Fault and feature fused industrial pump knowledge graph construction method

Abstract

The invention discloses an industrial pump knowledge graph construction method fusing faults and features. The method comprises the steps that firstly, after characteristics of unstructured data are analyzed, ontology construction and knowledge extraction are conducted on preprocessed data, and an industrial pump fault map is created through a neo4j map database; then, on the basis of structured data analysis, entity and attribute extraction of the feature map is completed through a system mechanism, experience knowledge and feature extraction; and then fault classification is carried out on related features by utilizing a machine learning and deep learning algorithm, a relation extraction method is provided for creation of a feature map, and the industrial pump feature map is created. And finally, fusing the fault atlas and the characteristic atlas through an atlas fusion method to form a unified knowledge atlas. And generating a fault report by taking the two partial maps as a search engine. The comprehensive diagnosis capability greatly improves the fault solving efficiency and accuracy, and a detailed fault diagnosis report can be automatically generated.

Description

Technical Field

[0001] The present invention relates to the field of rotating machinery, and in particular to a method for constructing an industrial pump knowledge graph integrating faults and features. Background Art

[0002] Industrial pumps are one of the indispensable key equipment in the chemical, petroleum, electric power, pharmaceutical and other industries, used to transport liquids, gases or other fluids. With the continuous improvement of industrial automation level, the operating status of industrial pumps is directly related to the safe and stable operation of the entire production line. Therefore, the demand for industrial pump status monitoring, fault diagnosis and preventive maintenance is increasing.

[0003] Some of the main problems in the condition monitoring and fault diagnosis of industrial pumps include: each monitoring system operates independently, data is scattered, and an effective information sharing mechanism cannot be formed; expert knowledge and experience have not been effectively integrated, making it difficult to inherit and develop; most of the existing fault diagnosis systems are based on single-dimensional data analysis and lack multi-angle and multi-level information support; traditional methods have poor adaptability to complex and changing working environments and are difficult to meet dynamic adjustment needs. At present, it mainly relies on data collection and analysis of a single sensor. Although this method can reflect the operating status of the equipment to a certain extent, it is often difficult to fully and accurately diagnose complex faults due to the lack of understanding of the overall picture of the equipment. Traditional fault diagnosis methods mainly include threshold-based alarm systems, signal processing-based methods, and machine learning-based methods. However, these methods have their own limitations: threshold-based alarm systems are prone to false alarms or missed alarms; although signal processing-based methods can provide detailed physical parameter analysis, they have limited detection capabilities for atypical faults; although machine learning-based methods have good generalization capabilities, they perform poorly when data is insufficient or feature selection is inappropriate.

[0004] In recent years, with the development of big data, artificial intelligence and Internet of Things technologies, more and more research has begun to focus on how to use these advanced technologies to improve the management and maintenance of industrial pumps. In particular, knowledge graphs, as a structured form of knowledge representation, have been widely used in many fields, including but not limited to recommendation systems and natural language processing. In the application scenario of industrial pumps, building knowledge graphs can effectively integrate resources such as equipment information, operating data, and expert knowledge, providing strong support for fault diagnosis. Summary of the invention

[0005] The main purpose of the present invention is to overcome the problem that data management in the prior art is not intuitive and comprehensive enough, and to provide a method for constructing an industrial pump knowledge graph that integrates faults and features.

[0006] To achieve the above purpose, the technical solution of the present invention is:

[0007] A method for constructing an industrial pump knowledge graph integrating faults and features includes the following steps:

[0008] Step 1: Analyze the data related to industrial pumps and establish an unstructured database of industrial pumps.

[0009] Step 2: Process the vibration signal data set and establish an industrial pump structured database.

[0010] Step 3: Manually extract relevant content from the unstructured database in step 1 and establish an ontology model based on expert knowledge in the field of industrial pumps.

[0011] Step 4: Extract knowledge from the basic information and fault maintenance obtained in step 3 to establish a fault triple.

[0012] Step 5: Store the triples in step 4 into the graph database to construct an industrial pump fault map.

[0013] Step 6: Extract and summarize relevant features of the data in the structured database in step 2 to establish the required nodes and node attributes.

[0014] Step 7: Input the features extracted in step 6 into the intelligent algorithm for fault classification and complete relationship extraction.

[0015] Step 8: Summarize the entities, attributes, and relationships obtained in steps 6 and 7 to establish related triples, and then store them in the graph database to construct an industrial pump feature map.

[0016] Step 9: Further optimize the industrial pump fault map constructed in step 5 using the map fusion method to enhance the internal connection of the map.

[0017] Step 10: The fault graph of step 9 and the feature graph constructed in step 8 are also fused using the graph fusion method to form a unified knowledge graph, and then a diagnosis report is generated.

[0018] Among them, the specific steps of step 1 include: the data of establishing the unstructured database of industrial pumps includes two parts: the first part is to sort out the basic information files of industrial pumps by collecting literature, mainly describing the main components of industrial pumps and the types of failures they may encounter. The second part is the relevant information on the specific industrial pump failure maintenance, mainly including two types of information: one is to obtain some common failures and treatment methods of clean water pumps through web crawlers, mainly involving equipment phenomena, causes of failures and treatment methods; the other is the common failures and treatment measures of slurry pumps collected from the instruction manual of the ZJ series slurry pumps of the industrial pump factory, which also contains equipment phenomena, causes of failures and treatment methods.

[0019] The specific steps of step 2 include: collecting relevant open source data sets, dividing and organizing them into csv files, and storing them in the local hard disk as a structured database.

[0020] The specific steps of step 3 include: the entities that need to be divided from the two parts of data include "components", "fault type" in the first part and "faulty equipment", "fault phenomenon", "fault cause" and "fault handling" in the second part, and the "research object" node is used as the starting point for constructing the relevant map.

[0021] The specific steps of step 4 include: The basic structure of building a knowledge graph usually adopts the form of "knowledge triples", which is a basic unit for representing the relationship between entities. The knowledge triples follow the format of <entity, relationship, entity> to clearly express a segment of entity relationship and complete the knowledge extraction of centrifugal pump text data.

[0022] Step 5 includes: In the created industrial pump fault map, the knowledge in the map should be frequently retrieved and updated. In order to facilitate effective storage and query, Neo4j graph database is selected as the carrier for building and displaying the knowledge map.

[0023] The specific steps of step 6 include: summarizing the relevant structure and mechanism knowledge of the data set, extracting the entity nodes related to it, selecting the commonly used frequency domain feature parameters, and performing one-dimensional feature extraction on the divided data set; using the DATSDP formula Generate a fusion image of the corresponding signal data and obtain the address of the image. Input the frequency domain feature parameters and image address into the same table file as the basic data for subsequent attribute extraction. According to the specific information of each node, input the corresponding attribute data to complete the attribute extraction of the feature map.

[0024] The specific steps of step 7 include: combining the recognition results of machine learning algorithms and deep learning technology to build a comprehensive analysis framework. Input the feature vector of the entire data set into the machine learning algorithm, and use the classification results as information for relationship extraction. At the same time, the classification results of the DATSDP fusion image input into the convolutional neural network are also used as information for relationship extraction. Based on the recognition results of the machine learning algorithm and deep learning, the relationship extraction of the feature map is completed.

[0025] The specific steps of step 8 include: summarizing the obtained entities, attributes, and relationships to establish relevant triples, and also selecting the Neo4j graph database as a carrier for building and displaying the knowledge graph to store triples and construct an industrial pump feature graph.

[0026] The specific steps of step 9 include: the flow of the graph fusion method is to first traverse the relevant type of node parts that need to be connected between the graphs, identify nodes with similar characteristics, and form similar node pairs; then, instead of directly merging the similar node pairs into a single node, establish similar relationships between them; this not only ensures the integrity of the original node information, but also effectively reflects the similarity of the nodes between the graphs; in addition, the other nodes and their relationships originally connected to this pair of similar nodes remain unchanged, and finally form a new graph, which contains the information of the original two graphs and enhances the connection between them. Starting from the fault graph and feature graph of the industrial pump, this method further optimizes the information graph and enhances the connection between the basic information and fault maintenance parts.

[0027] The specific steps of step 10 include: also using the graph fusion method, linking the fault type of the industrial pump component with the diagnosis result of the vibration signal. In this way, a direct connection between the fault type and the vibration signal feature can be established, and then a fault report can be generated.

[0028] The present invention first analyzes the characteristics of unstructured fault data, then constructs an ontology and extracts knowledge from the preprocessed data, and uses the neo4j graph database to create an industrial pump fault map. Then, based on the structured data analysis, the entity and attribute extraction of the feature map is completed through system mechanism, empirical knowledge and feature extraction; then, the relevant features are classified for fault classification using machine learning and deep learning algorithms, and a relationship extraction method is provided for the creation of the feature map to create an industrial pump feature map. Finally, the fault map and the feature map are fused through a graph fusion method to form a unified knowledge map. The two partial maps are used as search engines to generate a fault report. This comprehensive diagnostic capability greatly improves the efficiency and accuracy of fault resolution, and can automatically generate a detailed fault diagnosis report, helping technicians to better understand the problem and take appropriate measures, thereby successfully completing the entire diagnostic process. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0030] Figure 1 A schematic diagram of a process flow of a method for constructing an industrial pump knowledge graph integrating faults and features provided in an embodiment of the present invention;

[0031] Figure 2A schematic diagram of entities divided from the unstructured data of industrial pumps provided in an embodiment of the present invention;

[0032] Figure 3 A schematic diagram of entity relationships of an industrial pump fault map provided by an embodiment of the present invention;

[0033] Figure 4 A schematic diagram of an industrial pump fault spectrum provided by an embodiment of the present invention;

[0034] Figure 5 The construction process of the industrial pump characteristic map provided by the embodiment of the present invention;

[0035] Figure 6 A schematic diagram of a characteristic spectrum of an industrial pump provided by an embodiment of the present invention;

[0036] Figure 7 A schematic diagram of the knowledge graph fusion process provided by an embodiment of the present invention;

[0037] Figure 8 A schematic diagram of a flow chart of a graph fusion method provided in an embodiment of the present invention;

[0038] Fig. 9 A schematic diagram of an industrial pump knowledge graph that partially integrates faults and features provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0039] The present invention is further explained below in conjunction with the accompanying drawings and specific implementations. It should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, various equivalent forms of modifications to the present invention by those skilled in the art all fall within the scope defined by the claims attached to this application.

[0040] Reference Figure 1 The present invention discloses a method for constructing an industrial pump knowledge graph integrating faults and features, comprising the following steps:

[0041] Step 1: Analyze the data related to industrial pumps and establish an unstructured database of industrial pumps.

[0042] Step 2: Process the vibration signal data set and establish an industrial pump structured database.

[0043] Step 3: Manually extract relevant content from the unstructured database in step 1 and establish an ontology model based on expert knowledge in the field of industrial pumps.

[0044] Step 4: Extract knowledge from the basic information and fault maintenance obtained in step 3 to establish a fault triple.

[0045] Step 5: Store the triples in step 4 into the graph database to construct an industrial pump fault map.

[0046] Step 6: Extract and summarize relevant features of the data in the structured database in step 2 to establish the required nodes and node attributes.

[0047] Step 7: Input the features extracted in step 6 into the intelligent algorithm for fault classification and complete relationship extraction.

[0048] Step 8: Summarize the entities, attributes, and relationships obtained in steps 6 and 7 to establish related triples, and then store them in the graph database to construct an industrial pump feature map.

[0049] Step 9: Further optimize the industrial pump fault map constructed in step 5 using the map fusion method to enhance the internal connection of the map.

[0050] Step 10: The fault graph of step 9 and the feature graph constructed in step 8 are also fused using the graph fusion method to form a unified knowledge graph, and then a diagnosis report is generated.

[0051] This embodiment realizes the combination of existing knowledge and real-time data by constructing an industrial pump knowledge graph that integrates faults and features. The method first starts from two aspects: on the one hand, a fault graph is constructed based on unstructured data such as fault records and maintenance history of industrial pumps, and it is used as a knowledge engine; on the other hand, a feature graph is established using structured data such as sensor data and operating parameters as a data engine. Next, the knowledge engine and the data engine are reorganized and fused using graph fusion technology to form a unified knowledge graph. Finally, by traversing and analyzing this fused graph, the problem can be quickly located and a detailed fault report can be generated, thereby completing the entire fault diagnosis task.

[0052] The method is described in detail below.

[0053] Step 1: Establish an unstructured database of industrial pumps.

[0054] The collected data consists of two parts. The first part is to compile basic information files of industrial pumps by collecting literature, mainly describing the main components of industrial pumps and the types of failures they may encounter. The second part is the relevant information on specific industrial pump failure maintenance, mainly including two types of information: one is some common failures and treatment methods of clean water pumps obtained through web crawlers, mainly involving equipment phenomena, failure causes and treatment methods; the other is the common failures and treatment measures of slurry pumps collected from the ZJ series slurry pump instruction manual of the industrial pump factory, which also contains equipment phenomena, failure causes and treatment methods.

[0055] Step 2: Establish an industrial pump structured database.

[0056] This embodiment collects relevant open source data sets, divides and organizes them into csv files, and stores them in the local hard disk as a structured database.

[0057] Step 3: Ontology construction of unstructured database.

[0058] like Figure 2 It is a schematic diagram of the entities divided from the unstructured data of industrial pumps. According to the two-part data structure, the entities that can be divided include "components" and "fault types" in the first part and "faulty equipment", "fault phenomenon", "fault cause" and "fault handling" in the second part.

[0059] Step 4: Extract knowledge from unstructured data and create triples.

[0060] like Figure 3 It is an entity relationship diagram of the industrial pump fault map. The related entities are linked and extracted, and the relationship between the entities is extracted. The knowledge triples follow the format of <entity, relationship, entity> to clearly express a segment of entity relationship to complete the knowledge extraction of centrifugal pump text data. For example, "<bearing heating, cause, bearing pad scraping does not meet the requirements>", in this example, "cause" indicates the relationship type, "bearing heating" is the head entity, representing the problem or phenomenon; "bearing pad scraping does not meet the requirements" is the tail entity, indicating one of the causes of bearing heating. This representation method concisely and clearly reflects the relationship between bearing heating and its potential causes.

[0061] Step 5: Store the triples in the graph database and construct an industrial pump fault map.

[0062] like Figure 4 This is a schematic diagram of the industrial pump fault map. When building a knowledge map, the most suitable storage mode and expression method should be selected according to the characteristics and application scenarios of the data. In the industrial pump knowledge map created in this chapter, the knowledge in the map should be frequently retrieved and updated. In order to facilitate effective storage and query, the Neo4j graph database is selected as the carrier for building and displaying the knowledge map to store triples and construct an industrial pump fault map.

[0063] like Figure 5This is the construction process of the industrial pump feature map, which mainly includes the establishment of the feature map and the classification of information features. First, the structure, mechanism and knowledge of the data set are extracted as rule nodes; then the vibration data is feature extracted as node attributes, including the frequency domain parameters in the xy direction and the DATSDP fusion image extracted in the previous chapter, etc., and the node attributes are combined with the rule nodes to complete the entity extraction; next, the rule nodes are used as classification labels, and the two features are input into the diagnosis algorithm for classification training to complete the relationship extraction; finally, the feature map triples are formed according to the corresponding relationship of "node-relationship-node", and the industrial pump feature map is constructed through the triples.

[0064] Step 6: Extract features of structured data and establish required nodes and node attributes.

[0065] The relevant structure and mechanism knowledge of the data set are summarized and summarized, and the relevant entity nodes are extracted. After the node extraction is completed, the vibration signal data in two directions are subjected to a series of operations such as Fourier transform to obtain the corresponding power spectrum density function, and then the eight frequency domain feature parameters in two directions are calculated to obtain one-dimensional feature extraction; using the DATSDP formula Generate a fusion image of the corresponding signal data and obtain the address of the image. Input the frequency domain feature parameters and image address into the same table file as the basic data for subsequent attribute extraction. According to the specific information of each node, input the corresponding attribute data to complete the attribute extraction of the feature map.

[0066] Step 7: Input the extracted features into the intelligent algorithm for fault classification and complete relationship extraction.

[0067] A comprehensive analysis framework is constructed by combining the recognition results of machine learning algorithms and deep learning technology. The feature vector of the entire data set is input into the machine learning algorithm, and the classification results are used as information for relationship extraction. At the same time, the classification results of the DATSDP fused image input into the convolutional neural network are also used as information for relationship extraction. Based on the recognition results of machine learning algorithms and deep learning, the relationship extraction of feature maps is completed.

[0068] Step 8: Summarize the obtained entities, attributes, and relationships to establish relevant triples, and then store them in the graph database to construct the industrial pump feature map.

[0069] like Figure 6 It is a schematic diagram of the characteristic map of industrial pumps. The reference information map is constructed, and Neo4j graph data is also used for storage and display. The triples are exported in CSV file format, and the characteristic map of the triples is created using the Neo4j graph database.

[0070] like Figure 7This is a schematic diagram of the knowledge graph fusion process. It starts from two aspects: on the one hand, a fault graph is built based on unstructured data such as fault records and maintenance history of industrial pumps, and used as a knowledge engine; on the other hand, a feature graph is built using structured data such as sensor data and operating parameters as a data engine. Next, the knowledge engine and the data engine are reorganized and fused using graph fusion technology to form a unified knowledge graph. Finally, by traversing and analyzing this fused graph, problems can be quickly located and detailed diagnostic reports can be generated.

[0071] like Figure 8 It is a flowchart of the graph fusion method. First, the relevant types of nodes that need to be connected between graphs are traversed, nodes with similar characteristics are identified, and similar node pairs are formed. Then, instead of directly merging the similar node pairs into a single node, similar relationships are established between them. This not only ensures the integrity of the original node information, but also effectively reflects the similarity of the nodes between graphs. In addition, the other nodes originally connected to this pair of similar nodes and their relationships remain unchanged, and finally a new graph is formed, which contains the information of the original two graphs and enhances the connection between them.

[0072] The following is an introduction to the important steps in generating similar node pairs:

[0073] (1) Chinese-English mapping

[0074] Since the node names in the two graphs are expressed in different ways, usually in a mixture of English and Chinese, a Chinese-English mapping dictionary needs to be established to ensure the consistency and convenience of subsequent processing. Through this mapping dictionary, the English names in the graph can be converted into corresponding Chinese names, thereby achieving the purpose of name standardization. This not only eliminates the obstacles caused by language differences, but also simplifies the subsequent data processing process.

[0075] (2) Word segmentation and word vector generation

[0076] There are some useless words and conjunctions in the node text data. These words often do not carry substantive meaning, so the text needs to be segmented and denoised. At present, the more mature word segmentation technologies include Jieba word segmentation library in Python, ICTCLAS and IK word segmenters in Java, and ASPSeek in C++. In this section, Jieba word segmentation library is used to achieve this goal. Jieba is a very popular Chinese word segmentation tool. It supports multiple word segmentation strategies such as precise mode, full mode and search engine mode, and can effectively remove stop words, greatly improving the efficiency and accuracy of text processing. Through Jieba word segmentation, the original text can be segmented into meaningful word sequences, laying a good foundation for further text analysis.

[0077] After word segmentation, text vectors need to be represented. There are two main methods for representing text vectors. The first is one-hot encoding, but one-hot encoding may cause a large number of redundant sparse matrices and cause dimensionality disasters; the second is distributed encoding, among which Word2Vec is more commonly used. This section uses the distributed encoding Word2Vec model to obtain word vectors.

[0078] (3) Node similarity calculation

[0079] After the node name word vector is generated, the similarity between two nodes is judged. Cosine similarity is a similarity calculation method based on the vector space model, which is often used for text similarity calculation. Its calculation formula is: Among them, A and B are two vectors, ||A|| and ||B|| are their moduli respectively.

[0080] Step 9: Use the graph fusion method to further optimize the fault graph.

[0081] Starting from the fault map and characteristic map of industrial pumps, this method is used to further optimize the fault map and enhance the connection between the basic information and fault maintenance.

[0082] Step 10: Fuse the fault graph and feature graph to form a unified knowledge graph.

[0083] like Fig. 9 This is a schematic diagram of the industrial pump knowledge graph that partially integrates faults and features. The graph fusion method is also used to link the fault types of industrial pump components with the diagnostic results of vibration signals. In this way, a direct connection between the fault type and the vibration signal characteristics can be established, and then a diagnostic report can be generated.

Claims

1. A method for constructing an industrial pump knowledge graph integrating faults and features, characterized in that: Step 1: Analyze the data related to industrial pumps, including basic information files of industrial pumps and relevant data on specific industrial pump failure maintenance, and establish an unstructured database of industrial pumps; Step 2: Taking the open source bearing data set as an example, analyze the structure of the data set and perform preprocessing to establish an industrial pump structured database; Step 3: Manually extract relevant content from the unstructured database in step 1, and establish an ontology model of basic information and fault maintenance based on expert knowledge in the field of industrial pumps; Step 4: Extract knowledge from the ontology obtained in step 3 and establish triples; Step 5: Store the triples in step 4 into the graph database to construct an industrial pump fault map; Step 6: extract and summarize relevant features of the data in the structured database in step 2 to establish the required nodes and node attributes; Step 7: Input the features extracted in step 6 into the intelligent algorithm for fault classification and complete relationship extraction; Step 8: Summarize the entities, attributes, and relationships obtained in steps 6 and 7, establish related triples, and then store them in the graph database to construct the industrial pump feature map; Step 9: further optimize the industrial pump fault map constructed in step 5 by using the map fusion method to enhance the internal connection of the map; Step 10: The fault graph of step 9 and the feature graph constructed in step 8 are fused using the graph fusion method to form a unified knowledge graph, and then a diagnosis report is generated.

2. The method for analyzing data related to industrial pumps and establishing an unstructured database of industrial pumps according to claim 1, characterized in that: The data for establishing the unstructured database of industrial pumps includes two parts. The first part is to compile the basic information files of industrial pumps by collecting literature, which mainly describes the main components of industrial pumps and the types of failures they may encounter. The second part is the relevant information on the specific industrial pump failure maintenance, which mainly includes two types of information: one is the common failures and treatment methods of clean water pumps obtained through web crawlers, which mainly involve equipment phenomena, failure causes and treatment methods; the other is the common failures and treatment measures of slurry pumps collected from the instruction manual of the ZJ series slurry pumps of the industrial pump factory, which also contains equipment phenomena, failure causes and treatment methods.

3. The method for processing vibration signal data set and establishing industrial pump structured database according to claim 1, characterized in that: Collect relevant open source data sets, divide and organize them into csv files, and store them on the local hard disk as a structured database.

4. According to claim 1, the relevant contents in the unstructured database are manually extracted, and the ontology model is established in combination with the expert knowledge in the field of industrial pumps, characterized in that: The entities divided from the two parts of data include "components", "fault type" in the first part and "faulty equipment", "fault phenomenon", "fault cause" and "fault handling" in the second part, and the "research object" node is used as the starting point for the construction of the relevant map.

5. Extract knowledge and establish triples based on the ontology obtained according to claim 1, characterized in that: The basic structure for building a knowledge graph usually takes the form of "knowledge triples", which are the basic units for representing the relationship between entities. Knowledge triples follow the format of <entity, relationship, entity> to clearly express an entity relationship and complete the knowledge extraction of centrifugal pump text data.

6. According to claim 1, the triples are stored in a graph database to construct an industrial pump fault map, characterized in that: In the created industrial pump knowledge graph, the knowledge in the graph needs to be frequently retrieved and updated. In order to facilitate effective storage and query, Neo4j graph database is selected as the carrier for building and displaying the knowledge graph.

7. According to claim 1, extract and summarize relevant features of data in the structured database to establish required nodes and node attributes, characterized in that: Summarize the relevant structure and mechanism knowledge of the data set and extract the relevant entity nodes. Select the frequency domain feature parameters and perform one-dimensional feature extraction on the data set; Using the DAT SD P formula Generate a fusion image of the corresponding signal data and obtain the address of the image. Input the frequency domain feature parameters and image address into the same table file as the basic data for subsequent attribute extraction; input the corresponding attribute data according to the specific information of each node to complete the attribute extraction of the feature map.

8. According to claim 1, the extracted features are input into an intelligent algorithm for fault classification to complete relationship extraction, characterized in that: Combine the recognition results of machine learning algorithms and deep learning technology to build a comprehensive analysis framework; input the feature vector of the entire data set into the machine learning algorithm, and use the classification results as information for relationship extraction; at the same time, input the classification results of the DATSDP fused image into the convolutional neural network as information for relationship extraction. Based on the recognition results of machine learning algorithms and deep learning, complete the relationship extraction of feature maps.

9. According to claim 1, the entities, attributes and relationships obtained are summarized, relevant triples are established and stored, and a characteristic map of industrial pumps is constructed, characterized in that: The obtained entities, attributes, and relationships are summarized to establish relevant triples. The Neo4j graph database is also selected as the carrier for building and displaying the knowledge graph to store triples and construct the industrial pump feature graph.

10. The industrial pump fault map constructed according to claim 1 is further optimized by using a map fusion method, characterized in that: The process of the graph fusion method is to first traverse the relevant type of nodes that need to be connected between graphs, identify nodes with similar characteristics, and form similar node pairs; then, instead of directly merging similar node pairs into a single node, establish similar relationships between them; this not only ensures the integrity of the original node information, but also effectively reflects the similarity of nodes between graphs; in addition, other nodes originally connected to this pair of similar nodes and their relationships remain unchanged, and finally form a new graph that contains the information of the original two graphs and enhances the connection between them. Starting from the fault graph and feature graph of industrial pumps, this method further optimizes the information graph and enhances the connection between the basic information and fault maintenance parts.

11. According to claim 1, the fault graph and the feature graph are fused by using the graph fusion method to form a unified knowledge graph and then generate a diagnosis report, characterized in that: The graph fusion method is also used to link the fault type of industrial pump components with the diagnosis results of the vibration signal. In this way, a direct connection between the fault type and the vibration signal characteristics can be established, and a diagnostic report can be generated.

Images