Boiler fault self-diagnosis method and related device

Abstract

The invention discloses a boiler fault self-diagnosis method and related device, and the method comprises the steps: S1, collecting boiler operation parameters, generating virtual data in combination with a digital twinborn model, and constructing a multi-modal monitoring data set; s2, carrying out preprocessing and anomaly detection on the data set by utilizing an edge computing node, and obtaining a preliminary anomaly signal and a feature vector; s3, uploading the abnormal signal and the feature vector to a cloud end, and performing simulation verification through a digital twin engine; s4, inputting the feature vector and a verification result into a hybrid enhancement diagnosis model, and outputting a fault type and a probability; s5, reasoning a fault source and a propagation path in combination with the knowledge graph according to the fault type and the probability; and S6, generating a maintenance scheme based on the fault information, and optimizing the diagnosis model by using operation and maintenance feedback. According to the method, the boiler operation parameters are collected and combined with the digital twinborn model to generate the virtual sensor data, the multi-modal monitoring data set is constructed, synchronous monitoring of multiple data is achieved, and one-sidedness of single-parameter monitoring is avoided.

Description

Technical Field

[0001] The invention belongs to the technical field of equipment maintenance, and in particular relates to a boiler fault self-diagnosis method and related devices. Background Art

[0002] In industrial production, boilers are core power equipment, and their stable operation is crucial to production efficiency and safety. Traditional boiler fault diagnosis methods rely primarily on threshold alarms based on single physical parameters, such as monitoring only temperature and pressure. This approach is significantly one-sided and fails to fully capture potential anomalies under complex operating conditions. This can easily lead to misdiagnosis or missed faults, hindering the timely detection of potential faults in the early stages.

[0003] While the development of IoT technology allows for the collection of more operational data, data transmission and processing present significant challenges. Uploading large amounts of raw data directly to the cloud not only strains network bandwidth but also increases data processing costs. Furthermore, existing edge computing models are inefficient in feature extraction and anomaly detection, failing to meet real-time requirements.

[0004] In terms of fault analysis models, traditional methods often use a single algorithm to process data, making it difficult to effectively analyze both structured and unstructured features simultaneously. Furthermore, data sharing between different boiler nodes presents privacy and security issues. Existing models are weak at identifying rare faults and are unable to fully utilize multi-source data to improve the accuracy and generalization of fault diagnosis. Furthermore, there is a lack of intuitive and effective methods for analyzing fault propagation paths, making it difficult to accurately locate the source of the fault based on historical data and real-time monitoring. Therefore, a more advanced, comprehensive, and efficient boiler fault self-diagnosis method is urgently needed. Summary of the Invention

[0005] The purpose of the present invention is to provide a boiler fault self-diagnosis method and related devices to overcome the deficiencies of the prior art.

[0006] In a first aspect, the present invention provides a boiler fault self-diagnosis method, comprising the following steps: S1, collects the operating parameters of the boiler body, generates virtual sensor data through the digital twin model, and obtains a multimodal monitoring data set; S2, based on the multimodal monitoring dataset, uses edge computing nodes to perform data preprocessing and anomaly detection to obtain preliminary anomaly signals and corresponding feature vectors; S3, uploading the preliminary abnormal signal and characteristic vector to the cloud diagnosis system, performing multi-physics field coupling simulation through the digital twin engine, and obtaining a virtual verification result; S4, based on the feature vector and the virtual verification result, input the pre-trained hybrid enhanced diagnosis model to perform fault classification and obtain the fault type and probability distribution; S5, based on the fault type and probability distribution, combined with the fault propagation knowledge graph, causal reasoning is performed to obtain the fault root location and potential propagation path; S6, based on the fault root location and potential propagation path, generates a maintenance decision plan through the reinforcement learning optimization module, and dynamically updates the operation and maintenance feedback data to the hybrid enhanced diagnostic model to complete the closed-loop optimization.

[0007] Furthermore, the edge computing node in S2 runs a lightweight TinyTransformer model, specifically including: Perform time alignment and normalization on the multimodal monitoring dataset to obtain standardized data; Extracting time domain statistical features and frequency domain wavelet coefficients through a sliding window to form the feature vector; When the characteristic vector exceeds a preset health threshold, the preliminary abnormal signal is triggered.

[0008] Furthermore, the training method of the hybrid enhanced diagnostic model described in S4 includes: Generate simulation data sets covering normal states and typical fault states through the digital twin engine; A federated learning framework is used to aggregate edge computing data from multiple boiler nodes and jointly optimize model parameters.

[0009] Furthermore, the method for constructing the fault propagation knowledge graph described in S5 includes: Extract equipment topology from boiler design drawings; Mark the cause and effect relationship of failures based on historical maintenance records; Supplement rare fault propagation paths through expert rules.

[0010] Furthermore, in the causal reasoning described in S5, a Bayesian network is used to dynamically update the propagation path probability.

[0011] Furthermore, S6 specifically includes: Receive fault root location information and potential propagation path data from S5; Construct a reinforcement learning state space based on the boiler's current operating state parameters, production plan constraints, and maintenance resource availability; According to the preset reward function, a maintenance decision plan is selected from the action space; Execute the selected maintenance decision plan and collect equipment response data in real time to obtain the actual maintenance effect; Compare and verify the actual maintenance effect with the predicted propagation path to obtain the verification result; Dynamically update the hybrid enhanced diagnostic model parameters based on the verification results; If the verification result is a new fault propagation relationship, the new fault propagation relationship is added to the fault propagation knowledge graph of S5.

[0012] Furthermore, the hybrid enhanced diagnostic model described in S4 includes a two-stage series structure of a lightweight decision tree classifier and a graph neural network classifier.

[0013] In a second aspect, the present invention provides a boiler fault self-diagnosis system, comprising: Acquisition module: collects the operating parameters of the boiler itself, generates virtual sensor data through the digital twin model, and obtains a multimodal monitoring data set; Processing module: Based on the multimodal monitoring data set, edge computing nodes are used to perform data preprocessing and anomaly detection to obtain preliminary anomaly signals and corresponding feature vectors; Coupling module: uploads the preliminary abnormal signal and characteristic vector to the cloud diagnosis system, performs multi-physics field coupling simulation through the digital twin engine, and obtains virtual verification results; Enhancement module: Based on the feature vector and virtual verification results, the pre-trained hybrid enhanced diagnosis model is input to perform fault classification to obtain the fault type and probability distribution; Reasoning module: Based on the fault type and probability distribution, combined with the fault propagation knowledge graph, causal reasoning is performed to obtain the fault root location and potential propagation path; Optimization module: Based on the fault root location and potential propagation path, a maintenance decision plan is generated through the reinforcement learning optimization module, and the operation and maintenance feedback data is dynamically updated to the hybrid enhanced diagnostic model to complete closed-loop optimization.

[0014] In a third aspect, the present invention provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of a boiler fault self-diagnosis method when executing the computer program.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of a method for self-diagnosis of boiler faults.

[0016] Compared with the prior art, the present invention has the following beneficial technical effects: This invention provides a method for self-diagnosing boiler faults. By collecting boiler operating parameters and combining them with a digital twin model to generate virtual sensor data, a multimodal monitoring dataset is constructed. This allows for simultaneous monitoring of physical parameters such as temperature, pressure, and vibration, along with virtual simulation data, avoiding the one-sided nature of single-parameter monitoring. The digital twin model simulates the normal operation of the boiler. When discrepancies between actual monitoring data and virtual data occur, potential anomalies can be quickly identified, providing a more proactive approach than traditional threshold alarms.

[0017] Specifically, edge nodes run the TinyTransformer model to perform time alignment, normalization, and frequency domain feature extraction on the data. Only preliminary abnormal signals and feature vectors are uploaded to the cloud, reducing data transmission volume. Multi-physics coupled simulation using a digital twin engine verifies whether edge-detected anomalies are caused by real faults.

[0018] Specifically, a two-level serial structure of lightweight decision trees and graph neural networks is adopted. The decision tree can quickly process structured features, while the graph neural network is good at analyzing unstructured features in equipment topology relationships. When the federated learning framework aggregates data from multiple boiler nodes, there is no need to upload the original data, only the model parameters are shared, which not only ensures data privacy but also improves the model's recognition rate for rare faults.

[0019] Specifically, by extracting equipment topology from design drawings and annotating causal relationships based on historical maintenance records, the resulting knowledge graph can intuitively visualize fault propagation paths. A Bayesian network dynamically adjusts propagation path probabilities based on real-time monitoring data. For example, if a boiler induced draft fan experiences abnormal vibration, the system can accurately locate the source of the fault based on the vibration spectrum and historical case studies. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 Schematic diagram of a boiler fault self-diagnosis method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0022] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0023] See Figure 1 The present invention provides a boiler fault self-diagnosis method, comprising the following steps: S1, collects the operating parameters of the boiler body, generates virtual sensor data through the digital twin model, and obtains a multimodal monitoring data set; The operating parameters of the boiler itself include data collected by physical sensors: temperature parameters (feedwater temperature, steam temperature, and exhaust gas temperature); pressure parameters (drum pressure and furnace negative pressure); flow parameters (feedwater flow and steam flow); and water quality parameters (pH and dissolved oxygen content). A digital twin model is constructed that incorporates the boiler's geometry, material properties, and boundary conditions. These collected parameters are input into the digital twin model as boundary conditions. The multi-physics coupling equations are solved, and the following virtual data is output: the three-dimensional temperature field distribution within the furnace, the stress distribution on the heating surface tube wall, the flue gas flow velocity field, and the predicted remaining life of key components. The physical and virtual sensor data are timestamp-aligned, and the Kalman filter algorithm is used for data fusion to generate a multimodal monitoring dataset.

[0024] S2, based on the multimodal monitoring dataset, uses edge computing nodes to perform data preprocessing and anomaly detection to obtain preliminary anomaly signals and corresponding feature vectors; The multimodal monitoring dataset is normalized, and the dynamic time warping algorithm is used to perform time alignment of multi-source data. The time domain statistical features and frequency domain wavelet coefficients are extracted through a sliding window to form the feature vector. The lightweight TinyTransformer model is run to evaluate the feature vector. When the feature vector exceeds the preset health threshold, the preliminary abnormal signal is triggered.

[0025] S3, uploading the preliminary abnormal signal and characteristic vector to the cloud diagnosis system, performing multi-physics field coupling simulation through the digital twin engine, and obtaining a virtual verification result; Receive abnormal signals uploaded by edge nodes, start the digital twin engine, load the device status snapshot at the abnormal moment, set simulation parameters, and generate a simulation data set covering normal status and typical fault status through the digital twin engine; use the federated learning framework to aggregate edge computing data from multiple boiler nodes, jointly optimize model parameters, and generate virtual verification results.

[0026] S4, based on the feature vector and the virtual verification result, input the pre-trained hybrid enhanced diagnosis model to perform fault classification and obtain the fault type and probability distribution; The hybrid enhanced diagnosis model uses a two-stage cascade structure consisting of a lightweight decision tree classifier and a graph neural network classifier. This architecture fully combines the advantages of the decision tree classifier's rapid initial screening with the graph neural network classifier's deep mining of complex relationships, improving fault classification accuracy while ensuring diagnostic efficiency.

[0027] The feature vectors uploaded by the edge computing node and the output of the digital twin engine are received as input to the lightweight decision tree classifier. The decision tree classifier is constructed based on the information gain criterion. The information gain is obtained by calculating the information entropy difference of the data set before and after partitioning. During the construction process, starting from the root node, the information gain of all features is calculated for each node. The feature with the highest information gain is selected as the node's partitioning attribute, and the dataset is divided into multiple child nodes based on the different values ​​of this feature. This process is repeated until a stopping condition is met, such as the number of samples in a node falling below a preset threshold, the information gain falling below a specific value, or all samples belonging to the same category. Through the layered division of the decision tree, the input data is quickly filtered into different category nodes. For data that can be clearly attributed to a specific fault type, a preliminary classification result is directly output. For samples that cannot be accurately judged, they are passed to the next level of the graph neural network classifier for further analysis.

[0028] The graph neural network classifier constructs a graph structure representing the boiler's equipment structure and fault propagation relationships. The graph's nodes are represented by the boiler's individual equipment components, whose attributes include information such as component type, specifications, and operating parameters. The graph's edges, including the physical connections between components and the impact of faults, record information such as the connection type and impact level. A graph convolutional network (GCN) is used to learn features from graph-structured data. The graph convolution operation updates the feature representation of a node by aggregating information from its neighbors. Through multiple layers of graph convolution, high-level feature representations of nodes and edges are continuously learned, capturing the complex relationships between faults. Finally, the learned node features are input into a fully connected layer and a softmax function to calculate the probability of each sample belonging to a different fault type, outputting the final fault type and probability distribution.

[0029] S5, based on the fault type and probability distribution, combined with the fault propagation knowledge graph, causal reasoning is performed to obtain the fault root location and potential propagation path; Extract equipment topology from boiler design drawings. Utilize computer vision technology to analyze boiler design drawings and extract the physical connection relationships between equipment components through feature point matching and graph structure recognition algorithms. Construct a topology graph containing node attributes (equipment type, material, rated parameters) and edge attributes (connection type, flow direction, energy transfer relationships), forming a basic equipment network model. Mark the cause and effect relationship of failures based on historical maintenance records; Perform natural language processing on historical maintenance records and annotate causal relationships through the following steps: The BERT-BiLSTM-CRF model is used to identify entities such as fault components, fault phenomena, and repair measures. A fault causal relationship triple (source fault, causal relationship, target fault) is constructed based on dependency syntax analysis. The order of fault occurrence is mined through timestamp sequences to determine the causal direction.

[0030] Supplement rare fault propagation paths through expert rules; For rare fault scenarios, experts in the boiler field are invited to develop supplementary rules. These rules are stored in the knowledge graph in the form of production rules as prior knowledge of the Bayesian network.

[0031] The knowledge graph is converted into a Bayesian network. During the reasoning process, probability calculation is performed and the root cause of the fault is located using three indicators: causal strength, minimum cut set analysis, and time series consistency. Based on the probability distribution of Bayesian networks, the Dijkstra algorithm is used to calculate the most likely propagation path and supplement the rare fault propagation path.

[0032] S6, based on the fault root location and potential propagation path, generates a maintenance decision plan through the reinforcement learning optimization module, and dynamically updates the operation and maintenance feedback data to the hybrid enhanced diagnostic model to complete the closed-loop optimization.

[0033] Receive fault root location information and potential propagation path data from S5; Construct a reinforcement learning state space based on the boiler's current operating state parameters, production plan constraints, and maintenance resource availability; According to the preset reward function, a maintenance decision plan is selected from the action space; Execute the selected maintenance decision plan and collect equipment response data in real time to obtain the actual maintenance effect; Compare and verify the actual maintenance effect with the predicted propagation path to obtain the verification result; Dynamically update the hybrid enhanced diagnostic model parameters based on the verification results; If the verification result is a new fault propagation relationship, the new fault propagation relationship is added to the fault propagation knowledge graph of S5.

[0034] Specifically, the system receives the fault root location information and potential propagation path data from step S5 and constructs a reinforcement learning state space based on the boiler's current operating state parameters, production plan constraints, and maintenance resource availability. Operating state parameters include real-time data such as temperature, pressure, and flow, reflecting the boiler's current operating status. Production plan constraints include production task urgency and allowable downtime, representing production-level constraints. Maintenance resource availability includes factors such as the number of maintenance personnel, their skill levels, and spare parts inventory, representing available maintenance resources. These factors are integrated to form a state space containing information from multiple dimensions, with each state corresponding to the comprehensive condition of the boiler under a specific scenario. Based on a preset reward function, a maintenance decision plan is selected from the action space. The action space encompasses various maintenance operations, such as replacing parts, adjusting operating parameters, and increasing monitoring frequency. The reward function comprehensively considers factors such as maintenance cost, downtime losses, and fault repair effectiveness, quantifying the potential benefits of each action. The system calculates the expected reward for each action in the current state and selects the action with the highest reward as the maintenance decision plan. For example, if a fault is caused by component wear, replacing the part, while costly, can quickly and completely resolve the problem and reduce long-term downtime losses. The reward function calculates the reward value for this action based on these factors and compares it with other actions (such as temporary repairs) to ultimately determine the optimal solution. Once the solution is determined, the selected maintenance decision is executed, and sensors deployed in key areas of the boiler collect real-time equipment response data to determine the actual maintenance results. This data is used to measure whether the maintenance plan achieved its intended goals, such as whether equipment parameters have returned to normal ranges and whether the fault has been eliminated.

[0035] The actual maintenance results are compared and verified against the predicted propagation paths to generate a verification result. If the actual results match the predictions, the diagnosis and decision-making are effective; if there are deviations, the system indicates deficiencies in the diagnosis or decision-making stages. The parameters of the hybrid enhanced diagnosis model are dynamically updated based on the verification results. If the verification reveals inaccurate model predictions, the system uses new data to calculate the gradients of the model parameters and adjusts them using specific algorithms (such as stochastic gradient descent) to optimize the model's predictive capabilities, enabling more accurate fault diagnosis in the future. If the verification results reveal new fault propagation relationships, the system adds these new relationships to the S5 fault propagation knowledge graph. This process involves extracting key information from the new relationships, assessing their credibility, adding corresponding nodes and edges to the knowledge graph, and updating relevant attributes and connection weights. This ensures that the knowledge graph reflects the latest fault propagation patterns, providing a more comprehensive and accurate basis for subsequent causal reasoning. Through this closed-loop optimization process, the boiler fault self-diagnosis system can continuously learn and improve, adapting to complex and changing operating environments and improving the accuracy and effectiveness of fault diagnosis and maintenance decisions.

[0036] The present invention also provides a boiler fault self-diagnosis system, comprising: Acquisition module: collects the operating parameters of the boiler itself, generates virtual sensor data through the digital twin model, and obtains a multimodal monitoring data set; Processing module: Based on the multimodal monitoring data set, edge computing nodes are used to perform data preprocessing and anomaly detection to obtain preliminary anomaly signals and corresponding feature vectors; Coupling module: uploads the preliminary abnormal signal and characteristic vector to the cloud diagnosis system, performs multi-physics field coupling simulation through the digital twin engine, and obtains virtual verification results; Enhancement module: Based on the feature vector and virtual verification results, the pre-trained hybrid enhanced diagnosis model is input to perform fault classification to obtain the fault type and probability distribution; Reasoning module: Based on the fault type and probability distribution, combined with the fault propagation knowledge graph, causal reasoning is performed to obtain the fault root location and potential propagation path; Optimization module: Based on the fault root location and potential propagation path, a maintenance decision plan is generated through the reinforcement learning optimization module, and the operation and maintenance feedback data is dynamically updated to the hybrid enhanced diagnostic model to complete closed-loop optimization.

[0037] In one embodiment of the present invention, a computer device is provided. The computer device includes a processor and a memory, wherein the memory is configured to store a computer program, the computer program including program instructions, and the processor is configured to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or may also be another general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic device, discrete hardware component, etc. It is the computing core and control core of the terminal and is suitable for implementing one or more instructions, specifically loading and executing one or more instructions in the computer storage medium to implement a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used in a method for self-diagnosing faults in a boiler.

[0038] The present invention also provides a storage medium, specifically a computer-readable storage medium (Memory). The computer-readable storage medium is a memory device in a computer device, used to store programs and data. It is understood that the computer-readable storage medium herein may include both built-in storage media in the computer device and, of course, extended storage media supported by the computer device. The computer-readable storage medium provides storage space, which stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for being loaded and executed by a processor. These instructions may be one or more computer programs (including program code). It should be noted that the computer-readable storage medium herein may be high-speed RAM memory or non-volatile memory, such as at least one disk drive. The processor may load and execute the one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the boiler fault self-diagnosis method described in the above-mentioned embodiment.

[0039] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0040] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0041] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0042] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0043] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A boiler fault self-diagnosis method, characterized in that: The following steps are involved: S1, collects the operating parameters of the boiler body, generates virtual sensor data through the digital twin model, and obtains a multimodal monitoring data set; S2, based on the multimodal monitoring dataset, uses edge computing nodes to perform data preprocessing and anomaly detection to obtain preliminary anomaly signals and corresponding feature vectors; S3, uploading the preliminary abnormal signal and characteristic vector to the cloud diagnosis system, performing multi-physics field coupling simulation through the digital twin engine, and obtaining a virtual verification result; S4, based on the feature vector and the virtual verification result, input the pre-trained hybrid enhanced diagnosis model to perform fault classification and obtain the fault type and probability distribution; S5, based on the fault type and probability distribution, combined with the fault propagation knowledge graph, causal reasoning is performed to obtain the fault root location and potential propagation path; S6, based on the fault root location and potential propagation path, generates a maintenance decision plan through the reinforcement learning optimization module, and dynamically updates the operation and maintenance feedback data to the hybrid enhanced diagnosis model to complete the closed-loop optimization.

2. A boiler fault self-diagnosis method according to claim 1, characterized in that: The edge computing node described in S2 runs a lightweight TinyTransformer model, specifically including: Perform time alignment and normalization on the multimodal monitoring dataset to obtain standardized data; Extracting time domain statistical features and frequency domain wavelet coefficients through a sliding window to form the feature vector; When the characteristic vector exceeds a preset health threshold, the preliminary abnormal signal is triggered.

3. A boiler fault self-diagnosis method according to claim 2, characterized in that: The training method of the hybrid enhanced diagnostic model described in S4 includes: Generate simulation data sets covering normal states and typical fault states through the digital twin engine; A federated learning framework is used to aggregate edge computing data from multiple boiler nodes, jointly optimize model parameters, and generate virtual verification results.

4. A boiler fault self-diagnosis method according to claim 3, characterized in that: The method for constructing the fault propagation knowledge graph described in S5 includes: Extract equipment topology from boiler design drawings; Mark the cause and effect relationship of failures based on historical maintenance records; Supplement rare fault propagation paths through expert rules.

5. A boiler fault self-diagnosis method according to claim 4, characterized in that: In the causal reasoning described in S5, a Bayesian network is used to dynamically update the propagation path probability.

6. A boiler fault self-diagnosis method according to claim 5, characterized in that: S6 specifically includes: Receive fault root location information and potential propagation path data from S5; Construct a reinforcement learning state space based on the boiler's current operating state parameters, production plan constraints, and maintenance resource availability; According to the preset reward function, a maintenance decision plan is selected from the action space; Execute the selected maintenance decision plan and collect equipment response data in real time to obtain the actual maintenance effect; Compare and verify the actual maintenance effect with the predicted propagation path to obtain the verification result; Dynamically update the hybrid enhanced diagnostic model parameters based on the verification results; If the verification result is a new fault propagation relationship, the new fault propagation relationship is added to the fault propagation knowledge graph of S5.

7. A boiler fault self-diagnosis method according to claim 6, characterized in that: The hybrid enhanced diagnostic model described in S4 includes a two-stage series structure of a lightweight decision tree classifier and a graph neural network classifier.

8. A boiler fault self-diagnosis system, characterized in that: include: Acquisition module: collects the operating parameters of the boiler itself, generates virtual sensor data through the digital twin model, and obtains a multimodal monitoring data set; Processing module: Based on the multimodal monitoring data set, edge computing nodes are used to perform data preprocessing and anomaly detection to obtain preliminary anomaly signals and corresponding feature vectors; Coupling module: uploads the preliminary abnormal signal and characteristic vector to the cloud diagnosis system, performs multi-physics field coupling simulation through the digital twin engine, and obtains virtual verification results; Enhancement module: Based on the feature vector and virtual verification results, the pre-trained hybrid enhanced diagnosis model is input to perform fault classification to obtain the fault type and probability distribution; Reasoning module: Based on the fault type and probability distribution, combined with the fault propagation knowledge graph, causal reasoning is performed to obtain the fault root location and potential propagation path; Optimization module: Based on the fault root location and potential propagation path, a maintenance decision plan is generated through the reinforcement learning optimization module, and the operation and maintenance feedback data is dynamically updated to the hybrid enhanced diagnostic model to complete closed-loop optimization.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images