030] For the completed knowledge graph of steel product surface longitudinal crack defect traceability, an abstract relationship between the Cypher graph database query language and actual usage requirements is established.

[0031] For products with surface longitudinal crack defects, the knowledge graph is used to conduct finished product traceability analysis:

[0032] First, the billet number of the steel product with the surface longitudinal crack defect must be obtained, so as to know the processing steps that the billet number product has gone through and the production parameters in the corresponding processing equipment monitoring sensor.

[0033] Taking the steel product with the billet number "200900011" as an example, the converter process number "20092301" and the production process parameters of the product are obtained. Then, attribute filtering is performed in the knowledge graph, and the quality events belonging to the converter process and its subordinate abnormal events and the parameters involved are first filtered out. In this way, all the causal paths of the surface longitudinal crack defects of the billet number "200900011" product are generated. Finally, an analysis is performed to match the production process parameters of the "20092301" furnace number with the production rules to which the parameter entity node belongs. If the judgment rule is not met, it means that the corresponding production process parameters are abnormal during production and processing, which eventually leads to the appearance of surface longitudinal crack defects. The real cause of this defect is derived, and the causal path from the judgment rule entity node to the surface longitudinal crack defect entity node is output to provide support for quality decision-making.

Claims

1. A method for tracing the source of longitudinal crack defects on the surface of steel products based on knowledge graph, It is characterized in that The following steps are involved: (1) Constructing the knowledge graph ontology of the traceability of longitudinal crack defects on the surface of steel products based on expert experience knowledge; (2) Based on the ontology and actual production conditions, entity instances and relationship instances are constructed to form a knowledge graph for tracing the defects of longitudinal cracks on the surface of steel products: the instances corresponding to the models in the knowledge graph ontology are specific events that occur during the processing and production of steel products; the entities corresponding to the quality event model are the process steps of steel products in the production process, including "converter slag removal" and "converter re-blowing" in the converter process and "LF argon gas static blowing time mismatch" in the LF refining process; the entities corresponding to the abnormal event model are the abnormal conditions that may occur in the quality event entities in each process step, including "converter steel tapping temperature drop is too large" and "converter re-blowing twice" that may occur in the converter process; construct entities, each entity will correspond to a certain ontology model, and then establish specific relationships between entities for all entities based on the relationships between the ontology models; (3) Based on the constructed knowledge graph entity network, the possible cause paths of the surface longitudinal crack quality defects of steel products are searched: by inputting the "surface longitudinal crack" quality defect of steel products, all entity nodes and relationships pointing to the "surface longitudinal crack" node will be obtained, forming a graph network centered on the "surface longitudinal crack" node; then, according to the actual situation, the screening conditions are input, and the entity types and relationship types appearing in the network are attribute filtered to obtain the cause paths that may lead to the occurrence of surface longitudinal crack defects; Input the billet number of the steel product with surface longitudinal crack defects, and you will get the processing steps of the steel with this billet number, thereby screening out all the causal paths that may cause the surface longitudinal cracks in the processing steps of the steel product with this billet number; (4) Based on the possible paths searched out, combined with actual production data and events, analysis is performed to infer the real cause of the quality defects: According to the billet number of the steel product with quality defects, on the one hand, all processing steps of the billet number product and the abnormal events that may occur in the process are obtained, and then the production process parameter model and judgment rules that will cause the abnormal event are obtained; on the other hand, the actual production parameters of the steel product are obtained from the enterprise production process database, and matched with the judgment rules, so as to obtain the path that does not match the production rules and causes the quality defect; select the surface longitudinal cracks in the quality defects, and the system retrieves that in the production process of the product, in the converter process, the temperature drop of the converter steelmaking process is too large, T 终点温度 -T 大包温度 =73℃, which is greater than the critical value of 65℃ at the end point of the judgment rule, resulting in the appearance of surface longitudinal crack defects; after analysis, the output will be (: judgment rule {rule description: "When the converter is tapping steel, T 终点温度 -T 大包温度 ≥65℃”, rule id: 10})->(:parameter{parameter name:["T 终点温度 ”,"T 大包温度 ”], metadata: {Converter monitoring sensor database}})->(:abnormal event{abnormal event name: "Converter steel temperature drop is too large", occurrence location: "All produced ingots in this furnace", equipment: "Converter equipment name", process: "Converter"})->(:quality event{quality event name: "Converter steel temperature drop", process: "Converter"})->(:quality defect{defect name: "Surface longitudinal crack", process: "Continuous casting"}) this causal path; Through attribute filtering, a cross-process critical path of causes is generated for the longitudinal crack defects on the surface of steel products, and a specific quality event chain is output; at the same time, a critical path flow diagram for tracing the quality defects of steel products is established based on empirical probability and data-driven construction, and the attribution probability of each possible traceability path is quantitatively analyzed to provide support for quality decisions.

2. According to claim 1, the method for tracing the source of longitudinal crack defects on the surface of steel products based on knowledge graph, It is characterized in that The knowledge graph for tracing the source of longitudinal crack defects on the surface of steel products contains five knowledge graph ontology models and their relationships; the specific ontology models are as follows: Quality defect model: refers to the quality defect event that may occur in steel products, such as the "surface longitudinal crack" defect. The quality defect ontology model contains two attributes: the defect name and the process to which it belongs. The outgoing relationship "produced" points to the failed product model, and the relationship is accompanied by a weight. A certain quality defect may cause the steel product to fail during the use of the product, and the weight is the degree of influence of the product defect on the product failure. Quality event model: refers to the process events that steel products go through during processing; the quality event ontology model contains two attributes: the name of the quality event and the process to which it belongs; the outgoing relationship "leads to" points to the quality defect model, and the relationship is accompanied by a weight. A quality event in the process of steel product processing will lead to the appearance of a certain quality defect, and the weight refers to the degree of influence on the quality defect; The outgoing relationship "conduction" points to itself, and the relationship is accompanied by a weight. Each process consists of multiple quality events, and the quality events between different processes also have mutual influence. The weight refers to the degree of influence between different quality events; Abnormal event model: refers to the abnormal situation that may occur in a certain quality event during the processing of steel products; the abnormal event model contains four attributes: abnormal event name, occurrence location, belonging equipment, and belonging process. The output relationship "belongs to" points to the quality event model; Parameter model: refers to the influencing parameters for determining whether abnormal events will occur during the processing of steel products. The determination rule model contains two attributes: parameter name and metadata, and the output relationship points to the abnormal event; The metadata attribute records the location of the parameter in the database table and the access method or interface information; There are situations where multiple parameters jointly determine the occurrence of a certain abnormal event; Judgment rule model: refers to the judgment rule corresponding to the parameter. The judgment rule model contains two attributes: rule description and rule ID. The outgoing relationship points to the parameter model. The rule ID is the serial number of a rule in the rule base.

Images