A System and Method for Equipment Fault Diagnosis Based on Large Model and Knowledge Graph Collaborative Reasoning

Abstract

This invention provides a system and method for equipment fault diagnosis based on collaborative reasoning using a large model and knowledge graph. It relates to the field of equipment fault diagnosis technology and includes: a data layer for acquiring real-time operating data of the equipment under test and preprocessing the data; a knowledge layer connected to the data layer for analyzing the preprocessed real-time operating data to determine the presence of fault data; and a model layer connected to the knowledge layer for analyzing the fault data when the analysis results output by the knowledge layer determine the presence of fault data, thereby obtaining the cause of the fault and the corresponding maintenance plan. This invention deeply integrates data processing, fault identification, intelligent analysis, predictive warning, and lifespan assessment, achieving an upgrade in equipment fault diagnosis from passive response to proactive prevention, and from single-data judgment to multi-dimensional collaborative reasoning.

Description

Technical Field

[0001] This invention relates to the field of equipment fault diagnosis technology, and more specifically to an equipment fault diagnosis system and method based on collaborative reasoning using large models and knowledge graphs. Background Technology

[0002] Currently, industrial equipment is an indispensable and important component of modern industrial production. It is widely used in manufacturing, construction, energy, agriculture, and service industries, and is a key factor in promoting economic and social development and improving production efficiency and product quality. Regular health management of equipment is also an important measure to ensure its long-term stable operation.

[0003] However, although researchers have made some progress in equipment fault diagnosis, most of their work focuses on equipment fault analysis and lacks content on analyzing potential faults or problems of equipment, thus failing to achieve intelligent fault analysis of equipment throughout its entire life cycle.

[0004] Therefore, how to provide a device fault diagnosis system that can solve the above problems is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides a fault diagnosis system and method for equipment based on large model and knowledge graph collaborative reasoning. By constructing a fault diagnosis system for equipment based on large model and knowledge graph collaborative reasoning through hierarchical design, the system deeply integrates data processing, fault identification, intelligent analysis, prediction and early warning and life assessment, so as to realize the upgrade of equipment fault diagnosis from passive response to active prevention and from single data judgment to multi-dimensional collaborative reasoning.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A device fault diagnosis system based on large model and knowledge graph collaborative reasoning includes: The data layer is used to acquire real-time operating data of the device under test and to preprocess the real-time operating data. The knowledge layer, which is connected to the data layer, is used to analyze the preprocessed real-time working data to determine whether there is fault data. The analysis of the real-time working data includes constructing a knowledge graph of equipment faults and performing reasoning analysis through a graph neural network. When abnormal data is received, the association weight between the node corresponding to the abnormal data and each fault node is calculated through a graph attention network, and fault nodes with association weights higher than a preset threshold are selected as fault data output. The model layer, which is connected to the knowledge layer, is used to analyze the fault data when the analysis results output by the knowledge layer determine that fault data exists, in order to obtain the cause of the fault and the corresponding maintenance plan.

[0007] Preferred options also include: The application layer, which is connected to the knowledge layer and the model layer, is used to perform fault prediction on the preprocessed real-time working data when the analysis result output by the knowledge layer determines that there is no fault data, to obtain the equipment fault occurrence prediction result, the equipment fault occurrence probability and the corresponding actual occurrence probability threshold. When the equipment fault occurrence probability meets the actual occurrence probability threshold, the preprocessed real-time working data is input to the model layer for fault probability analysis.

[0008] Preferred options also include: The data analysis layer, which is connected to the application layer, is used to acquire equipment maintenance data and technical parameter data when the probability of equipment failure does not meet the actual probability threshold, and to predict the remaining lifespan of the equipment in combination with the real-time working data.

[0009] Preferred options also include: A digital twin simulation verification layer, connected to the knowledge layer, the model layer, and the application layer, is used to establish a three-dimensional simulation model of the device under test. When the analysis results output by the knowledge layer determine the existence of fault data, the fault data, the cause of the fault, and the repair plan are received, and the fault data, the cause of the fault, and the repair plan are input into the three-dimensional simulation model for simulation to verify the rationality of the cause of the fault and the repair plan. When the analysis results output by the knowledge layer do not show fault data, the preprocessed real-time working data is input into the three-dimensional simulation model for simulation to verify the possibility of the fault occurring.

[0010] This invention also provides a fault diagnosis method for devices based on collaborative reasoning using large models and knowledge graphs, comprising the following steps: S1: Obtain the real-time working data of the device under test through the data layer, and preprocess the real-time working data; S2: The preprocessed real-time working data is analyzed through the knowledge layer to determine whether there is faulty data; S3: When the analysis results output by the knowledge layer determine that there is fault data in the preprocessed real-time working data, the model layer is used to analyze the fault data to obtain the cause of the fault and the corresponding maintenance plan. S4: When there is no fault data in the preprocessed real-time working data, the real-time working data is analyzed, and fault prediction is performed on the preprocessed real-time working data through the application layer.

[0011] Preferred options also include: S5: Construct a three-dimensional simulation model of the device under test through a digital twin simulation verification layer. When the analysis results output by the knowledge layer determine that there is fault data, receive the fault data, the cause of the fault, and the repair plan, and input the fault data, the cause of the fault, and the repair plan into the three-dimensional simulation model for simulation to verify the rationality of the cause of the fault and the repair plan.

[0012] Preferably, S4 further includes: When the analysis results output by the knowledge layer determine that there is no fault data, the application layer performs fault prediction on the preprocessed real-time working data to obtain the equipment fault occurrence prediction result, the equipment fault occurrence probability and the corresponding actual occurrence probability threshold. When the equipment fault occurrence probability meets the actual occurrence probability threshold, the preprocessed real-time working data is input to the model layer for fault probability analysis.

[0013] Preferably, the implementation process of S2 includes: The preprocessed real-time working data is analyzed to determine whether there is faulty data in the preprocessed real-time working data. When faulty data exists, a knowledge graph is constructed, and reasoning analysis is performed using the knowledge graph to determine whether faulty data exists within the abnormal data. The specific implementation process of the knowledge graph includes: Knowledge graphs perform global reasoning on fault data, specifically expressed as: ; In the formula, This represents the reasoning score, where T represents the number of hops in the fault path. Let t represent the t-th fault data, sim() represent the similarity between the t-th fault data and the (t+1)-th fault data, and ATT() represent... Attribute confidence, Indicates the strength of the relationship; The knowledge graph outputs the probability of the eventual failure, specifically expressed as: ; In the formula, To determine the confidence level for the existence of fault f in a knowledge graph. This represents the score of the i-th faulty path, where M is the total number of candidate paths. Indicates the importance weight of the path; It can also dynamically update the knowledge graph, including incremental updates and weight adaptive updates. The specific expression for the incremental update is as follows: ; In the formula, λ represents the historical knowledge forgetting factor, and 0 < λ < 1. This indicates the strength of the relationship calculated from the newly added samples. This indicates the strength of the relationship calculated from historical samples; The specific expression for adaptive weight update is: ; In the formula, Indicates the distance attenuation coefficient. This represents the Gaussian kernel standard deviation. This represents the label consistency reward coefficient. This indicates the consistency between the fault labels and the output of the knowledge graph. This represents the weight of the relationship between node i and node j in the knowledge graph at time t.

[0014] Preferably, the implementation process of S4 includes: The Bayesian fault prediction model is used to predict faults in the preprocessed real-time working data. The specific expression is as follows: ; In the formula, This represents the parameter vector of the fault prediction model. Indicates a given historical observation data The posterior predicted probability of a failure occurring at time t. Indicates a fault event. This represents a sequence of working data.

[0015] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a device fault diagnosis system and method based on collaborative reasoning of large models and knowledge graphs, which has the following beneficial effects: 1. This invention is designed in a progressive and interconnected manner, consisting of a data layer, a knowledge layer, a model layer, an application layer, a data analysis layer, and a digital twin simulation verification layer. It relies on a knowledge graph to perform preliminary fault identification on preprocessed real-time working data. Utilizing the structured knowledge association capabilities of the knowledge graph, it quickly locates faulty data and eliminates invalid interference, achieving preliminary fault screening and judgment. When faulty data is confirmed, the deep reasoning capabilities of the large model are used to perform multi-dimensional analysis of the fault data. This not only accurately uncovers the root cause of the fault but also directly outputs corresponding repair solutions, improving fault handling efficiency. For scenarios where the knowledge layer determines there is no faulty data, the application layer further performs fault prediction on the real-time working data, outputting the fault occurrence prediction results, probabilities, and corresponding thresholds, enabling early detection of potential equipment faults. When the fault probability reaches the threshold, the data is automatically input into the model layer for in-depth analysis of fault probability, proactively identifying potential faults and improving equipment operational stability. Simultaneously, the system is integrated with the equipment's digital twin model to achieve interactive fault diagnosis between the virtual and physical worlds.

[0016] 2. The data analysis layer of this invention is designed for scenarios where the failure probability does not reach the threshold. It combines equipment maintenance data, technical parameter data, and real-time working data for multi-dimensional fusion analysis to achieve accurate prediction of the remaining lifespan of the equipment, improve the efficiency and service life of the equipment, and reduce the overall operation and maintenance costs of the equipment.

[0017] 3. This invention achieves automated data transmission and analysis throughout the entire process, from real-time working data acquisition and preprocessing at the data layer, to fault data determination at the knowledge layer, fault cause analysis and maintenance plan generation at the model layer, to fault prediction at the application layer, and remaining life prediction at the data analysis layer. This reduces subjective errors caused by manual operation and improves the efficiency of data processing and analysis. It is especially suitable for scenarios with multiple devices and high-frequency data monitoring, such as industrial production and large equipment operation and maintenance, thereby enhancing the practicality and universality of the system. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 This is a structural principle block diagram of the fault diagnosis system for equipment based on large model and knowledge graph collaborative reasoning provided by the present invention; Figure 2 The flowchart of the fault diagnosis system for equipment based on large model and knowledge graph collaborative reasoning provided by the present invention is shown.

[0020] Figure label: 1-Data layer; 2-Knowledge layer; 3-Model layer; 4-Application layer; 5-Data analysis layer; 6-Digital twin simulation verification layer. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] See Figure 1 As shown, this embodiment of the invention discloses a fault diagnosis system for devices based on large models and knowledge graph collaborative reasoning, including: Data layer 1 is used to acquire real-time working data of the device under test and to preprocess the real-time working data. Knowledge layer 2, which is connected to data layer 1, is used to analyze the preprocessed real-time working data to determine whether there is faulty data. Model layer 3, which is connected to knowledge layer 2, is used to analyze the fault data when the analysis results output by knowledge layer 2 determine that fault data exists, and to obtain the cause of the fault and the corresponding maintenance plan.

[0023] Specifically, the implementation process of data layer 1 includes: The real-time working data may include various types of data such as equipment operating parameter data, working environment data, component action status data, and energy consumption data. Preprocessing operations may include steps such as data deduplication, outlier removal, missing value completion, data standardization and normalization.

[0024] Specifically, the normalization expression in the preprocessing operation is as follows: ; In the formula, x represents the original real-time working data. , These represent the maximum and minimum values ​​corresponding to the original real-time working data, respectively. The specific expression for missing value completion is: ; In the formula, Let represent the i-th valid data point, and n represent the number of valid data samples. This represents the data after filling.

[0025] The specific implementation process of knowledge layer 2 includes: (1) Multiple abnormal thresholds for real-time working data are set in the knowledge layer 2. Based on the abnormal thresholds, the real-time working data is divided into working data that meets the threshold and abnormal data that does not meet the threshold. In the knowledge layer 2, a knowledge graph of equipment faults is constructed and reasoning analysis is performed through a graph neural network. Equipment components, fault modes, fault characteristics, and working environment are taken as entities, and the causal relationship, temporal relationship, and membership relationship between entities are taken as edges to construct a structured knowledge graph. The relevant information of equipment faults is extracted by feature extraction, and graph neural network algorithm and graph attention network algorithm are integrated to increase the connection between multiple factors and quickly and comprehensively find the relationship between fault entities and entities. Therefore, when abnormal data is received, the association weight between the nodes corresponding to the abnormal data is calculated by the graph attention network, and the fault nodes with association weights higher than the preset threshold are selected as fault data output. The specific process of reasoning analysis by the knowledge graph construction and analysis module through the graph attention network is as follows: First, the feature vector of each entity node in the knowledge graph is initialized. For any node and its neighboring nodes, the node features are concatenated after linear transformation and input into the LeakyReLU activation function to calculate the attention coefficient. Then, the attention coefficients are normalized using the softmax function to obtain normalized attention weights. These weights are used as aggregation weights to weight and sum the transformation features of each neighboring node, followed by non-linear activation, to update the feature representation of the current node. To enhance the model's expressive power, a multi-head attention mechanism is employed, setting multiple independent attention heads to perform the above calculations in parallel. The output features of each attention head are concatenated into an intermediate layer or an average output layer to obtain the final node feature representation. When abnormal data is received, the correlation weights between abnormal nodes and each faulty node are calculated, and faulty nodes with correlation weights higher than a preset threshold are filtered out and output.

[0026] (2) If both working data and abnormal data exist, then the clustering algorithm is used to filter out data that is close to the abnormal threshold from the working data and combine it with the abnormal data to form the final abnormal data; If only working data exists, then data close to the anomaly threshold is selected from the working data as anomaly data for subsequent analysis (e.g., when the difference between the working data and the anomaly threshold is within a certain fixed range); if no data close to the anomaly threshold exists, then the working data is output.

[0027] Specifically, the expression for filtering data close to the anomaly threshold from the working data using a clustering algorithm is as follows: ; In the formula, represents the Euclidean distance between data i and data j, represents the clustering degree, m represents the data dimension, and data with a distance less than the clustering threshold are considered faulty data.

[0028] The specific implementation process of model layer 3 includes: A fault diagnosis model and a fault repair solution decision model are constructed separately. When fault data exists, the fault data is input into the fault diagnosis model and the fault repair solution decision model for processing to obtain the corresponding fault cause and the corresponding repair solution. The fault diagnosis model can be implemented using a Bayesian fault model, and the fault repair solution decision model can be implemented using a combined model of support vector machine and decision tree. Furthermore, the above models can be optimized. The specific process is as follows: Historical equipment failure data and historical maintenance plan data are acquired through big data analytics. These data are then used to optimize the fault diagnosis model and the fault maintenance plan decision model. When fault data is available, it is input into the fault diagnosis model and fused with the corresponding entities in the knowledge graph for reasoning. Simultaneously, the fault data is input into the fault maintenance plan decision model for processing to obtain the corresponding fault cause and the corresponding maintenance plan.

[0029] The process of optimizing the fault diagnosis model and the fault repair plan decision model using the historical equipment fault data and historical maintenance plan data can be achieved through a relevant loss function, the specific expression of which is: ; In the formula, N represents the number of historical fault samples. Indicates the actual fault label. L represents the model's prediction result, and L represents the loss function.

[0030] In one specific embodiment, it also includes: Application layer 4, connected to knowledge layer 2 and model layer 3, is used to perform fault prediction on the preprocessed real-time working data when the analysis result output by knowledge layer 2 determines that no fault data exists. This yields a fault prediction result, a fault probability, and a corresponding actual fault probability threshold. When the fault probability meets the actual fault probability threshold, the preprocessed real-time working data is input to model layer 3 for fault probability analysis. The specific implementation process includes: A fault prediction decision model is constructed, and the preprocessed real-time working data is input into the fault prediction decision model for processing to obtain the corresponding equipment fault occurrence prediction result, equipment fault occurrence probability, and corresponding actual occurrence probability threshold.

[0031] In one specific embodiment, it also includes: Data analysis layer 5, connected to application layer 4, is used to acquire equipment maintenance data and technical parameter data when the probability of equipment failure does not meet the actual probability threshold, and to predict the remaining lifespan by combining the data with the real-time operating data. The specific implementation process includes: A remaining life prediction model is constructed. When the probability of equipment failure does not meet the actual probability threshold, the maintenance data, technical parameter data, and preprocessed real-time working data are input into the remaining life prediction model for processing to obtain the corresponding remaining life prediction result.

[0032] In one specific embodiment, it also includes: The digital twin simulation verification layer 6 is connected to the knowledge layer 2, the model layer 3, and the application layer 4. It is used to establish a three-dimensional simulation model of the device under test. When the analysis results output by the knowledge layer 2 determine that fault data exists, it receives the fault data, the cause of the fault, and the repair plan, and inputs the fault data, the cause of the fault, and the repair plan into the three-dimensional simulation model for simulation to verify the rationality of the cause of the fault and the repair plan. When the analysis results output by the knowledge layer 2 do not show fault data, it inputs the preprocessed real-time working data into the three-dimensional simulation model for simulation to verify the possibility of the fault occurring.

[0033] This invention also provides a diagnostic method for the equipment fault diagnosis system based on large model and knowledge graph collaborative reasoning, as described in any of the above embodiments. Figure 2 As shown, it includes the following steps: S1: Obtain the real-time working data of the device under test through data layer 1, and preprocess the real-time working data; S2: The preprocessed real-time working data is analyzed through knowledge layer 2 to determine whether there is fault data; S3: When the analysis results output by the knowledge layer 2 determine that there is fault data in the preprocessed real-time working data, the model layer 3 is used to analyze the fault data to obtain the cause of the fault and the corresponding maintenance plan. S4: When there is no fault data in the preprocessed real-time working data, the real-time working data is analyzed, and fault prediction is performed on the preprocessed real-time working data through application layer 4.

[0034] In one specific embodiment, it also includes: S5: Construct a three-dimensional simulation model of the device under test through the digital twin simulation verification layer 6. When the analysis results output by the knowledge layer 2 determine that there is fault data, receive the fault data, the cause of the fault, and the repair plan, and input the fault data, the cause of the fault, and the repair plan into the three-dimensional simulation model for simulation to verify the rationality of the cause of the fault and the repair plan.

[0035] In one specific embodiment, S4 further includes: When the analysis results output by the knowledge layer 2 determine that there is no fault data, the application layer 4 performs fault prediction on the preprocessed real-time working data to obtain the equipment fault occurrence prediction result, the equipment fault occurrence probability and the corresponding actual occurrence probability threshold. When the equipment fault occurrence probability meets the actual occurrence probability threshold, the preprocessed real-time working data is input to the model layer 3 for fault probability analysis.

[0036] In one specific embodiment, the implementation process of S2 includes: The preprocessed real-time working data is analyzed to determine whether there is faulty data in the preprocessed real-time working data. When faulty data exists, a knowledge graph is constructed, and reasoning analysis is performed using the knowledge graph to determine whether faulty data exists within the abnormal data. The specific implementation process of the knowledge graph includes: Knowledge graphs perform global reasoning on fault data, specifically expressed as: ; In the formula, This represents the reasoning score, where T represents the number of hops in the fault path. Let t represent the t-th fault data, sim() represent the similarity between the t-th fault data and the (t+1)-th fault data, and ATT() represent... Attribute confidence, Indicates the strength of the relationship; The knowledge graph outputs the probability of the eventual failure, specifically expressed as: ; In the formula, To determine the confidence level for the existence of fault f in a knowledge graph. This represents the score of the i-th faulty path, where M is the total number of candidate paths. Indicates the importance weight of the path; It can also dynamically update the knowledge graph, including incremental updates and weight adaptive updates. The specific expression for the incremental update is as follows: ; In the formula, λ represents the historical knowledge forgetting factor, and 0 < λ < 1. This indicates the strength of the relationship calculated from the newly added samples. This indicates the strength of the relationship calculated from historical samples; The specific expression for adaptive weight update is: ; In the formula, Indicates the distance attenuation coefficient. This represents the Gaussian kernel standard deviation. This represents the label consistency reward coefficient. This indicates the consistency between the fault labels and the output of the knowledge graph. This represents the weight of the relationship between node i and node j in the knowledge graph at time t.

[0037] In one specific embodiment, the implementation process of S4 includes: The Bayesian fault prediction model is used to predict faults in the preprocessed real-time working data. The specific expression is as follows: ; In the formula, This represents the parameter vector of the fault prediction model. Indicates a given historical observation data The posterior predicted probability of a failure occurring at time t. Indicates a fault event. This represents a sequence of working data.

[0038] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0039] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A fault diagnosis system for equipment based on large-scale model and knowledge graph collaborative reasoning, characterized in that: include: The data layer (1) is used to acquire the real-time working data of the device under test and to preprocess the real-time working data. Knowledge layer (2), which is connected to the data layer (1), is used to analyze the preprocessed real-time working data to determine whether there is fault data; The analysis of the real-time working data includes constructing a knowledge graph of equipment faults and performing reasoning analysis through a graph neural network; when abnormal data is received, the association weight between the node corresponding to the abnormal data and each fault node is calculated through a graph attention network, and fault nodes with association weights higher than a preset threshold are selected as fault data output. Model layer (3), which is connected to knowledge layer (2), is used to analyze the fault data when the analysis results output by knowledge layer (2) determine that there is fault data, and to obtain the fault cause and the corresponding maintenance plan; Application layer (4), which is connected to knowledge layer (2) and model layer (3), is used to perform fault prediction on the preprocessed real-time working data when the analysis result output by knowledge layer (2) determines that there is no fault data, to obtain equipment fault occurrence prediction result, equipment fault occurrence probability and corresponding actual occurrence probability threshold, and when the equipment fault occurrence probability meets the actual occurrence probability threshold, the preprocessed real-time working data is input to model layer (3) for fault probability analysis; The digital twin simulation verification layer (6) is connected to the knowledge layer (2), the model layer (3), and the application layer (4) to establish a three-dimensional simulation model of the device under test. When the analysis results output by the knowledge layer (2) determine that there is fault data, the fault data, the fault cause, and the repair plan are received, and the fault data, the fault cause, and the repair plan are input into the three-dimensional simulation model for simulation to verify the rationality of the fault cause and the repair plan. And: When the analysis results output by the knowledge layer (2) do not contain fault data, the preprocessed real-time working data is input into the three-dimensional simulation model for simulation operation to verify the possibility of fault occurrence.

2. The fault diagnosis system for equipment based on large model and knowledge graph collaborative reasoning according to claim 1, characterized in that, Also includes: The data analysis layer (5) is connected to the application layer (4) and is used to obtain the equipment maintenance data and technical indicator parameter data when the probability of the equipment failure does not meet the actual probability threshold, and to predict the remaining lifespan in combination with the real-time working data.

3. A diagnostic method for a device fault diagnosis system based on collaborative reasoning using large models and knowledge graphs according to any one of claims 1-2, characterized in that, Includes the following steps: S1: Obtain the real-time working data of the device under test through the data layer (1), and preprocess the real-time working data; S2: Analyze the preprocessed real-time working data through the knowledge layer (2) to determine whether there is fault data; S3: When the analysis results output by the knowledge layer (2) determine that there is fault data in the preprocessed real-time working data, the model layer (3) is used to analyze the fault data to obtain the fault cause and the corresponding maintenance plan. S4: When there is no fault data in the preprocessed real-time working data, the real-time working data is analyzed, and fault prediction is performed on the preprocessed real-time working data through the application layer (4).

4. The diagnostic method according to claim 3, characterized in that, Also includes: S5: Construct a three-dimensional simulation model of the device under test through the digital twin simulation verification layer (6). When the analysis results output by the knowledge layer (2) determine that there is fault data, receive the fault data, the fault cause and the repair plan, and input the fault data, the fault cause and the repair plan into the three-dimensional simulation model for simulation to verify the rationality of the fault cause and the repair plan.

5. The diagnostic method according to claim 3, characterized in that, S4 further includes: When the analysis results output by the knowledge layer (2) determine that there is no fault data, the application layer (4) performs fault prediction on the preprocessed real-time working data to obtain the equipment fault occurrence prediction result, the equipment fault occurrence probability and the corresponding actual occurrence probability threshold. When the equipment fault occurrence probability meets the actual occurrence probability threshold, the preprocessed real-time working data is input to the model layer (3) for fault probability analysis.

6. The diagnostic method according to claim 3, characterized in that, The implementation process of S2 includes: The preprocessed real-time working data is analyzed to determine whether there is faulty data in the preprocessed real-time working data. When faulty data exists, a knowledge graph is constructed, and reasoning analysis is performed using the knowledge graph to determine whether faulty data exists within the abnormal data. The specific implementation process of the knowledge graph includes: Knowledge graphs perform global reasoning on fault data, specifically expressed as: ； In the formula, This represents the reasoning score, where T represents the number of hops in the fault path. Let t represent the t-th fault data, sim() represent the similarity between the t-th fault data and the (t+1)-th fault data, and ATT() represent... Attribute confidence, Indicates the strength of the relationship; The knowledge graph outputs the probability of the eventual failure, specifically expressed as: ； In the formula, To determine the confidence level for the existence of fault f in a knowledge graph. This represents the score of the i-th faulty path, where M is the total number of candidate paths. Indicates the importance weight of the path; It can also dynamically update the knowledge graph, including incremental updates and weight adaptive updates. The specific expression for the incremental update is as follows: ； In the formula, λ represents the historical knowledge forgetting factor, and 0 < λ < 1. This indicates the strength of the relationship calculated from the newly added samples. This indicates the strength of the relationship calculated from historical samples; The specific expression for adaptive weight update is: ； In the formula, Indicates the distance attenuation coefficient. This represents the Gaussian kernel standard deviation. This represents the label consistency reward coefficient. This indicates the consistency between the fault labels and the output of the knowledge graph. This represents the weight of the relationship between node i and node j in the knowledge graph at time t.

7. The diagnostic method according to claim 3, characterized in that, The implementation process of S4 includes: The Bayesian fault prediction model is used to predict faults in the preprocessed real-time working data. The specific expression is as follows: ； In the formula, This represents the parameter vector of the fault prediction model. Indicates a given historical observation data The posterior predicted probability of a failure occurring at time t. Indicates a fault event. This represents a sequence of working data.

Images