An Excavator Fault Auxiliary Decision-making Method Based on Graph Neural Network

Abstract

An embodiment of the present invention discloses a method for auxiliary decision-making of excavator faults based on a graph neural network, which relates to the field of excavator fault diagnosis and can solve problems such as poor timeliness and low accuracy caused by manual diagnosis and handling of faults. The present invention includes: preprocessing the received fault work orders of the excavator, and obtaining a fault knowledge graph by using the preprocessed fault work orders; querying the processing scheme data corresponding to the fault work orders, and constructing an auxiliary decision matrix according to the fault work orders and the corresponding processing scheme data; constructing a graph neural network model, and training the graph neural network model through the fault knowledge graph and the auxiliary decision matrix; receiving the text data reported by the currently faulty excavator, inputting the trained graph neural network model, and outputting the processing scheme data for the currently faulty excavator.

Description

Technical Field

[0001] The present invention relates to the field of excavator fault diagnosis, and in particular to an excavator fault auxiliary decision-making method based on a graph neural network. Background Art

[0002] Crawler excavators have advantages such as strong power and high flexibility. They are the most widely used type of excavators and are known as the "barometer of economic activities". They are widely used in various engineering activities such as civil construction, mineral extraction, and infrastructure construction. Crawler excavators operate under relatively harsh conditions and complex working conditions, and many factors may cause them to malfunction. Moreover, the maintenance of crawler excavators is difficult. If not handled properly, it will not only cause secondary damage to the equipment, but also pose a threat to the personal safety of the operators. If these problems are not solved, it is very likely to limit the development of crawler excavators. Summarizing the common excavator fault problems during the operation process and selecting the most appropriate maintenance method during on-site maintenance can promote the smooth progress of construction operations on the basis of ensuring the personal safety of the operators, and help improve the construction efficiency of enterprises, extend the service life of excavators, and increase the economic benefits of enterprises. Therefore, researching the fault diagnosis auxiliary decision-making method for crawler excavators is of crucial significance.

[0003] At present, the field of fault diagnosis related to crawler excavators lacks a complete and reliable knowledge base, and there is still much room for development in the reasoning strategies that can assist decision-making. When an excavator malfunctions at the job site, relying solely on the experience of the operator to troubleshoot is inefficient and unreliable. Summary of the Invention

[0004] Embodiments of the present invention provide an excavator fault auxiliary decision-making method based on a graph neural network, which can solve problems such as poor timeliness and low accuracy caused by manual diagnosis and fault handling.

[0005] To achieve the above object, the embodiments of the present invention adopt the following technical solutions:

[0006] S1. Preprocess the received fault work order of the excavator, and obtain a fault knowledge graph by using the preprocessed fault work order;

[0007] S2. Query the processing scheme data corresponding to the fault work order, and construct an auxiliary decision matrix according to the fault work order and the corresponding processing scheme data;

[0008] S3. Construct a graph neural network model, and train the graph neural network model by using the fault knowledge graph and the auxiliary decision matrix;

[0009] S4. Receive the text data reported by the currently malfunctioning excavator, input it into the trained graph neural network model, and output the processing solution data for the currently malfunctioning excavator.

[0010] An embodiment of the present invention discloses a method for auxiliary decision-making of excavator faults based on a graph neural network, which relates to the field of fault diagnosis of crawler excavators. The present invention includes: extracting entities and relationships from semi-structured fault work orders to construct an excavator fault knowledge graph; constructing an auxiliary decision matrix based on fault phenomena and corresponding methods for handling faults; constructing a graph neural network framework, sending the fault knowledge graph and the auxiliary decision matrix into the network model, and optimizing the parameters to be trained in the model; obtaining the fault text data during the operation of the excavator, extracting the fault phenomena after preprocessing, and sending them into the trained model to predict the corresponding fault handling methods. By using the model framework of the graph neural network, the efficiency and accuracy of decision-making can be effectively improved, and the problems of poor timeliness and low accuracy caused by manual diagnosis and fault handling are solved. Description of the Drawings

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0012] Figure 1 It is a schematic diagram of the overall method flow provided by the embodiment of the present invention;

[0013] Figure 2 It is a partial schematic diagram of the excavator fault knowledge graph provided by the embodiment of the present invention;

[0014] Figure 3 It is a schematic diagram of the principle of the receptive field of the graph neural network provided by the embodiment of the present invention;

[0015] Figure 4 It is a schematic diagram of the information aggregation framework of the graph neural network provided by the embodiment of the present invention. Detailed Embodiments

[0016] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. The embodiments of the present invention will be described in detail hereinafter. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be construed as a limitation to the present invention. Those skilled in the art of the present technology can understand that unless specifically stated otherwise, the singular forms "a", "an", "the" and "said" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the specification of the present invention means the presence of the described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we say that an element is "connected" or "coupled" to another element, it can be directly connected or coupled to other elements, or there may also be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or coupling. The phrase "and / or" used herein includes any and all combinations of one or more of the associated listed items. Those skilled in the art of the present technology can understand that unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as commonly understood by those of ordinary skill in the art to which the present invention belongs. It should also be understood that terms such as those defined in a general dictionary should be understood to have a meaning consistent with the meaning in the context of the prior art, and will not be interpreted in an idealized or overly formal sense unless defined as herein.

[0017] At present, there is a lack of a complete and reliable knowledge base in the field of fault diagnosis related to crawler excavators, and there is also a large room for development in the reasoning strategy that can assist decision-making. When a fault occurs at the operation site of the excavator, relying solely on the experience of the operator to troubleshoot the fault is inefficient and unreliable. The design idea given in this embodiment is: construct a fault knowledge graph of the excavator, and on this basis, design a reasoning algorithm to provide efficient and reliable processing suggestions for the operator when a fault occurs, analyze and eliminate common faults of the excavator, which is of great significance for ensuring the safe operation of the excavator and improving work efficiency.

[0018] An embodiment of the present invention provides an excavator fault auxiliary decision-making method based on a graph neural network, including:

[0019] S1. Preprocess the received fault work order of the excavator, and obtain a fault knowledge graph by using the preprocessed fault work order.

[0020] Among them, existing trouble tickets can be used to extract entities and relationships, construct RDF triples and store them in the csv file format, and import them into the Neo4j graph database to complete the construction of the basic fault knowledge graph. After that, the newly generated trouble tickets during the operation process are preprocessed to complement and improve the original fault knowledge graph.

[0021] S2. Query the processing solution data corresponding to the trouble ticket, and construct an auxiliary decision matrix according to the trouble ticket and the corresponding processing solution data.

[0022] Among them, the fault phenomena described in the trouble ticket and the processing methods for this fault can be extracted and used as the rows and columns of the matrix respectively to construct an auxiliary decision matrix.

[0023] S3. Construct a graph neural network model, and train the graph neural network model through the fault knowledge graph and the auxiliary decision matrix.

[0024] Among them, a graph neural network framework can be constructed. By feeding the knowledge graph and the auxiliary decision matrix into the network, the parameters are optimized to improve the performance of the model, and the validation set is used to test the accuracy.

[0025] S4. Receive the text data reported by the excavator with a current fault, input it into the trained graph neural network model, and output the processing solution data for the excavator with the current fault.

[0026] Specifically, obtain the fault text data during the operation of the excavator, preprocess it, extract the fault phenomenon text data, and feed it into the trained model to predict the corresponding fault processing method to assist the operator in making decisions.

[0027] This embodiment designs an excavator fault auxiliary decision-making scheme based on a graph neural network. The experience and knowledge in the trouble ticket are used to construct an excavator fault knowledge graph, and the original graph is continuously improved during actual use, providing a good knowledge base foundation for the reasoning task. In terms of the training strategy, the excavator fault knowledge graph is combined with the auxiliary decision matrix, and a graph neural network framework is used to improve the generalization ability of the model. The present invention can realize the online auxiliary fault diagnosis and processing decision-making functions of the excavator, with high reliability and efficiency.

[0028] In this embodiment, in S1, it includes: extracting entity text data and text data recording the relationships between various entities from the trouble ticket. After classifying the entities through a text convolutional neural network model, RDF triples of the relationships between the entities and various entities are generated. The generated RDF triples are imported into the Neo4j graph database and the knowledge graph is built.

[0029] Further, it also includes: when a new fault occurs, using the new fault work order to update the constructed knowledge graph.

[0030] Specifically, in the stage of constructing the fault knowledge graph, semi-structured excavator fault work orders can be collected, and entity texts and relationship texts between various entities can be extracted from the fault work orders. Use an efficient text convolutional neural network model to classify entities, construct RDF triples of entities and relationships based on rules, and import the triples into the Neo4j graph database to complete the construction of the knowledge spectrum. When a new fault occurs, preprocess the fault text data, use the trained convolutional neural network model to automatically classify the entities and import them into the original fault knowledge graph, and complete the complementation and improvement of the graph.

[0031] For example, in the process of constructing the fault knowledge graph: existing fault work orders can be used to construct the excavator fault knowledge graph. Entities and relationships are extracted from the fault work orders and labeled, and a convolutional neural network for entity classification is trained to optimize the network parameters to prepare for subsequent entity automatic classification tasks. Construct RDF triples of entities and relationships and store them in the csv file format, and import them into the Neo4j graph database to complete the construction of the basic fault knowledge graph. Then, continue to improve and supplement the fault knowledge graph. Extract entities from the newly generated fault work orders during the operation process, use the convolutional neural network model trained in step 1 to automatically classify the entities, construct RDF triples according to the classification results based on rules, and add them to the original fault knowledge graph to complete its complementation and improvement.

[0032] In this embodiment, in S2, it includes:

[0033] Taking the processing solutions and fault phenomena as the rows and columns of a matrix respectively, to obtain an auxiliary decision matrix Y ∈ R M×N . When the processing solution can solve the current fault phenomenon, the corresponding element value in the auxiliary decision matrix is 1, otherwise it is 0.

[0034] Among them, M kinds of fault phenomena are represented as Representing N kinds of processing solutions as Both M and N are positive integers.

[0035] Specifically, in the process of the offline training stage: it is necessary to first construct a graph neural network; then screen samples, and divide the training set, validation set, and test set in the ratio of 6:2:2 in the optimal solutions; use the training set to optimize the parameters of the multi-task learning model; use the validation set to test the accuracy of the model; the trained model has good generalization ability and can predict the possible processing methods corresponding to the newly generated fault phenomena.

[0036] Among them, each layer of the graph neural network mainly contains two operations. First, it is necessary to calculate the degree of association between entities and relationships in the knowledge graph to provide a weight reference for subsequent aggregation operations. The other is the aggregation operation, which aggregates entities and their associated entities in the form of vector representations and converges the feature information to the center.

[0037] (i) Association degree algorithm: Considering each pair of fault phenomena f and handling methods s in the auxiliary decision matrix Y ∈ R M×N , use to represent all entities associated with the handling method entity in the knowledge graph, to represent entity s i and entity s j relationship between.

[0038] In this embodiment, in the process of constructing the graph neural network model, it includes:

[0039] Represent the degree of association between the fault phenomenon entity and various entities as an inner product function g, where where f ∈ R d , r ∈ R d is the vector representation of the fault phenomenon entity f and the relationship r, d represents the dimension of the vector, represents the importance of the relationship r for the fault phenomenon f.

[0040] In order to describe the approximate topological structure of the handling method entity s in the knowledge graph, it is necessary to calculate the linear combination of the association degrees of the neighborhood of s:

[0041]

[0042] where e is the vector representation of the neighborhood entity of the handling method entity s in the knowledge graph, is the normalization form of , and

[0043] Among them, the aggregation operation is the core operation form of the graph neural network, and an addition aggregator can be set in each layer of the graph neural network sum , and aggregate the vector representation s of the handling method entity and the vector representation of its neighborhood entity into vectors with the same dimension:

[0044]

[0045] where W is the network parameter to be trained, b is the bias term, and σ is the ReLU activation function.

[0046] After aggregation, a handling method entity vector is bound to its neighborhood entities, aggregating the features of all neighborhood entities.

[0047] In this embodiment, the process of training the graph neural network model includes:

[0048] To improve the computational efficiency, a negative sampling strategy is adopted during the training process. Specifically, the loss function is minimized for each batch of input samples:

[0049]

[0050] where is the cross-entropy loss function, P is the negative sampling distribution and follows a uniform distribution, T f is the number of negative samples of the fault phenomenon f, y fs indicates that the corresponding processing solution can handle such faults in the actual situation, represents the predicted probability that the processing solution is effective, indicates that the negative sample of the processing solution cannot handle such faults, represents the predicted probability that the negative sample of the processing solution is effective, s i represents the negative sample of the fault processing solution, indicates that the negative sample of the fault processing solution follows the negative sampling distribution.

[0051] When the value of the loss function stabilizes and no longer decreases, the training stops.

[0052] Specifically, during the online testing phase, the fault text data generated during the operation of the excavator is obtained. After preprocessing, the fault information is supplemented into the fault knowledge graph, and the structured fault information is sent into the trained model, so as to predict the corresponding fault processing method to assist the operator in making decisions.

[0053] The embodiment of the present invention can adjust the application mode according to the specific scenario. For example, in some application scenarios, the solution can be divided into two stages, as Figure 1 shown. In the first stage, a graph neural network model is established and trained. In the second stage, online auxiliary decision-making is carried out. The steps include:

[0054] Specifically, in the first stage, as Figure 3 shown, 1 construct a fault knowledge graph, 2 construct an auxiliary decision matrix, 3 construct a graph neural network model framework, and train the model.

[0055] Specifically, the 1 constructing a fault knowledge graph includes:

[0056] Step 1.1: Construct an excavator fault knowledge graph using existing fault work orders. Extract entities and relationships from the fault work orders and label them. Train a convolutional neural network for entity classification, optimize the network parameters, and prepare for subsequent entity automatic classification tasks. Construct RDF triples from the entities and relationships and store them in the csv file format, and import them into the Neo4j graph database to complete the construction of the basic fault knowledge graph.

[0057] Step 1.2: Improve and supplement the fault knowledge graph. Extract entities from the newly generated fault work orders during operation, use the trained convolutional neural network model to automatically classify the entities, construct RDF triples according to the rules based on the classification results, and add them to the original fault knowledge graph to complement and perfect it.

[0058] Specifically, construct an auxiliary decision matrix, including:

[0059] Construct an auxiliary decision matrix based on different fault phenomena and corresponding methods for handling faults. Describe M fault phenomena as Describe N handling methods as Taking the fault phenomena and handling methods as the rows and columns of the matrix respectively, the auxiliary decision matrix Y∈R M×N can be obtained. When the fault handling method can solve the current fault, the corresponding element value in the matrix is 1, otherwise it is 0.

[0060] Specifically, construct a graph neural network model framework and train the model, including:

[0061] Step 3.1: Construct a graph neural network. Each layer of the graph neural network mainly contains two operations. First, it is necessary to calculate the degree of association between entities and relationships in the knowledge graph to provide a weight reference for subsequent aggregation operations. The other is the aggregation operation, which aggregates entities and their associated entities in the form of vectors, and converges the feature information to the center.

[0062] (i) Association degree algorithm: Considering each pair of fault phenomena f and handling methods s in the auxiliary decision matrix Y∈R M×N , use to represent all entities associated with the handling method entity in the knowledge graph, to represent the relationship between entity s i and entity s j . Define an inner product function g to calculate the degree of association between the fault phenomenon entity and the relationship:

[0063]

[0064] where f∈R d , r∈R d are the vector representations of the fault phenomenon entity f and the relationship r, and d is the dimension of the representation vector. Indicates the importance level of relationship r for fault phenomenon f.

[0065] To approximately describe the topological structure of processing method entity s in the knowledge graph, it is necessary to calculate the linear combination of the association degrees of the neighborhood of s:

[0066]

[0067] where e is the vector representation of the neighborhood entities of processing method entity s in the knowledge graph, is the normalized form of :

[0068]

[0069] (ii) Information aggregation operation: The aggregation operation is the core operation form of the graph neural network. An additive aggregator sum is introduced in each layer of the graph neural network to aggregate the vector representation s of the processing method entity and the vector representations of its neighborhood entities into a vector with the same dimension:

[0070]

[0071] where W are the parameters to be trained in the network, b is the bias term, and σ is the ReLU activation function. After aggregation, a processing method entity vector is bound to the entities in its neighborhood, aggregating all the neighborhood entity features.

[0072] Step 3.2, Train the model. To improve the computational efficiency, a negative sampling strategy is adopted during the training process. Minimize the loss function on each batch of input samples:

[0073]

[0074] where is the cross-entropy loss function, P is the negative sampling distribution, following a uniform distribution. T f is the number of negative samples of fault phenomenon f. When the value of the loss function stabilizes and no longer decreases, stop the training.

[0075] Second stage, online auxiliary decision-making: Obtain the fault text data generated during the operation of the excavator, generate test examples after preprocessing, and input the test examples into the trained auxiliary decision-making model to obtain the predicted processing method.

[0076] Specifically, obtaining the fault text data generated during the operation of the excavator, generating test examples after preprocessing, and inputting the test examples into the auxiliary decision-making model trained in 3 to obtain the predicted processing method, including:

[0077] Obtain the fault text data generated during the operation of the excavator. After preprocessing, use the same method as before to supplement the fault information into the fault knowledge graph. Send the structured fault information into the trained model to predict the corresponding fault handling methods to assist the operator in making decisions.

[0078] The excavator fault auxiliary decision-making system proposed in this embodiment, which is based on multi-task learning and knowledge graph embedding, can be used to provide efficient and reliable diagnostic suggestions for maintenance personnel when the excavator fails. This method can extract the associated potential features of the two by performing interactive operations on the word vectors of the head entity text of the fault phenomenon category in the fault knowledge graph and the text of the handling methods for this fault phenomenon in the auxiliary decision-making matrix. By using the graph neural network model, the generalization ability of the model can be improved and the problems of cold start and data sparsity can be effectively solved. In addition, this framework has good transferability and still has high accuracy when making auxiliary decisions for other mechanical faults.

[0079] Each embodiment in this specification is described in a progressive manner. The same or similar parts among the embodiments can be referred to each other, and the differences between each embodiment and other embodiments are emphasized. In particular, for the device embodiments, since they are basically similar to the method embodiments, they are described relatively simply, and the relevant parts can be referred to the partial description of the method embodiments. The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. An auxiliary decision-making method for excavator faults based on graph neural network, characterized in that, Including: S1. Preprocess the received fault work orders of the excavator, and obtain a fault knowledge graph by using the preprocessed fault work orders; S2. Query the processing scheme data corresponding to the fault work order, and construct an auxiliary decision matrix according to the fault work order and the corresponding processing scheme data; S3. Construct a graph neural network model, and train the graph neural network model by using the fault knowledge graph and the auxiliary decision matrix; S4. Receive the text data reported by the excavator with a current fault, input the trained graph neural network model, and output the processing scheme data for the excavator with the current fault; In S2, it includes: Taking the processing solutions and fault phenomena as the rows and columns of a matrix respectively, an auxiliary decision matrix Y ∈ R M×N is obtained, where M fault phenomena are represented as and N processing solutions are represented as Both M and N are positive integers; When the processing scheme can solve the current fault phenomenon, the corresponding element value in the auxiliary decision matrix is 1, otherwise it is 0; In the process of constructing the graph neural network model, it includes: The correlation degree between the fault phenomenon entity and various entities is expressed as an inner product function g, where where f ∈ R d , r ∈ R d is the vector representation of the fault phenomenon entity f and the relationship r, d represents the dimension of the vector, represents the importance of the relationship r for the fault phenomenon f; Calculate the linear combination of the association degrees of the neighborhood of s: where e is the vector representation of the neighborhood entities of the processing method entity s in the knowledge graph, is the normalized form of and It also includes: Set an additive aggregator in each layer of the graph neural network sum , and aggregate the vector representation s of the processing method entity and the vector representations of its neighboring entities into vectors with consistent dimensions: Wherein, W is the network parameter to be trained, b is the bias term, and σ is the ReLU activation function; The process of training the graph neural network model includes: During the training process, a negative sampling strategy is adopted. Among them, the loss function is minimized for each batch of input samples: Among them, is the cross-entropy loss function, P is the negative sampling distribution and follows a uniform distribution, T f is the number of negative samples of the fault phenomenon f, y fs indicates that the corresponding processing solution can handle such faults in the actual situation, represents the predicted probability that the processing solution is effective, indicates that the negative sample of the processing solution cannot handle such faults, represents the predicted probability that the negative sample of the processing solution is effective, s i represents the negative sample of the fault processing solution, indicates that the negative sample of the fault processing solution follows the negative sampling distribution; When the value of the loss function stabilizes and no longer decreases, stop training.

2. The method according to claim 1, characterized in that, In S1, it includes: Extract entity text data and text data recording the relationships between various entities from the fault work order; After classifying the entities through the text convolutional neural network model, generate RDF triples of the relationships between the entities and various entities; Import the generated RDF triples into the Neo4j graph database and build a knowledge graph.

3. The method according to claim 2, characterized in that, It also includes: When a new fault occurs, use the new fault work order to update the built knowledge graph.

Images