Belt conveyor fault auxiliary maintenance knowledge graph construction method

Abstract

The invention provides a method for constructing a fault auxiliary maintenance knowledge graph of a belt conveyor. The method comprises the following steps: firstly, collecting fault text data of the belt conveyor; analyzing the collected data so as to design a knowledge graph mode for guiding the construction of a subsequent knowledge graph; performing named entity recognition on sentences in the fault text by utilizing a BERT pre-training model to obtain all entities in the sentences; then, an entity relation triple is extracted by adopting a cascade dichotomy marking framework, namely, all possible subjects (head entities) are recognized from the entities through a subject marker, and objects (tail entities) related to the subjects under a specific relation are recognized from the entities through a specific relation object marker; and finally, completing the construction of the belt conveyor fault auxiliary maintenance knowledge graph through the obtained relation triple. The knowledge graph is constructed based on the pre-training model and the combined learning method, and the belt conveyor fault auxiliary maintenance function can be achieved.

Description

Technical Field

[0001] The present invention relates to the field of belt conveyors, and in particular to a method for constructing a knowledge graph for auxiliary maintenance of belt conveyors. Background Art

[0002] Belt conveyors are widely used in mining, metallurgy, building materials, chemicals, electricity, food processing and other industries due to their advantages such as stable operation, low noise, long-distance transportation, simple maintenance and easy operation, and low energy consumption. Belt conveyors are complex electrical and mechanical coupling systems that operate in harsh working environments for a long time, so they are prone to various failures. If quick and effective measures are not taken, it will lead to significant economic losses and even casualties. In order to ensure that belt conveyors are reliably maintained and failures are effectively handled, a method that can assist maintenance personnel in overhaul is needed.

[0003] As a method of modeling the association between knowledge, knowledge graph can explicitly express the logic of these associations. By combining with natural language processing technology and using text data such as belt conveyor maintenance records and operation logs, knowledge graph can establish the relationship between belt conveyor fault knowledge, and maintenance personnel can make full use of these accumulated historical experience knowledge to assist themselves in maintenance. Triples, as the basic elements of knowledge graphs, are generally obtained through extraction technology. However, the currently commonly used extraction methods based on pipeline learning have problems such as error accumulation, entity redundancy, lack of interaction between subtasks, and entity overlap, resulting in low quality of extracted triples and poor efficiency of the extraction process. In summary, it is necessary to design a knowledge graph construction method based on joint learning to construct a knowledge graph of belt conveyor faults and realize the function of fault-assisted maintenance. Summary of the invention

[0004] The purpose of the present invention is to provide a belt conveyor fault auxiliary maintenance knowledge graph construction method, which can solve the problems of error accumulation, entity redundancy, lack of interaction between subtasks, and entity overlap in the current existing belt conveyor construction method based on pipeline extraction. This method realizes entity extraction and relationship extraction at the same time through a single algorithm, reduces error propagation, enhances the connection between subtasks, improves extraction efficiency, constructs a higher quality knowledge graph, and realizes the belt conveyor fault auxiliary maintenance function in combination with web technology.

[0005] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is as follows: The present invention provides a method for constructing a knowledge graph for auxiliary maintenance of belt conveyor faults, comprising the following steps:

[0006] S1: Collect text data of belt conveyor maintenance logs and operation logs, analyze text content and design a knowledge graph for belt conveyor fault auxiliary maintenance based on application requirements;

[0007] S2: Use the Chinese BERT pre-trained model to perform named entity recognition tasks on the sentences in the collected belt conveyor text data to obtain all entities in the sentences;

[0008] S3: The cascade binary classification tagging framework is used to process the entity. The subject tagger is used to identify all possible subjects from the entity. The specific relation object tagger is used to identify the objects related to the subject under a specific relation from the entity, and finally the entity relation triples are obtained.

[0009] S4: Based on the entity-relationship triples obtained by joint extraction, the knowledge graph of belt conveyor fault auxiliary maintenance is constructed using the Neo4j graph database;

[0010] S5: Utilize the constructed knowledge graph of belt conveyor fault auxiliary maintenance and combine it with web technology to realize historical fault query, fault handling measures query and fault cause query functions to assist maintenance personnel in repairing the belt conveyor.

[0011] Preferably, in step S1, entity types of the pattern include fault location, fault cause, treatment measures and fault phenomenon, entity attributes include operation data and phenomenon details, relationship types include occurrence, cause and treatment, and relationship attributes include fault date.

[0012] Preferably, in step S2, the named entity recognition task using the Chinese BERT pre-trained model is specifically as follows:

[0013] S21: Use the word segmenter to convert the input sequence into a token_id sequence, and then convert the token_id sequence into a high-dimensional representation token embedding through the Embedding Layer. At the same time, the learned position embedding and segment embedding are combined with it to generate the input representation of BERT;

[0014] S22: The bidirectional encoder based on the self-attention mechanism encodes the input representation from the input layer. The bidirectional encoder captures the context information of the input word vector and generates an encoding vector containing rich semantic information after processing by multiple layers of encoders.

[0015] S23: BERT performs named entity recognition from the perspective of sequence labeling problems, and finally outputs the vector representation h of all recognized entitiesN .

[0016] Preferably, in step S3, entity relationship triples are extracted as follows:

[0017] S31: Use the subject tagger to directly decode the encoded vector h output by BERT N , that is, assign a binary label to each token to detect the start and end positions respectively;

[0018] The specific operations for detecting the start position are as follows:

[0019]

[0020] The specific operations for detecting the end position are as follows:

[0021]

[0022] in, represents the probability that the i-th token is recognized as the start position of the subject, represents the probability that the i-th token is recognized as the end position of the subject, X i is the encoding representation of the i-th token in the input sequence, W (·) represents the trainable weights, b (·) represents bias, σ represents activation function;

[0023] S32: The subject tagger optimizes the following likelihood function to identify the subject:

[0024]

[0025] Where L represents the length of the sentence, I{·} is the indicative function, The binary tag indicating the start position of the subject of the i-th token, The binary label representing the end position of the subject of the i-th token, and the parameter θ is {W start ,b start ,W end ,b end};

[0026] S33: Use a specific relation object marker to detect the start and end position of each token object:

[0027] The specific operations for detecting the start position are as follows:

[0028]

[0029] The specific operations for detecting the end position are as follows:

[0030]

[0031] in, represents the probability that the i-th token is recognized as the start position of the object, represents the probability that the i-th token is recognized as the end position of the object, represents the encoded representation of the kth subject identified by the subject tagger; The superscript r means that the matrix W is associated with a specific relation r.

[0032] S34: The object tagger for relation r optimizes the following likelihood function to identify the object:

[0033]

[0034] in, The binary label indicating the starting position of the object of the i-th token, The binary label representing the end position of the object of the i-th token, parameter φ r for

[0035] Preferably, in step S4, the knowledge graph of belt conveyor fault auxiliary maintenance is constructed using the Neo4j graph database as follows:

[0036] S41: Use the LOAD CSV command in the cypher query language to import the csv file containing the node information data into neo4j, and use the CREATE command to create the node;

[0037] S42: Use the LOAD CSV command in the cypher query language to import the csv file containing the relationship information data into neo4j, and use the MATCH command to create relationships between nodes.

[0038] Preferably, in step S5, the auxiliary maintenance function is realized by using the belt conveyor fault auxiliary maintenance knowledge graph and combining web technology as follows:

[0039] S51: Use web technology to build a front-end and back-end platform, with the front-end realizing interaction and visualization with users, and the back-end responsible for processing instructions and data;

[0040] S52: Further, the platform is divided into three modules, namely, data module, business module and application module, wherein the data module is responsible for data storage and management, the business module is responsible for the specific implementation of each function, and the application module is responsible for providing specific application functions and UI interface.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] 1. The present invention uses a pre-trained BERT model to perform named entity recognition tasks. Since BERT uses a bidirectional encoding structure, it can capture word vector context information and generate an encoding vector containing rich semantic information after being processed by a multi-layer encoder;

[0043] 2. The present invention uses a cascaded binary classification labeling framework to process entities, which can reduce error propagation, enhance the connection between subtasks, improve extraction efficiency, and obtain entity relationship triples with better quality;

[0044] 3. The present invention adopts a method based on the belt conveyor fault knowledge graph combined with web technology to realize the belt conveyor fault auxiliary maintenance function, which is convenient for maintenance personnel to operate and use. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0046] Figure 1 It is a flow chart of the method for constructing a knowledge graph for auxiliary maintenance of belt conveyor faults based on joint learning in the present invention;

[0047] Figure 2 Schematic diagram of the knowledge graph model in the present invention;

[0048] Figure 3 Schematic diagram of the structure of the cascaded binary classification labeling framework model in the present invention;

[0049] Figure 4 This is a knowledge graph and application diagram of the belt conveyor fault auxiliary maintenance of the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. Of course, the specific embodiments described here are only used to explain the present invention and are not used to limit the present invention.

[0051] like Figure 1 As shown, an embodiment of the present invention provides a method for constructing a knowledge graph for auxiliary maintenance of a belt conveyor fault, comprising the following steps:

[0052] S1, collect text data such as belt conveyor maintenance logs and operation logs, analyze the text content and design the model of belt conveyor fault auxiliary maintenance knowledge graph in combination with application requirements. The model entity types include fault location, fault cause, treatment measures, fault phenomenon, entity attributes include operation data, phenomenon description, relationship types include occurrence, cause, treatment, and relationship attributes include fault date; the designed belt conveyor fault auxiliary maintenance knowledge graph model is as follows: Figure 2 shown.

[0053] S2, using the Chinese BERT pre-trained model to perform named entity recognition tasks on sentences in the collected text data such as belt conveyor maintenance logs and operation logs, and obtain all entities in the sentences; specifically:

[0054] S21, use the word segmenter to convert the input sequence into a token_id sequence, and then convert the token_id sequence into a high-dimensional representation token embedding through the Embedding Layer, and combine the learned position embedding and segment embedding with it to generate the input representation of BERT. Specifically, the sentence in the text data is used as the input sequence, and the sequence length is set to a fixed value and the value is large. If the actual length of the sentence is insufficient, the sequence is padded with 0. After processing by the Embedding Layer, the word embedding sequence of the sentence is obtained, and then combined with the position embedding and segment embedding to generate the input representation of BERT, where the position embedding selects the sine-cosine position encoding method;

[0055] S22, the bidirectional encoder based on the self-attention mechanism encodes the input representation from the input layer. The bidirectional encoder captures the context information of the input word vector and generates an encoding vector containing rich semantic information after processing by multiple layers of encoders.

[0056] S23, BERT performs named entity recognition from the perspective of sequence labeling problems, and finally outputs the vector representation h of all recognized entities N .

[0057] S3, uses a cascaded binary classification tagging framework to process entities, uses a subject tagger to identify all possible subjects from entities, and uses a specific relationship object tagger to identify objects related to the subject under a specific relationship from entities, and finally obtains entity-relationship triples;

[0058] S31, use the subject tagger to directly decode the encoded vector h output by BERT N , that is, assign a binary label to each token to detect the start and end positions respectively;

[0059] The specific operations for detecting the start position are as follows:

[0060]

[0061] The specific operations for detecting the end position are as follows:

[0062]

[0063] in, represents the probability that the i-th token is recognized as the start position of the subject, represents the probability that the i-th token is recognized as the end position of the subject, X i is the encoding representation of the i-th token in the input sequence, W (·) represents the trainable weights, b (·) represents bias, σ represents activation function;

[0064] S32, the subject tagger optimizes the following likelihood function to identify the subject:

[0065]

[0066] Where L represents the length of the sentence, I{·} is the indicative function, The binary tag indicating the start position of the subject of the i-th token, The binary label representing the end position of the subject of the i-th token, and the parameter θ is {W start ,b start ,W end ,b end};

[0067] S33, using a specific relation object marker to detect the start and end position of each token object:

[0068] The specific operations for detecting the start position are as follows:

[0069]

[0070] The specific operations for detecting the end position are as follows:

[0071]

[0072] in, represents the probability that the i-th token is recognized as the start position of the object, represents the probability that the i-th token is recognized as the end position of the object, represents the encoded representation of the kth subject identified by the subject tagger;

[0073] S34, the object tagger for relation r optimizes the following likelihood function to identify the object:

[0074]

[0075] in, The binary label indicating the starting position of the object of the i-th token, The binary label representing the end position of the object of the i-th token, parameter φ r for The cascade binary classification labeling framework structure is as follows Figure 3 shown.

[0076] S4, based on the entity-relationship triples obtained by joint extraction, the knowledge graph of belt conveyor fault auxiliary maintenance is constructed using the Neo4j graph database;

[0077] S41, using the LOAD CSV command in the cypher query language to import the CSV file containing the node information data into neo4j, and using the CREATE command to create the node;

[0078] S42, using the LOAD CSV command in the cypher query language to import the CSV file containing the relationship information data into neo4j, and using the MATCH command to create relationships between the nodes.

[0079] S5, using the constructed belt conveyor fault auxiliary maintenance knowledge graph and combining it with web technology to realize historical fault query, fault handling measures query, and fault cause query functions, to assist maintenance personnel in repairing the belt conveyor.

[0080] S51, using web technology to build a front-end and back-end platform, the front-end realizes interaction and visualization with users, and the back-end is responsible for command and data processing;

[0081] S52, further, the platform is divided into three modules, namely data module, business module, and application module, where the data module is responsible for data storage and management, the business module is responsible for the specific implementation of each function, and the application module is responsible for providing specific application functions and UI interface; the knowledge graph of belt conveyor fault auxiliary maintenance and its application diagram are shown in Figure 4 shown.

[0082] In summary, the present invention constructs a knowledge graph for auxiliary maintenance of belt conveyor faults based on a joint learning extraction method. The belt conveyor fault text data is collected, and the collected data is analyzed to design a knowledge graph model; the sentences in the fault text are named entity recognized using the BERT pre-training model to obtain all entities in the sentence; the entity relationship triples are extracted using a cascaded binary classification labeling framework, that is, all possible subjects (head entities) are identified from the entity through a subject marker, and objects (tail entities) related to the subject under a specific relationship are identified from the entity through a specific relationship object marker; finally, the construction of the knowledge graph for auxiliary maintenance of belt conveyor faults is completed through the obtained relationship triples, and the belt conveyor fault auxiliary maintenance function is realized in combination with web technology. The above description is only a preferred embodiment of the present invention, and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for constructing a knowledge graph for auxiliary maintenance of belt conveyor faults, characterized in that: The following steps are involved: S1: Collect text data of belt conveyor maintenance logs and operation logs, analyze text content and design a knowledge graph for belt conveyor fault auxiliary maintenance based on application requirements; S2: Use the Chinese BERT pre-trained model to perform named entity recognition tasks on the sentences in the collected belt conveyor text data to obtain all entities in the sentences; S3: The cascade binary classification tagging framework is used to process the entity. The subject tagger is used to identify all possible subjects from the entity. The specific relation object tagger is used to identify the objects related to the subject under a specific relation from the entity, and finally the entity relation triples are obtained. S4: Based on the entity-relationship triples obtained by joint extraction, the knowledge graph of belt conveyor fault auxiliary maintenance is constructed using the Neo4j graph database; S5: Utilize the constructed knowledge graph of belt conveyor fault auxiliary maintenance and combine it with web technology to realize historical fault query, fault handling measures query and fault cause query functions to assist maintenance personnel in repairing the belt conveyor.

2. A belt conveyor fault auxiliary maintenance knowledge graph construction method according to claim 1, characterized in that: In step S1, the entity types of the pattern include fault location, fault cause, treatment measures and fault phenomenon, the entity attributes include operation data and phenomenon details, the relationship types include occurrence, cause and treatment, and the relationship attributes include fault date.

3. A belt conveyor fault auxiliary maintenance knowledge graph construction method according to any one of claims 1-2, characterized in that: In step S2, the named entity recognition task using the Chinese BERT pre-trained model is specifically as follows: S21: Use the word segmenter to convert the input sequence into a token_id sequence, and then convert the token_id sequence into a high-dimensional representation token embedding through the Embedding Layer. At the same time, the learned position embedding and segment embedding are combined with it to generate the input representation of BERT; S22: The bidirectional encoder based on the self-attention mechanism encodes the input representation from the input layer. The bidirectional encoder captures the context information of the input word vector and generates an encoding vector containing rich semantic information after processing by multiple layers of encoders. S23: BERT performs named entity recognition from the perspective of sequence labeling problems, and finally outputs the vector representation h of all recognized entities N .

4. A belt conveyor fault auxiliary maintenance knowledge graph construction method according to any one of claims 1 to 3, characterized in that: In step S3, entity relationship triples are extracted as follows: S31: Use the subject tagger to directly decode the encoded vector h output by BERT N , that is, assign a binary label to each token to detect the start and end positions respectively; The specific operations for detecting the start position are as follows: The specific operations for detecting the end position are as follows: in, represents the probability that the i-th token is recognized as the start position of the subject, represents the probability that the i-th token is recognized as the end position of the subject, X i is the encoding representation of the i-th token in the input sequence, W (·) represents the trainable weights, b (·) represents bias, σ represents activation function; S32: The subject tagger optimizes the following likelihood function to identify the subject: Where L represents the length of the sentence, I{·} is the indicative function, The binary tag indicating the start position of the subject of the i-th token, The binary label representing the end position of the subject of the i-th token, and the parameter θ is {W start ,b start ,W end ,b end }; S33: Use a specific relation object marker to detect the start and end position of each token object: The specific operations for detecting the start position are as follows: The specific operations for detecting the end position are as follows: in, represents the probability that the i-th token is recognized as the start position of the object, represents the probability that the i-th token is recognized as the end position of the object, represents the encoded representation of the kth subject identified by the subject tagger; S34: The object tagger for relation r optimizes the following likelihood function to identify the object: in, The binary label indicating the starting position of the object of the i-th token, The binary label representing the end position of the object of the i-th token, parameter φ r for 5. A belt conveyor fault auxiliary maintenance knowledge graph construction method according to any one of claims 1 to 4, characterized in that: In step S4, the knowledge graph of belt conveyor fault auxiliary maintenance is constructed using the Neo4j graph database as follows: S41: Use the LOAD CSV command in the cypher query language to import the csv file containing the node information data into neo4j, and use the CREATE command to create the node; S42: Use the LOAD CSV command in the cypher query language to import the csv file containing the relationship information data into neo4j, and use the MATCH command to create relationships between nodes.

6. A belt conveyor fault auxiliary maintenance knowledge graph construction method according to any one of claims 1 to 5, characterized in that: In step S5, the auxiliary maintenance function is realized by using the belt conveyor fault auxiliary maintenance knowledge graph and combining web technology as follows: S51: Use web technology to build a front-end and back-end platform, with the front-end realizing interaction and visualization with users, and the back-end responsible for processing instructions and data; S52: Further, the platform is divided into three modules, namely, data module, business module and application module, wherein the data module is responsible for data storage and management, the business module is responsible for the specific implementation of each function, and the application module is responsible for providing specific application functions and UI interface.

Images