An industrial fault diagnosis method and system with intelligent causal correction

Abstract

The present invention relates to the technical field of fault detection, and particularly to an industrial fault diagnosis method and system with intelligent causal correction. The method includes obtaining monitoring data and user task requirements; constructing task structure information based on the obtained user task requirements; processing the monitoring data through a three-level cascaded framework to generate metadata; constructing an initial causal graph structure by using the task structure information and the metadata; deeply optimizing the initial causal graph structure through a graph neural network to obtain a causal composite relationship graph; performing fault diagnosis and information retrieval by using the causal composite relationship graph in combination with a knowledge base; and generating a structured diagnostic report. Through the innovative structure integrating the LLM and the graph neural network, the system can dynamically optimize the causal relationship model, automatically learn and correct fault associations, enhance the interpretability of the causal graph by using the semantic reasoning of the LLM and the graph attention mechanism, and significantly improve the robustness and accuracy of diagnosis.

Description

Technical Field

[0001] The present invention relates to the technical field of fault detection, and more particularly to an industrial fault diagnosis method and system with intelligent causal correction. Background Art

[0002] With the rapid development of Industry 4.0 and artificial intelligence technologies, industrial fault diagnosis systems play a crucial role in enhancing equipment reliability and production efficiency, especially in the field of power equipment, such as the maintenance of transformers, generators, and transmission lines. However, fault diagnosis of power equipment faces significant challenges, mainly due to complex working environments, multiple fault modes, and data uncertainty. These problems include the ambiguity of fault causality, knowledge gaps (such as unknown variable correlations and potential fault chains), and complex interactions between multiple variables, resulting in reduced diagnostic accuracy and response delays. For example, in a power system, traditional methods may not be able to effectively capture the dynamic association of "voltage fluctuations causing temperature rise and then leading to insulation failure", especially when data variation or new fault modes occur.

[0003] Traditional fault diagnosis methods usually rely on predefined thresholds or static models, unable to dynamically adapt to changing environments, ignoring potential fault chains and cross - impacts, and thus performing poorly when dealing with unknown faults or data variation. This limitation not only reduces the interpretability and reliability of diagnosis but also increases maintenance costs and risks. Specifically, it is manifested in the following aspects:

[0004] Monitoring data of power equipment generally has noise interference, data loss, and abnormal fluctuations caused by harsh environments, seriously affecting data reliability; at the same time, key diagnostic information is scattered in multi - source heterogeneous data, and traditional methods lack effective mechanisms for deep fusion and high - quality processing, resulting in insufficient exploration of data value and limiting diagnostic accuracy and robustness;

[0005] When existing systems are based on statistical methods for causal relationship modeling, they lack robustness, are easily limited by data noise, outliers, and task information limitations, resulting in unstable or unreliable construction of causal graphs, and thus reducing the overall performance and anti - interference ability of fault diagnosis;

[0006] Traditional methods based on preset rules or single statistical models are difficult to dynamically capture the complex, non - linear, and time - lagged fault causal chains generated by power equipment in a high - dimensional and time - varying operating environment. When existing systems are based on neural network methods for causal relationship modeling, the interpretability is not high, and they cannot fully utilize the semantic reasoning of large language models (LLMs) and the deep learning capabilities of graph neural networks, resulting in incomplete and lack of transparency in causal graphs, affecting the reliability of diagnostic results and user confidence;

[0007] The power equipment fault diagnosis system lacks a real-time knowledge update and feedback optimization mechanism, and the knowledge base cannot be dynamically expanded and corrected according to new data or user feedback. As a result, the system has poor adaptability when facing new fault modes or environmental changes, the diagnostic results may be lagged or inaccurate, and a closed-loop optimization of continuous learning cannot be achieved.

[0008] Traditional diagnostic results are often simple fault labels or alarms, lacking clear and logical explanations for the causes, propagation paths, and potential impacts of faults. At the same time, the confidence level of diagnostic conclusions and the uncertainty of prediction results are not effectively quantified, reducing users' trust in diagnostic suggestions and decision-making efficiency. Summary of the Invention

[0009] To solve the above-mentioned problems, the present invention provides an industrial fault diagnosis method and system with intelligent causal correction. By integrating a large language model (LLM), a graph neural network (GNN), and an expert knowledge base, the automatic construction, dynamic optimization, and real-time correction of causal relationships are realized, thus significantly improving the robustness, interpretability, and adaptability of diagnosis.

[0010] In a first aspect, an industrial fault diagnosis method with intelligent causal correction provided by the present invention adopts the following technical solutions:

[0011] An industrial fault diagnosis method with intelligent causal correction includes:

[0012] Obtain monitoring data and user task requirements;

[0013] Construct task structure information based on the obtained user task requirements;

[0014] Process the monitoring data through a three-level cascaded framework to generate metadata;

[0015] Use the task structure information and metadata to construct an initial causal graph structure;

[0016] Deeply optimize the initial causal graph structure through a graph neural network to obtain a causal composite relationship graph;

[0017] Use the causal composite relationship graph and combine it with the knowledge base for fault diagnosis and information retrieval;

[0018] Generate a structured diagnostic report.

[0019] Furthermore, the constructing of task structure information based on the obtained user task requirements includes, for the current task T1, calculating the similarity with one or more historical tasks T2 in the historical task knowledge base, and through a hybrid similarity mechanism to comprehensively consider the similarity of the semantic content of the task and the set of structured variables, expressed as:

[0020]

[0021] Among them, emb ( T ) represents the key entity embedding vector of the task T . var ( T ) represents the set of standardized monitoring variables in the task T . α 1 represents the weight parameter, Cosine represents the cosine similarity between vectors, reflecting the proximity of the task in the semantic space, Jaccard represents the proportion of shared variables between sets, reflecting the degree of structural overlap of the task in key variables.

[0022] Furthermore, the generation of metadata by processing the monitoring data through a three - level cascaded framework includes first performing feature normalization on the monitoring data using the Robust Scaling method, scaling based on the quartiles and median of the data, and for the missing regions in the time - series feature matrix, interpolating using the weighted average of several latest valid observations before the missing points; then estimating the noise level in the monitoring data through the median absolute deviation, and performing multi - scale decomposition using wavelet transform on the high - frequency components detected based on the estimated noise level, decomposing the data sequence into low - frequency approximation coefficients and high - frequency detail coefficients; for low - frequency noise, non - Gaussian distributed noise or structural noise, suppressing it using a GAN - based denoising model.

[0023] Furthermore, the construction of the initial causal graph structure using the task structure information and metadata includes first constructing a causal graph using the Granger causality test algorithm G Granger , generating a directed graph by analyzing the time - series prediction relationship between variables, expressed as:

[0024]

[0025] Among them, RSS resticted is the sum of squared residuals fitted using only the historical data of B , RSS unresticted is the sum of squared residuals fitted using the historical data of i and j , k represents the lag order, dynamically selected based on the autocorrelation function analysis, p represents the autoregressive order, and the algorithm assumes that if the past values of variable j can significantly improve the prediction of variable i , that is, F is higher than the set threshold, then there exists j → iCausal relationship

[0026] Furthermore, when constructing the initial causal graph structure by using task structure information and metadata, it further includes using the PC algorithm based on conditional independence test to assume that there may be potential causal relationships between all variables, and gradually increasing the conditional set starting from a fully connected graph k , testing the conditional independence of variables i and j ; by iteratively increasing k, the algorithm gradually removes redundant edges to construct a sparse undirected skeleton graph, and infers the edge direction in combination with the Meek rule to construct a directed graph GPC. Among them, for the initial conditional set k, the correlation coefficient of each variable pair (i, j) is calculated r ij∣k , and its conditional independence score is tested Z :

[0027]

[0028] Among them, r ij∣k is the partial correlation coefficient between variables i and j under the given conditional set k , n is the sample size. The larger the absolute value of the z value, the stronger the correlation, thus rejecting the conditional independence hypothesis

[0029] Furthermore, when constructing the initial causal graph structure by using task structure information and metadata, it further includes using the GES algorithm to assume that there is no direct causal connection between variables, and starting from an empty graph, adopting a greedy search strategy to construct a graph structure G GES , adding edges through forward search and removing edges through backward search to maximize a global scoring function. Among them, starting from an empty graph, calculate the scoring gain of all edges, add the edge with the highest score in the forward stage until no operation can optimize the scoring function, and remove redundant edges in the backward stage to further optimize the scoring function to ensure that the local optimal solution approximates the global optimal solution, expressed as:

[0030]

[0031] Among them, is the probability of variable i given its parent node , | E | is the number of edges, λ is the regularization parameter

[0032] Further, the step of constructing the initial causal graph structure using the task structure information and metadata further includes preliminarily optimizing the four candidate causal graphs using an LLM and an industrial knowledge graph. Specifically, a high-dimensional text embedding model, BERT, is used to map each variable of interest into a high-dimensional continuous vector space in the knowledge prior emb , which serves as the basic input for the LLM to perform semantic evaluation. Then, the metadata vector features X total are combined with emb to generate an embedding vector X . The combined confidence is calculated by integrating multi-source information confidence: multiple candidate causal graphs optimized by the LLM are fused to generate a preliminary comprehensive causal graph G. The fusion process is based on a Bayesian network framework, incorporating data-driven confidence, LLM semantic confidence, and domain knowledge prior to ensure the connectivity and acyclicity of the graph structure.

[0033] Further, the step of deeply optimizing the initial causal graph structure through a graph neural network to obtain a causal composite relationship graph includes using the preliminarily constructed causal graph and hybrid features as inputs, and successively implementing graph embedding representation, edge weight optimization, and causal path enhancement based on the graph neural network. An LLM is used to perform semantic completion and inference correction on potential causal structures. Specifically, the optimized graph of the current causal graph and all relevant historical causal graphs in the system-maintained historical causal graph library are used to generate graph-level global node embeddings and edge embeddings, and an asymmetric graph similarity evaluation function S ( G opt , G hist ) is established to quantify the structural and semantic dual similarity between the current optimized graph G opt and the historical causal graph G hist , which is expressed as:

[0034]

[0035] where represents the Jaccard similarity of the node set, represents the normalized Hamming distance of the edge set, and represents the cosine similarity of the core topological feature vectors of the graph spectrum.

[0036] Further, the step of deeply optimizing the initial causal graph structure through a graph neural network to obtain a causal composite relationship graph further includes integrating the LLM into the training loop of the GNN to form an interactive feedback mechanism. After the GNN is trained for several rounds, the current graph structure and node and edge representations are fed back to the LLM. The LLM combines the semantic verification process to evaluate the rationality of the edges or paths and outputs a correction signal. The formula is updated by the joint loss:

[0037]

[0038] Among them, L GNN represents the GNN prediction loss, represents the weight parameter, L LLM_feedback represents the supervised loss generated by the LLM, and the formula is as follows:

[0039]

[0040] Among them, C LLM ( e ) represents the edge e LLM semantic confidence of, p ( e ) represents the GNN predicted edge probability, entropy ( G ) represents the graph entropy, represents the weight parameter.

[0041] Furthermore, the use of the causal composite relationship graph and the knowledge base for fault diagnosis and information retrieval includes retrieving the historical fault instance most similar to the current diagnosis result in the local knowledge base through keyword extraction and graph matching technology, calculating the similarity using the joint criterion of semantic vector embedding and causal structure matching. When the similarity is higher than the preset threshold, the system extracts the key information of the matching case and generates a natural language diagnosis report. If the retrieval fails, relevant entries are extracted using uncertainty sampling and the LLM generates structured metadata, which is incrementally integrated into the local knowledge base and fed back to the causal graph to expand the coverage range of the causal graph, recalculating the similarity and generating the report. The similarity is expressed as:

[0042]

[0043] Among them, R is the current diagnosis result, C is the historical fault case, Sim emb ( R , C ) represents the semantic similarity of the fault description, Sim topo ( R , C ) represents the structural similarity of the fault causal link, represents the weight parameter.

[0044] In the second aspect, an industrial fault diagnosis system with intelligent causal correction includes:

[0045] A data acquisition module, configured to acquire monitoring data and user task requirements;

[0046] A task structure module, configured to construct task structure information based on the acquired user task requirements;

[0047] A metadata module, configured to generate metadata by processing monitoring data through a three - level cascaded framework;

[0048] A causal graph module, configured to construct an initial causal graph structure by using the task structure information and metadata;

[0049] A deep optimization module, configured to deeply optimize the initial causal graph structure through a graph neural network to obtain a causal composite relationship graph;

[0050] A fault diagnosis module, configured to perform fault diagnosis and information retrieval by using the causal composite relationship graph and combining with a knowledge base;

[0051] A report module, configured to generate a structured diagnostic report.

[0052] In a third aspect, the present invention provides a computer - readable storage medium, in which multiple instructions are stored, and the instructions are adapted to be loaded and executed by a processor of a terminal device for the industrial fault diagnosis method with intelligent causal correction described above.

[0053] In a fourth aspect, the present invention provides a terminal device, including a processor and a computer - readable storage medium. The processor is used to implement each instruction; the computer - readable storage medium is used to store multiple instructions, and the instructions are adapted to be loaded and executed by the processor for the industrial fault diagnosis method with intelligent causal correction described above.

[0054] In summary, the present invention has the following beneficial technical effects:

[0055] Compared with the prior art, the industrial fault diagnosis system with intelligent causal correction of the present application has the following beneficial effects: Through the innovative structure integrating the LLM and the graph neural network, the system can dynamically optimize the causal relationship model, automatically learn and correct fault associations, enhance the interpretability of the causal graph by using the semantic reasoning of the LLM and the graph attention mechanism, and significantly improve the robustness and accuracy of diagnosis; In addition, the closed - loop optimization mechanism of the system realizes real - time knowledge update and task scheduling, flexibly responds to multi - variable interactions and dynamic environmental changes, improves the diagnosis efficiency and reliability, and provides a solid foundation for subsequent decision - making, maintenance and optimization of power equipment fault diagnosis.

[0056] In view of the limitations of traditional causal graph construction methods, this system further uses large language models to perform semantic reasoning and knowledge completion, and combines graph neural networks for feature extraction and fault correlation mining, enabling the causal graph to more comprehensively adapt to multivariable interactions and dynamic fault scenarios. Based on the enhanced causal graph, the system generates structured diagnostic reports and supports real-time knowledge updates and decision assistance, realizing the intelligence and closed-loop optimization of power equipment fault diagnosis. Brief Description of the Drawings

[0057] Figure 1 is a schematic diagram of an intelligent causal correction industrial fault diagnosis method according to Embodiment 1 of the present invention;

[0058] Figure 2 is an example diagram of a causal graph according to Embodiment 1 of the present invention. Detailed Description of the Embodiments

[0059] The present invention will be further described in detail below with reference to the accompanying drawings.

[0060] Embodiment 1

[0061] Referring to Figure 1 , an intelligent causal correction industrial fault diagnosis method of this embodiment includes:

[0062] Obtain monitoring data and user task requirements;

[0063] Construct task structure information based on the obtained user task requirements;

[0064] Process the monitoring data through a three-level cascaded framework to generate metadata;

[0065] Construct an initial causal graph structure using the task structure information and metadata;

[0066] Deeply optimize the initial causal graph structure through a graph neural network to obtain a causal composite relationship graph;

[0067] Use the causal composite relationship graph and combine it with the knowledge base for fault diagnosis and information retrieval;

[0068] Generate a structured diagnostic report.

[0069] Specifically:

[0070] Step S1, Task Definition:

[0071] For the task information input by the user in natural language form, the system parses the input text through the Prompt Engineering strategy, constructs three specific prompt templates: diagnostic objectives, variables of interest, and output requirements, identifies and extracts the key entities in the text, and maps the extracted entity names and descriptions to a predefined standardized term set and variable list. After being processed by the natural language processing and structured engine, the unstructured task information input by the user is transformed into preliminary structured data and stored in the task information file.

[0072] The system maintains a historical task knowledge base T which stores the previously successfully processed and structured task definition files and their related processing configurations, intermediate results, or even the optimized processing flow parameters. For the current task T1, the system calculates its similarity with one or more historical tasks T2 in the historical task knowledge base, using a hybrid similarity mechanism to comprehensively consider the similarity of the semantic content and the set of structured variables of the tasks:

[0073]

[0074] where emb ( T ) represents the key entity embedding vector of T , var ( T ) represents the set of standardized monitoring variables in task T , α 1 represents the weight parameter, Cosine represents the cosine similarity between vectors, reflecting the proximity of tasks in the semantic space, Jaccard represents the proportion of shared variables between sets, reflecting the degree of structural overlap of key variables between tasks.

[0075] If the system detects that there is one or more historical tasks T 2 , whose hybrid similarity with T 1 is higher than the preset similarity threshold of 0.95, then the system determines that T 1 shares variable dependencies and core diagnostic patterns with T 2 . Secondly, the system triggers the historical resource loading mechanism and directly loads the processing configurations and model parameters related to T 2 , specifically including the preliminary causal graph structure skeleton constructed by S3 and the GNN model weights trained in step S4.

[0076] Step S2, Metadata Generation:

[0077] The core function of step S2 is to construct a highly robust and high-quality time series feature matrix, which serves as a reliable input for subsequent causal modeling and fault diagnosis modules. The system adopts a three-level cascaded framework to process the data of monitoring devices, covering three sub-modules: unified data format and completion, adaptive noise suppression, and data trend prediction.

[0078] Among them, the unified format and completion sub-module reads multi-source heterogeneous data through an automated script of the pandas library, maps it to a standard CSV structure, and adds a timestamp field in the UTC standard format to achieve time alignment. A two-dimensional feature matrix is constructed with the timestamp field as the primary index X , and the dimension of the monitoring variable is set to d , and the number of sampling points is n .

[0079] Considering the order of magnitude differences in industrial data due to different physical units, dimensions, or measurement ranges, and the possible existence of extreme values or outliers, the Robust Scaling method is used to perform feature normalization processing, scaling based on the quartiles and median of the data:

[0080]

[0081] Among them, median ( X ) represents the median of the feature, IQR ( X ) = Q 3 ( X ) - Q 1 ( X ) represents the interquartile range, Q 3 ( X ) represents the first quartile, Q 1 ( X ) represents the third quartile. The center of the data is shifted to 0 and scaled according to its interquartile range, making the processed data distribution more concentrated and maintaining the consistency of the monitoring variables in the structured data matrix X on the numerical scale.

[0082] For the missing regions commonly existing in the time series feature matrix X due to reasons such as transmission interruption and sensor failure, following the principle of inherent time continuity and local correlation of industrial time series data, interpolation is performed using the weighted average of several latest valid observed values before the missing point, and the weight decays exponentially with time:

[0083]

[0084] Among them, Control the attenuation of historical weights, x t Indicates the missing points to be interpolated, x t-k At time t - k The effective observed value, K Is the selected historical window length. The parameter α 2 Is determined according to the autocorrelation characteristics of the data to retain the local data fluctuations and trends within a short time.

[0085] As a further implementation,

[0086] The adaptive noise suppression sub-module in the present invention aims to handle the significant noise problems caused by high-voltage environmental electromagnetic interference and sensor accuracy limitations in power equipment monitoring data. These noise characteristics usually manifest as high-frequency pulse interference, low-frequency drift, or non-Gaussian distributed noise, which are easy to mask the true signal characteristics and amplify the risk of false anomalies. The sub-module first estimates the noise level σ By median absolute deviation:

[0087]

[0088] Where, x i Represents the i th sampling point in the data sequence, median Represents the median operation. Based on the high-frequency components detected by the estimated noise level, wavelet transform is used for multi-scale decomposition, and the data sequence is decomposed into low-frequency approximation coefficients and high-frequency detail coefficients. The decomposition process can be expressed as:

[0089]

[0090] Where, Represents the wavelet coefficient, s > 0 Represents the scale parameter, which is used to control the fineness of the decomposition. A smaller s corresponds to high-frequency detail coefficients, and a larger s corresponds to low-frequency approximation coefficients. Represents the time translation parameter, Represents the conjugate of the wavelet function. The low-frequency approximation coefficients remain unchanged, and soft threshold filtering is applied to the high-frequency detail coefficients to adaptively remove the noise components:

[0091]

[0092] Where, λ Based on the noise standard deviation σ Calculated. The inverse wavelet transform is performed on the filtered wavelet coefficients to reconstruct the signal suppressing high-frequency noise:

[0093]

[0094] Among them, C ψ is a constant of the wavelet function, and the reconstructed signal f denoised ( t ) is used as the denoising output to replace the original signal sequence to reduce high-frequency noise interference.

[0095] For low-frequency noise, non-Gaussian distributed noise or structural noise that is difficult to effectively process by wavelet transform, the system adopts a GAN-based denoising model. The generator of GAN receives the noisy data X noisy and attempts to generate a denoising output X clean that approximates the original clean data :

[0096]

[0097] The discriminator D is trained to distinguish between the real clean data X clean and the denoised data generated by the generator . The training objective function is the KL divergence between and X clean . The system optimizes the model on a historical dataset of power equipment containing typical noise patterns, and the training objective function adopts the standard minimax adversarial loss:

[0098]

[0099] Among them, P data represents the clean data distribution, and P noisy is the noisy data distribution. The optimized generator regenerates the denoised data X denoised for the input noisy data, and combines it with the wavelet transform output f denoised ( t ) for fusion, and finally outputs a signal X with comprehensive suppression.

[0100] As a further implementation method,

[0101] The data trend prediction module addresses the problem of delayed diagnosis caused by strong periodicity and mutation characteristics (such as load fluctuations, seasonal electricity consumption patterns, or fault precursor signals) in power equipment monitoring data. It adopts a hybrid prediction framework of ARIMA and LSTM networks to achieve adaptive prediction of future data trends, improving the prediction accuracy and the reliability of anomaly detection.

[0102] Specifically, the ARIMA model captures linear trends and seasonality through parameters θ =( p , d , q ), where p represents the autoregressive order, d represents the differencing order used to eliminate non-stationarity, q represents the moving average order, and the selection of parameter θ is based on information:

[0103]

[0104] where, θ =( p , d , q ) is the parameter vector, p ( x t ∣ x t -1, θ ) is the conditional probability density given historical data and parameters. Feature prediction is performed based on the obtained parameters:

[0105]

[0106] where, and θ j are the autoregressive and moving average coefficients respectively, is the white noise residual. Parameter estimation uses maximum likelihood estimation to fit historical data through numerical optimization, generating a preliminary linear prediction sequence. To address the limitations of ARIMA in dealing with non-linear mutation events, an LSTM network is introduced for supplementation, and the outputs of ARIMA and LSTM are combined using a weighted fusion framework:

[0107]

[0108] where, horizon represents the prediction step, w ARIMA and w LSTM are the adaptive weights, satisfying w ARIMA + wLSTM = 1, calculate the prediction errors of ARIMA and LSTM on historical data to dynamically update the weights, ensuring that the ARIMA weight is higher during the stable period and the LSTM weight increases during mutation events. Generate prediction data X pred and confidence level P , and combine with the original detection data X to form the total input X total = X , X pred .

[0109] Upload the structured task file and X total to the system. For the key objectives in the task file, the system uses a fuzzy matching algorithm to perform field matching and consistency verification on the X total set of column names. All successfully matched data columns are bound to the task information to generate a complete structured task metadata structure.

[0110] Step S3, Causal graph construction:

[0111] The said step S3 includes three components: a data-driven causal hypothesis generation component, an LLM semantic verification and preliminary correction component, and a multi-source information integration component, aiming to infer the causal dependencies between variables from power equipment monitoring data to support subsequent fault diagnosis and prediction. The overall process starts with purely statistical-driven hypothesis generation and gradually incorporates semantic and domain knowledge to achieve the evolution from candidate graphs to robust comprehensive causal graphs.

[0112] Among them, the data-driven causal hypothesis generation component, based on the set of variables of interest in the metadata, uses four complementary statistical causal structure learning algorithms to batch test all variable pairs and generate diverse candidate causal graphs.

[0113] First, use the Granger causality test algorithm to construct a causal graph G Granger , and generate a directed graph by analyzing the time series prediction relationship between variables:

[0114]

[0115] Among them, RSS resticted is the sum of squared residuals fitted using only the B historical data, RSS unresticted is the sum of squared residuals fitted using the i and j historical data, kdenotes the lag order, dynamically selected based on autocorrelation function analysis, p denotes the autoregressive order. The algorithm assumes that if the past values of variable j can significantly improve the prediction of variable i , that is F the value is higher than the set threshold, then there exists j → i causal relationship.

[0116] Using the PC algorithm based on conditional independence test, it is assumed that there may be potential causal relationships between all variables. Starting from a fully connected graph, the conditional set k is gradually increased, and the conditional independence of variables i and j is tested. For the initial conditional set k, the correlation coefficient r ij∣k of each variable pair (i, j) is calculated Z and its conditional independence score

[0117]

[0118] is tested: r ij∣k where i and j is the partial correlation coefficient of variables k under the given conditional set n is the sample size. The larger the absolute value of the z-value, the stronger the correlation, thus rejecting the conditional independence hypothesis. By iteratively increasing k , the algorithm gradually removes redundant edges (edges with z-values below the threshold) and constructs a sparse undirected skeleton graph, and infers the edge directions using Meek's rules to construct a directed graph G PC .

[0119] Using the GES algorithm, it is assumed that there is no direct causal connection between variables. Starting from an empty graph, a greedy search strategy is adopted to construct the graph structure G GES , adding edges through forward search and removing edges through backward search to maximize a global scoring function:

[0120]

[0121] where is the probability of variable i given its parent nodes , | E | is the number of edges, λis the regularization parameter. The algorithm starts from an empty graph, calculates the scoring gain for all edges, adds the edge with the highest score in the forward phase until no operation can optimize the scoring function, and removes redundant edges in the backward phase to further optimize the scoring function, ensuring that the local optimal solution approximates the global optimal solution.

[0122] Regarding the lag effect and autocorrelation characteristics of power time series data, the PCMCI+ algorithm is used to construct a graph structure starting from a fully connected directed time series graph. For the current variable x , calculate the mutual information of relevant variables in the set of variables of interest:

[0123]

[0124] where y represents a single variable in the subset of variables composed of past time steps y t-i and the current time step y t , y t or y t-i , p ( x , y ) is the joint probability distribution, p ( x ) and p ( y ) are marginal probability distributions. If the mutual information I is below the threshold, it does not conform to the causal hypothesis, and non-causal edges are removed. Finally, the constructed graph G PCMCI is a sparse directed graph, with nodes including variable and time information, and edges containing time delay information.

[0125] However, pure statistical methods may be limited by the limitations of task information and cannot fully explore the deep causal relationships between variables or capture implicit physical mechanisms. Further, the LLM semantic verification and preliminary correction component introduces LLM and industrial knowledge graphs to optimize the four candidate causal graphs. Domain knowledge priors and LLM are used for semantic reasoning and probability evaluation to perform semantic rationality verification, conflict identification, and structural adjustment on the graph edges.

[0126] To enable the LLM to understand and operate on the variables of interest and their domain knowledge in its internal semantic space, the system first uses the high-dimensional text embedding model BERT pre-trained specifically for the industrial domain or general domain to map the detailed domain knowledge description text of each variable of interest in the knowledge prior to a high-dimensional continuous vector space emb , as the basic input for the LLM to perform semantic evaluation ( S semantic function). The metadata vector featuresX total Combine with emb to generate an embedding vector X For each candidate edge in the candidate graph, combine multi-source information confidence to calculate the comprehensive confidence:

[0127]

[0128] Among them, e represents the edge that actually exists in the candidate graph, x and y respectively represent the starting point and the ending point of the edge e , C LLM ( e ) represents the comprehensive confidence of the edge e , α 3 , α 4 , α 5 are adjustable weight parameters. S semantic ( x , y ) represents the semantic similarity and represents the node confidence in the semantic space. The calculation formula is as follows:

[0129]

[0130] Among them, emb x and emb y respectively represent the BERT embedding vectors of the nodes x and y , that is, the semantic representation, to quantify the knowledge correlation degree of the variables x and y in the knowledge graph, S semantic ( x , y ) The larger the value, the higher the semantic relevance. Attention ( x , y ) represents the Transformer-based attention and represents the node confidence in the LLM inference space:

[0131]

[0132] Among them, d k is the vector dimension, and the attention weight is calculated by the dot product of the nodes x and y . By calculating the variables x andy The attention weights between the semantic embedding vectors are used to reflect the association strength under a more complex non - linear mapping. R counterfactual Adopt the embedding vector X Evaluate the causal relationship strength, representing the causal counterfactual rationality evaluation score:

[0133]

[0134] Among them, do( X = x ) represents the causal intervention operation, E Y |do( X = x )] is to force X to take the value of x the expected value after the intervention. Distinguish correlation and causal relationship through causal intervention, so as to identify the true causal path. Combine the Y calculated by the LLM C LLM ( e ) with the original weights of the candidate graph to generate an optimized candidate graph. Specifically, consider three cases:

[0135] ① If the C LLM ( e ) calculated by the LLM is in the same direction as that in the candidate graph, enhance the confidence by adding the weights, and limit the final weight value in [0, 1].

[0136] ② If there is x → y in the candidate graph, but C LLM ( e ) infers y → x , the new weight is calculated by the following formula:

[0137]

[0138] Among them, γ is the attenuation factor, and if the direction conflict is reduced by subtraction.

[0139] ③ If C LLM ( e ) > θ but there is no edge e in the candidate graph, then directly add a new edge, and the initial weight W new ( e ) = C LLM ( e ​), and generate new nodes through counterfactual reasoning.

[0140] As a further implementation,

[0141] The multi-source information integration component fuses multiple candidate causal graphs optimized by the LLM to generate a preliminary comprehensive causal graph G. The fusion process is based on the Bayesian network framework, combining data-driven confidence, LLM semantic confidence, and domain knowledge prior to ensure the connectivity and acyclicity of the graph structure.

[0142] First, construct a set containing all potential causal edges, including the existing edges in all candidate causal graphs ( G Granger ,G PC , G GES ,G PCMCI+ ). For any edge e , its final confidence P ( e |all) is calculated by combining the evidence probabilities from different sources:

[0143]

[0144] where, P ( Data | e ) is the edge existence probability based on the data-driven algorithm, calculated based on feature similarity, P ( LLM | e ) is the confidence obtained based on the LLM semantic verification, and the value is derived from C LLM ( e ), P ( Knowledge | e ) is the probability of edge existence determined based on the domain knowledge graph, obtained through Bayesian inference. By calculating the probability values for all existing edges and normalizing, the final comprehensive confidence P ( e | Data, LLM, Knowledge ) ∈ [0, 1] is obtained.

[0145] Perform directed cycle detection on the fused graph structure using depth-first search (DFS). If a directed cycle is detected, the system identifies all the edges that form the cycle and removes the edge with the lowest combined confidence in the cycle. Repeat the process of detecting and removing the lowest-confidence edge until the graph no longer contains any directed cycles. The optimized causal graph G, as a directed acyclic graph, has nodes as the variables of interest and edges with combined confidence, providing prior information for the graph neural network training in step S4.

[0146] Step S4: Optimization of the causal graph based on graph neural network:

[0147] To support the deep learning and modeling of causal relationships by the GNN, the system constructs a hybrid feature representation of the nodes H , consisting of temporal features X time , semantic features X semantic , edge features E and structural features X structure . Temporal features X time are used to capture the dynamic statistical metrics of variables, such as mean, variance, rate of change, autocorrelation, and historical volatility patterns; semantic features X semantic are generated based on encoding the variable names, descriptions, and related knowledge entries using LLM and BERT models, with dimensions ranging from 128 to 512; edge features E include the type of the edge and the initial confidence obtained through LLM semantic verification in S4 C LLM ( e ); structural features X structure are calculated based on the topological properties of the causal graph G output in S4, and its key metrics include the degree of the node ki , betweenness centrality CB ( v ). Betweenness centrality CB ( v ) measures the bridging role of a node in the shortest paths, that is, how many shortest paths pass through the node. The calculation formula is as follows:

[0148]

[0149] where σ st represents the total number of shortest paths of the node pair ( s , t ), and σ st ( v ) represents the number of shortest paths passing through the node vThe number of shortest paths. Hybrid features H Through attention-weighted fusion:

[0150]

[0151] Wherein, Represents the weight based on feature correlation.

[0152] As a further implementation,

[0153] The causal graph enhancement and optimization training module in the present invention includes three parts: a causal graph embedding layer, a graph structure optimization layer, and a causal relationship reconstruction layer. This module takes the preliminarily constructed causal graph and hybrid features as inputs, realizes graph embedding representation, edge weight optimization, and causal path enhancement based on graph neural networks, and uses the LLM to perform semantic completion and inference correction on potential causal structures.

[0154] Among them, the causal graph embedding layer receives the causal graph G =( V , E ) and its corresponding node feature matrix X ∈ R n*d , and first performs self-looping and normalization processing on the causal graph to obtain a normalized adjacency matrix:

[0155]

[0156] Wherein, A Is the original adjacency matrix, I Is the identity matrix, D Is the node degree matrix, α 6 Self-loop adjustment factor. Introduce a multi-scale receptive field aggregation mechanism:

[0157]

[0158] Wherein, W (l) Is the trainable weight matrix of the l th layer, ReLU Represents the activation function, FFN Represents the feed-forward neural network, || Represents vector concatenation, k Represents the hop distance on the graph, K Is the maximum considered hop count to cover the multi-level cascading fault propagation range in the power system, Represents the node i 's neighbor set, α ij Represents the attention weight, calculated through a learnable power flow sensitivity matrix:

[0159]

[0160] Among them, P ij It is shown that the physical prior term derived from the power flow sensitivity matrix is to retain the initial feature information, and a residual connection mechanism is introduced:

[0161]

[0162] Among them, represents the node representation after applying the residual connection, represents the node-specific residual weight factor, represents the residual transformation matrix, l represents the number of model layers. Each layer of the model performs multi-scale receptive field aggregation and residual connection. Through l layers of the model, the node features are obtained. The edge weights are adjusted according to the embedding similarity of the two end nodes of each edge to improve the rationality and distinguishability of the causal structure. Let the node pair ( i , j ) have embedding vectors h i , h j A high-order similarity function is used to define the edge weights:

[0163]

[0164] Among them, represents the estimation of the feature covariance matrix, represents the mean vector of all node embeddings. Combining the known causal paths annotated in the training set, the edge weights are learned and updated through a weighted edge classification loss function:

[0165]

[0166] Among them, y ij ∈ {0, 1} is the edge label, and the edge label comes from the semantic reasoning of the domain knowledge graph and the LLM. The GNN iteratively updates the edge weights, adds or deletes edges according to w ij and outputs the optimized causal graph G opt .

[0167] As a further implementation,

[0168] The causal relationship reconstruction layer aims to go beyond the limitations of a single causal graph through cross-graph analysis, deeply analyze the association patterns among multiple causal graphs accumulated historically and related to the current diagnosis task, and construct a similarity network among causal graphs to mine composite causal links across time or scenarios. To compare and associate different causal graphs in a unified vector space, the current optimized graph G opt and all relevant historical causal graphs in the historical causal graph library maintained by the system G hist generate global node embeddings and edge embeddings at the graph level:

[0169]

[0170] Establish an asymmetric graph similarity evaluation function S ( G opt , G hist ), which is used to quantify the dual structural and semantic similarity between the current optimized graph G opt and the historical causal graph G hist :

[0171]

[0172] Among them, represents the Jaccard similarity of the node set, represents the normalized Hamming distance of the edge set, represents the cosine similarity of the core topological feature vectors of the graph spectrum. By setting an adaptive threshold τ sim (G opt ):

[0173]

[0174] Among them, density ( G opt ) represents the edge density of the current graph, μ and η are adjustable parameters. Screen high-similarity historical graphs to form a knowledge transfer source set G sim ={ G hist ∣S( G opt , G hist )>τ sim (G opt ).

[0175] For Gsim Each figure in G hist , standardize the matching of all variables with the set of variables of interest in the current figure G opt , and analyze the consistency of the antecedent causal chains of the matching nodes to decide whether to migrate and update the structure of Gopt. The antecedent causal chain is defined as a sequence of paths recursively extended in the in-degree direction from the target variable. For each matching variable node v, construct the antecedent causal chain on both graphs and traverse recursively as follows: start from the current variable, extend along the in-degree edge, and if the weight of the edge is greater than the threshold , include the node and repeat until the weight is below the threshold or the maximum depth of 3 is reached to prevent infinite recursion:

[0176]

[0177] where, represents the union operation, Chainpre(v) represents the antecedent causal chain of node v , represents the set of nodes of the antecedent causal chain of node v for node u that meets the requirements. If the chain sequences are exactly the same (i.e., the number of nodes and the edge directions are the same), then perform weighted average of the edge weights; if they are inconsistent, prefer the longer chain and directly add and update the non-repeating part to G opt and G hist corresponding parts, and the confidence of the newly added edge is obtained based on the mutual information of historical data.

[0178] On this basis, the system performs multi-graph collaborative inference, optimizes through the following formula to achieve unified integration of cross-graph knowledge:

[0179]

[0180] where, L align is the inter-graph alignment loss based on KL divergence, is the trade-off parameter. Based on the constructed causal graph similarity network, the causal relationship reconstruction layer performs a data-driven causal knowledge transfer and multi-graph collaborative inference process. Different from inferring only relying on the local or global information of the current graph, this method regards the current graph as a node in the network and uses the causal graph with high similarity to it as an external knowledge source to assist in the reconstruction and optimization of the current graph and construct a "composite causal chain" to identify multi-hop paths such as "fan failure → temperature increase → motor overload".

[0181] To achieve intelligent causal optimization, the system integrates the LLM into the training loop of the GNN to form an interactive feedback mechanism: after the GNN is trained for several rounds, the current graph structure and node / edge representations are fed back to the LLM, and the LLM evaluates the rationality of edges or paths in combination with the semantic verification process of S4 and outputs a correction signal, and the formula is updated by the joint loss:

[0182]

[0183] where, L GNN represents the GNN prediction loss, represents the weight parameter, L LLM_feedback represents the supervised loss generated by the LLM, and the formula is as follows:

[0184]

[0185] where, C LLM ( e ) represents the LLM semantic confidence of edge e , p ( e ) represents the GNN predicted edge probability, entropy ( G ) represents the graph entropy, represents the weight parameter. The output of the LLM is converted into a supervised signal for the GNN, such as adding a loss term to penalize conflicting edges or strengthening high-confidence paths.

[0186] The final optimized causal graph G opt includes enhanced node / edge representations, dynamic weights, and cross-graph association information as the input for S5 fault diagnosis.

[0187] Step S5, Fault Diagnosis and Knowledge Retrieval:

[0188] Since the causal graph output by S4 only provides structured potential causal paths, lacks specific historical context, risk assessment, and natural language expression, and cannot directly generate a reliable diagnostic report, the system needs to combine the local knowledge base to generate a fault identification result. Specifically, the system retrieves the historical fault instance most similar to the current diagnosis result in the local knowledge base through keyword extraction and graph matching technology. The similarity calculation adopts a joint criterion of semantic vector embedding and causal structure matching, and the following similarity function is defined:

[0189]

[0190] where, R is the current diagnosis result, C is the historical fault case, Simemb ( R , C ) represents the semantic similarity of the fault description, Sim topo ( R , C ) represents the structural similarity of the fault causal link, represents the weight parameter. When the similarity is higher than the preset threshold, the system extracts the key information of the matching cases and generates a natural language diagnosis report, the content of which includes the fault location, potential causes, handling suggestions, and risk levels and alarm suggestions.

[0191] If the retrieval fails, the system triggers an external knowledge completion mechanism to handle novel fault modes. First, the system queries multiple source external knowledge sources online, extracts relevant entries using uncertainty sampling, and generates structured metadata with the LLM. The new knowledge is incrementally integrated into the local knowledge base and fed back to the causal graph generation module in S3 to expand the coverage of the causal graph. The update is limited to the relevant parts of the current task to avoid global recomputation. Based on the updated knowledge base, the similarity is recalculated and the report is generated.

[0192] Finally, the module pushes the diagnosis results in both structured data and natural language forms to the user interface and the upper-layer control system, providing an accurate decision-making basis for fault response.

[0193] Step S6, System verification and feedback optimization:

[0194] The system actively obtains the feedback annotations based on the user's professional judgment. For positive feedback, the system will solidify the processing path, model configuration, and key parameters of this task as an optimization reference example to enhance the performance and efficiency of future similar tasks. For negative feedback, the system starts a reverse analysis process, correlates the user feedback information with the diagnosis process log to form a labeled historical task sample library, and constructs a closed-loop optimization system in combination with the active learning strategy and the feedback explanation assisted by the LLM. First, causal deviation identification is performed, and deviation indicators are calculated for the GNN model in S4. Calculate the deviation indicators:

[0195]

[0196] If the deviation When it exceeds the threshold, the system identifies potential problem edges or subgraphs and triggers the correction process. After the GNN model parameters are corrected based on the feedback data, a more optimized causal graph that better conforms to the user's cognition will be generated. The system uses this improved causal graph to re-execute the diagnostic process in step S5 until the user gives positive feedback on the regenerated diagnostic report or reaches the preset maximum number of iterations. In addition, the system introduces an active learning mechanism that preferentially updates the historical knowledge base when the diagnostic confidence is low or there is a knowledge conflict. The active learning mechanism is a global optimization triggered by feedback analysis, and the external knowledge completion mechanism targets the local knowledge gaps of the current task.

[0197] As Figure 2 shown, it describes multiple causal paths that lead to an overload fault. Figure 2 It is shown in [reference] that the overload fault is directly caused by the increase in winding temperature, which may be due to two factors: the decrease in heat dissipation capacity or the excessive input voltage. Among them, the decrease in heat dissipation capacity can be caused by the abnormal fan, forming a compound causal chain.

[0198] Embodiment 2

[0199] This embodiment provides an industrial fault diagnosis system with intelligent causal correction. The system adopts a modular architecture design and includes six core functional modules: a task definition and standardization module, a metadata generation module, a causal relationship graph construction module, a causal graph optimization module based on graph neural networks, a fault diagnosis and knowledge retrieval module, and a system verification and feedback optimization module. Data interaction and process coordination between modules are achieved through standardized interfaces, and a complete fault diagnosis closed-loop is formed through seven consecutive processing steps: Step S1, first receive and standardize the user's task requirements; Step S2, comprehensively preprocess and improve the quality of the uploaded monitoring data, and generate structured metadata as the unified entry for subsequent processing; Step S3, on this basis, use multi-source algorithms and LLM to construct an initial causal relationship graph; Step S4, deeply optimize and correct the causal structure through graph neural networks, and construct a causal composite relationship graph; Step S5, perform accurate fault diagnosis and information retrieval in combination with the knowledge base; Step S6, finally achieve system verification and continuous optimization through user feedback. This process design ensures that the system has the ability of adaptive learning and continuous evolution characteristics, and can effectively handle complex fault scenarios in the industrial environment.

[0200] A computer-readable storage medium stores multiple instructions, and the instructions are suitable for being loaded and executed by a processor of a terminal device to perform the industrial fault diagnosis method with intelligent causal correction described above.

[0201] A terminal device includes a processor and a computer-readable storage medium. The processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, and the instructions are adapted to be loaded and executed by the processor for the industrial fault diagnosis method of intelligent causal correction.

[0202] The above are all preferred embodiments of the present invention, and the protection scope of the present invention is not limited thereby. Therefore, all equivalent changes made according to the structure, shape, and principle of the present invention shall be covered within the protection scope of the present invention.

Claims

1. An industrial fault diagnosis method with intelligent causal correction, characterized in that, Including: Obtain monitoring data and user task requirements; Construct task structure information based on the obtained user task requirements; Process the monitoring data through a three-level cascaded framework to generate metadata; Construct an initial causal graph structure using the task structure information and metadata; Deeply optimize the initial causal graph structure through a graph neural network to obtain a causal composite relationship graph; Use the causal composite relationship graph and combine it with a knowledge base for fault diagnosis and information retrieval; Generate a structured diagnostic report; The three-level cascaded framework processes the monitoring data, covering three sub-modules: unified and complemented data format, adaptive noise suppression, and data trend prediction; The construction of the causal composite relationship graph includes three components: a data-driven causal hypothesis generation component, an LLM semantic verification and preliminary correction component, and a multi-source information integration component, aiming to infer the causal dependence relationship between variables from the power equipment monitoring data to support subsequent fault diagnosis and prediction; The overall process starts with pure statistic-driven hypothesis generation and gradually integrates semantic and domain knowledge to achieve the evolution from a candidate graph to a robust comprehensive causal graph; The data-driven causal hypothesis generation component, based on the set of variables of interest in the metadata, uses four complementary statistical causal structure learning algorithms to batch test all variable pairs and generate diverse candidate causal graphs; The LLM semantic verification and preliminary correction component introduces an LLM and an industrial knowledge graph to optimize four candidate causal graphs. The domain knowledge prior and the LLM perform semantic reasoning and probability evaluation to perform semantic rationality verification, conflict identification, and structural adjustment on the graph edges; The multi-source information integration component fuses multiple candidate causal graphs optimized by the LLM to generate a preliminary comprehensive causal graph G; The fusion process is based on a Bayesian network framework, combined with data-driven confidence, LLM semantic confidence, and domain knowledge prior to ensure the connectivity and acyclicity of the graph structure; The deep optimization of the initial causal graph structure through a graph neural network to obtain a causal composite relationship graph includes using the preliminarily constructed causal graph and hybrid features as inputs, and successively realizing graph embedding representation, edge weight optimization, and causal path enhancement based on the graph neural network, and using the LLM to perform semantic completion and inference correction on potential causal structures.

2. The industrial fault diagnosis method with intelligent causal correction according to claim 1, characterized in that The process of generating metadata by processing the monitoring data through a three-level cascaded framework includes first performing feature normalization on the monitoring data using the Robust Scaling method, scaling based on the quartiles and median of the data, and interpolating the missing regions in the time series feature matrix using the weighted average of several latest valid observed values before the missing points; then estimating the noise level in the monitoring data through the median absolute deviation, and performing multi-scale decomposition on the detected high-frequency components in the monitoring data based on the estimated noise level using wavelet transform, decomposing the data sequence into low-frequency approximation coefficients and high-frequency detail coefficients; for low-frequency noise, non-Gaussian distributed noise, or structural noise, use a GAN-based denoising model to suppress it.

3. An industrial fault diagnosis method with intelligent causal correction according to claim 2, characterized in that, Said constructing an initial causal graph structure by using task structure information and metadata includes first constructing a causal graph by using the Granger causality test algorithm G Granger , generating a directed graph by analyzing the time series prediction relationship between variables, expressed as: ， Among them, RSS resticted is the sum of squared residuals fitted using only B historical data, RSS unresticted is the sum of squared residuals fitted using i and j historical data, k represents the lag order, dynamically selected based on autocorrelation function analysis, p represents the autoregressive order. The algorithm assumes that if the past values of variable j can significantly improve the prediction of variable i i.e., F the value is higher than the set threshold, then there exists j → i causal relationship.

4. An industrial fault diagnosis method with intelligent causal correction according to claim 3, characterized in that The step of constructing an initial causal graph structure by using task structure information and metadata further includes using the PC algorithm based on conditional independence testing to assume that there may be potential causal relationships between all variables, and gradually increasing the conditional set starting from a fully connected graph k , testing the variables i and j for conditional independence; By iteratively increasing k, the algorithm gradually removes redundant edges, constructs a sparse undirected skeleton graph, and infers the edge directions in combination with Meek's rules to construct a directed graph GPC. Among them, for the initial condition set k, the correlation coefficient of each variable pair (i, j) is calculated r ij∣k , and its conditional independence score is tested Z : ， Among them, r ij∣k is a variable i and j is the partial correlation coefficient under a given set of conditions k , n is the sample size. The larger the absolute value of the z-value, the stronger the correlation, thus rejecting the conditional independence hypothesis.

5. An industrial fault diagnosis method with intelligent causal correction according to claim 4, characterized in that, The construction of the initial causal graph structure using task structure information and metadata further includes assuming no direct causal connection between variables using the GES algorithm, and constructing the graph structure starting from an empty graph using a greedy search strategy G GES , adding edges through forward search and removing edges through backward search to maximize a global scoring function. Among them, the scoring gain of all edges is calculated starting from the empty graph, the edge with the highest score is added in the forward stage until no operation can optimize the scoring function, and redundant edges are removed in the backward stage to further optimize the scoring function to ensure that the local optimal solution approximates the global optimal solution, which is expressed as: ， Among them, is a variable i given the probability of its parent node , | E | is the number of edges, λ is the regularization parameter.

6. The industrial fault diagnosis method with intelligent causal correction according to claim 5, wherein The construction of the initial causal graph structure using task structure information and metadata further includes preliminarily optimizing the four candidate causal graphs using an LLM and an industrial knowledge graph. Among them, the high-dimensional text embedding model BERT is used to map each variable of interest into a high-dimensional continuous vector space in the knowledge prior emb , as the basic input for the LLM to perform semantic evaluation; then the metadata vector features X total are combined with emb to generate an embedding vector X , and the comprehensive confidence is calculated by combining the multi-source information confidence: the multiple candidate causal graphs optimized by the LLM are fused to generate a preliminary comprehensive causal graph G. The fusion process is based on the Bayesian network framework, combining data-driven confidence, LLM semantic confidence, and domain knowledge prior to ensure the connectivity and acyclicity of the graph structure.

7. An industrial fault diagnosis method with intelligent causal correction according to claim 6, characterized in that, Generate graph-level global node embeddings and edge embeddings for the optimized graph of the current causal graph and all relevant historical causal graphs in the historical causal graph library maintained by the system, and establish an asymmetric graph similarity evaluation function S ( G opt , G hist ), which is used to quantify the current optimized graph G opt and the historical causal graph G hist The dual structural and semantic similarity between them is expressed as: ， Among them, represents the Jaccard similarity of the node set, represents the normalized Hamming distance of the edge set, represents the cosine similarity of the spectral core topological feature vector.

8. An industrial fault diagnosis method with intelligent causal correction according to claim 7, characterized in that The initial causal graph structure is deeply optimized through a graph neural network to obtain a causal composite relationship graph, which also includes integrating the LLM into the training loop of the GNN to form an interactive feedback mechanism. After the GNN is trained for several rounds, the current graph structure and node-edge representations are fed back to the LLM. The LLM combines the semantic verification process to evaluate the rationality of edges or paths and outputs a correction signal. The formula is the joint loss update: ， Among them, L GNN represents the GNN prediction loss, represents the weight parameter, L LLM_feedback represents the supervision loss generated by the LLM, and the formula is as follows: ， Among them, C LLM ( e ) represents an edge e of the LLM semantic confidence, p ( e ) represents the GNN predicted edge probability, entropy ( G ) represents the graph entropy, represents the weight parameter.

9. An industrial fault diagnosis method with intelligent causal correction according to claim 8, characterized in that, Utilizing the causal composite relationship graph and combining it with the knowledge base for fault diagnosis and information retrieval includes, through keyword extraction and graph matching techniques, retrieving the historical fault instances most similar to the current diagnosis result in the local knowledge base, calculating the similarity using the joint criterion of semantic vector embedding and causal structure matching. When the similarity is higher than the preset threshold, the system extracts the key information of the matching case and generates a natural language diagnosis report. If the retrieval fails, relevant entries are extracted using uncertainty sampling and the LLM is used to generate structured metadata, which is incrementally integrated into the local knowledge base and fed back to the causal graph to expand the coverage of the causal graph, recalculating the similarity and generating a report. The similarity is expressed as: ， Among them, R is the current diagnosis result, C is the historical fault case, Sim emb ( R , C ) represents the semantic similarity of the fault description, Sim topo ( R , C ) represents the structural similarity of the fault causal link, represents the weight parameter.

10. An industrial fault diagnosis system with intelligent causal correction, which executes an industrial fault diagnosis method with intelligent causal correction as described in claim 1, characterized in that, including: A data acquisition module configured to acquire monitoring data and user task requirements; A task structure module configured to construct task structure information based on the acquired user task requirements; A metadata module configured to generate metadata by processing monitoring data through a three-level cascaded framework; A causal graph module configured to construct an initial causal graph structure using the task structure information and metadata; A deep optimization module configured to deeply optimize the initial causal graph structure through a graph neural network to obtain a causal composite relationship graph; A fault diagnosis module configured to perform fault diagnosis and information retrieval using the causal composite relationship graph and combining it with the knowledge base; A report module configured to generate a structured diagnosis report.

Images