Fault diagnosis model determination method, application method and device based on fault relation constraint for data scarce scene

Abstract

This application discloses a method, application method, and apparatus for determining a fault diagnosis model based on fault relationship constraints in data-scarce scenarios. It relates to the field of fault diagnosis technology. The determination method includes: acquiring multi-source information; standardizing the multi-source information and mapping it to nodes, then constructing a knowledge graph with causal and influence relationships as edges; the causal relationships are determined based on fault causes and measurement points; the influence relationships are determined based on fault modes and measurements; extracting key information from the knowledge graph to obtain a set of fault causes and a set of measurement points; constructing a relationship matrix between fault causes and measurement points based on the set of fault causes and the set of measurement points; constructing a hybrid dataset based on the relationship matrix; and training a supervised learning model on the hybrid dataset to obtain a fault diagnosis model. This method can solve the problems of traditional data-driven methods, such as reliance on large amounts of labeled data, lack of mechanistic constraints, and insufficient diagnostic accuracy and adaptability.

Description

Technical Field

[0001] This application relates to the field of fault diagnosis technology, and in particular to a fault diagnosis model determination method, application method and apparatus based on fault relationship constraints for data-scarce scenarios. Background Technology

[0002] Fault diagnosis is a key technology for ensuring the safe and stable operation of complex systems. Current mainstream data-driven methods (such as deep learning) typically rely on a large amount of well-labeled historical fault data to train high-precision models. However, in new systems, high-reliability equipment, or application scenarios with extremely high failure costs, historical fault data is often extremely scarce or even difficult to obtain, thus generally facing the application constraints of "small sample" or even "zero sample".

[0003] To address the problem of data scarcity, academia and industry have proposed various technical approaches, but these still have significant limitations in practical applications for fault diagnosis in complex systems. Firstly, in terms of data augmentation and generation, traditional methods (such as noise addition and scaling) and generative models (such as GANs and VAEs) primarily rely on learning and reconstructing existing data distributions. When fault samples are extremely limited, these methods struggle to effectively extrapolate to fault modes that exist physically outside the distribution, and they lack constraints on the system's fault propagation mechanisms, easily introducing samples that violate physical laws, thus affecting the model's reliability and interpretability.

[0004] Secondly, while transfer learning and meta-learning methods can alleviate the problem of insufficient samples by utilizing source domain data or cross-task experience, their performance is highly dependent on the similarity between the source and target domains. When the target system has a complex structure or unique mechanism, domain differences will significantly reduce model performance. Furthermore, these methods often struggle to effectively integrate knowledge from system design documents, FMECA analysis results, and expert experience, leading to a disconnect between "data-driven" and "knowledge utilization."

[0005] Furthermore, while mechanistic model-based methods generate data by constructing physical or mathematical models and possess a certain degree of interpretability, they often face challenges such as high modeling costs, long development cycles, and insufficient flexibility. For large and complex systems, constructing complete and accurate mechanistic models is typically difficult, and timely updates are challenging when the system's operating state changes or its configuration is adjusted.

[0006] Furthermore, existing methods also have shortcomings in modeling multiple concurrent faults and uncertainties. Most methods mainly focus on single fault scenarios and lack a systematic characterization of the coupling, superposition, and mutual influence of multiple faults. In terms of uncertainty handling, simple random perturbations are usually used to simulate sensor noise or measurement errors, failing to combine the different roles of different measurement points in the fault propagation process for modeling, resulting in deviations between the results and actual operating conditions.

[0007] Furthermore, existing methods are still mainly "purely data-driven" or "purely mechanism-driven," lacking a unified mechanism that can effectively utilize prior knowledge under conditions of scarce data and achieve continuous performance optimization as real data is gradually accumulated. It is difficult to balance the interpretability, reliability, and adaptability of the model.

[0008] In summary, existing technologies still face the following key challenges when applied to small-sample or zero-sample fault diagnosis of highly reliable and complex systems: difficulty in effectively utilizing multi-source prior knowledge, difficulty in generating effective samples that conform to mechanistic constraints, insufficient ability to model multiple faults and uncertainties, and lack of mechanisms to support continuous model evolution. These problems collectively limit the effectiveness of fault diagnosis methods in practical engineering applications.

[0009] Therefore, it is necessary to explore a new technical approach that, under conditions of extreme data scarcity, can fully utilize the existing domain knowledge of the system to effectively express and constrain fault-related relationships, and on this basis, construct a diagnostic model with good interpretability and consistency, while also having the ability to be gradually optimized and evolved with the accumulation of data, thereby improving the reliability and adaptability of fault diagnosis in complex systems. Summary of the Invention

[0010] The purpose of this application is to provide a fault diagnosis model determination method, application method and device based on fault relationship constraints for data-scarce scenarios. It can solve the technical problems of existing fault diagnosis methods that are difficult to effectively utilize multi-source prior knowledge under conditions of no sample or small sample, lack of fault mechanism constraints in generated data, insufficient ability to model multiple fault concurrency and uncertainty, and lack of model evolution ability with the accumulation of real data.

[0011] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a method for determining a fault diagnosis model based on fault relationship constraints for data-scarce scenarios, the method comprising: Acquire multi-source information; the multi-source information includes: fault cause, fault mode, probability of fault occurrence and measurement points.

[0012] After standardizing the multi-source information, it is mapped to nodes, and a knowledge graph is constructed with causal and influence relationships as edges; the causal relationship is the relationship determined based on the fault cause and the measurement point; the influence relationship is the relationship determined based on the fault mode and the measurement.

[0013] Key information is extracted from the knowledge graph to obtain a set of fault causes and a set of measurement points.

[0014] Based on the set of fault causes and the set of measurement points, a relationship matrix between fault causes and measurement points is constructed.

[0015] Based on the aforementioned relationship matrix, a hybrid dataset is constructed.

[0016] The supervised learning model is trained based on the hybrid dataset to obtain a fault diagnosis model.

[0017] Secondly, this application provides an application method for a fault diagnosis model based on fault relationship constraints for data-scarce scenarios, the application method including: Obtain the measurement status of the object to be diagnosed.

[0018] The status of the measurement point to be diagnosed is input into the fault diagnosis model to obtain the corresponding fault diagnosis result; the fault diagnosis model is a model obtained based on the fault diagnosis model determination method based on fault relationship constraints for data-scarce scenarios described above.

[0019] Based on the fault diagnosis results, the fault diagnosis model is continuously optimized using incremental learning or periodic retraining.

[0020] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method for determining a fault diagnosis model based on fault relationship constraints for data-scarce scenarios or the method for applying a fault diagnosis model based on fault relationship constraints for data-scarce scenarios.

[0021] Fourthly, this application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described method for determining a fault diagnosis model based on fault relationship constraints for data-scarce scenarios or the method for applying a fault diagnosis model based on fault relationship constraints for data-scarce scenarios.

[0022] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method for determining a fault diagnosis model based on fault relationship constraints for data-scarce scenarios or the method for applying a fault diagnosis model based on fault relationship constraints for data-scarce scenarios.

[0023] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a fault diagnosis model determination, application method, and apparatus based on fault relationship constraints for data-scarce scenarios. By comprehensively integrating multiple core multi-source information such as fault causes, fault modes, fault occurrence probabilities, and measurement points, it can completely cover all dimensions of fault diagnosis related data, laying a comprehensive and complete foundation of data support for subsequent full-process processing. By standardizing and organizing multi-source information to eliminate heterogeneous data differences, and then constructing a structured knowledge graph with standardized nodes and precisely defined causal and influence relationships, it can clarify the internal correlation logic of various fault-related information and achieve the structured and visualized organization of unstructured fault information. By extracting key information in a targeted manner from the constructed knowledge graph, it can accurately... This method effectively filters out redundant and invalid information, efficiently identifies core related elements for fault diagnosis, and simplifies the objects and complexity of subsequent data processing. Based on the filtered core set, a standardized relation matrix is ​​constructed, transforming implicit relationships in the knowledge graph into quantifiable and computable structured data, thus facilitating data adaptation between the knowledge graph and machine learning models. A hybrid dataset is built based on the standardized relation matrix, enriching the data dimensions and sample completeness for model training, optimizing the quality and adaptability of training data, and improving the effectiveness of model learning. Training a supervised learning model on a high-quality hybrid dataset allows the model to fully learn the precise association rules between fault causes and measurement points, ultimately outputting a highly adaptable and accurate dedicated fault diagnosis model. This application, by fusing multi-source fault knowledge to construct a mechanism-constrained knowledge graph, generating training samples that conform to physical logic, and combining them with weighted training using real data, enables highly reliable, interpretable, and continuously optimized fault diagnosis in zero-sample / small-sample scenarios, significantly improving the accuracy, adaptability, and engineering practicality of fault diagnosis in complex systems. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is an application environment diagram of a fault diagnosis model determination method based on fault relationship constraints for data-scarce scenarios, as described in one embodiment of this application. Figure 2 A flowchart illustrating a method for determining a fault diagnosis model based on fault relationship constraints for data-scarce scenarios, provided in an embodiment of this application; Figure 3 A flowchart illustrating an application method for a fault diagnosis model based on fault relationship constraints in a data-scarce scenario, provided as an embodiment of this application; Figure 4 A schematic diagram of the overall structure of a fault diagnosis method based on fault relationship constraints for the absence of sample conditions, provided in an embodiment of this application; Figure 5 A flowchart illustrating the generation of a hybrid dataset according to an embodiment of this application; Figure 6 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0026] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0027] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0028] The fault diagnosis model determination method based on fault relationship constraints for data-scarce scenarios provided in this application can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on another server. Terminal 102 can send multi-source information to server 104, including: fault causes, fault modes, probability of fault occurrence, and measurement points. After receiving the multi-source information, server 104 standardizes the information, maps it to nodes, and constructs a knowledge graph with causal and influence relationships as edges. The causal relationships are determined based on fault causes and measurement points; the influence relationships are determined based on fault modes and measurement points. Key information is extracted from the knowledge graph to obtain a set of fault causes and a set of measurement points. Based on the set of fault causes and the set of measurement points, a relationship matrix between fault causes and measurement points is constructed. Based on the relationship matrix, a hybrid dataset is constructed. A supervised learning model is trained on the hybrid dataset to obtain a fault diagnosis model. Server 104 can feed back the obtained fault diagnosis model to terminal 102. Furthermore, in some embodiments, the fault diagnosis model determination method based on fault relationship constraints for data-scarce scenarios can also be implemented separately by the server 104 or the terminal 102. For example, the terminal 102 can directly determine the fault diagnosis model based on fault relationship constraints for data-scarce scenarios based on multi-source information, or the server 104 can obtain multi-source information from the data storage system and determine the fault diagnosis model based on fault relationship constraints for data-scarce scenarios based on the multi-source information.

[0029] The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.

[0030] In one exemplary embodiment, such as Figure 2 As shown, a fault diagnosis model determination method based on fault relationship constraints is provided for data-scarce scenarios. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the following steps are included: S1: Obtain multi-source information; the multi-source information includes: fault cause, fault mode, probability of fault occurrence and measurement point.

[0031] S2: After standardizing the multi-source information, it is mapped to nodes, and a knowledge graph is constructed with causal and influence relationships as edges; the causal relationship is the relationship determined based on the fault cause and the measurement point; the influence relationship is the relationship determined based on the fault mode and the measurement.

[0032] S3: Extract key information from the knowledge graph to obtain a set of fault causes and a set of measurement points.

[0033] S4: Based on the set of fault causes and the set of measurement points, construct a relationship matrix between fault causes and measurement points.

[0034] S5: Based on the aforementioned relation matrix, construct a hybrid dataset.

[0035] S6: Train the supervised learning model based on the hybrid dataset to obtain the fault diagnosis model.

[0036] By implementing steps S1 to S6 above, and through the entire process from multi-source information integration, knowledge structuring, key information extraction, relationship quantification, sample generation to model training, highly reliable, interpretable, and accurate fault diagnosis can be achieved under conditions of no or small samples. This solves the problems of traditional data-driven methods that rely on a large amount of labeled data, lack mechanistic constraints, and have insufficient diagnostic accuracy and adaptability.

[0037] As an optional implementation, in step S5, a hybrid dataset is constructed based on the relation matrix, specifically including: S51: Extract the standard measurement point response pattern corresponding to a single fault based on the relationship matrix.

[0038] S52: Based on the response pattern, randomly select several fault causes from the probability of the fault occurrence to generate a candidate measurement point state vector.

[0039] S53: Perform constraint consistency correction on the state vector of the candidate measurement point and project it to the feasible solution space that satisfies the physical and logical constraints to obtain a multi-fault combination.

[0040] S54: Based on the aforementioned multiple fault combinations, calculate the probability of each measuring point occurring in all fault modes.

[0041] S55: Calculate the information entropy of the measurement points based on the probability of occurrence of each measurement point in all fault modes.

[0042] S56: Normalize the information entropy of the measurement point to obtain the normalized information entropy.

[0043] S57: Adjust the perturbation probability of the measuring point based on the normalized information entropy.

[0044] S58: Based on the perturbation probability, the multi-fault combination is perturbed to generate noisy samples.

[0045] S59: Determine if real data exists; S510: If so, the real data, the multi-fault combination, and the noisy sample are weighted and fused to obtain a hybrid dataset.

[0046] S511: If not, the multi-fault combination and the noisy sample are weighted and fused to obtain a hybrid dataset.

[0047] As an optional implementation, in step S6, the supervised learning model is trained based on the hybrid dataset to obtain a fault diagnosis model, specifically including: S61: Based on the hybrid dataset, the fault mode corresponding to the measurement point status is converted into a fault label vector.

[0048] S62: Input the measured point state into the supervised learning model to obtain the prediction result.

[0049] S63: Based on the prediction result, the fault label vector corresponding to the measurement point state, and the determined loss function, determine the loss value.

[0050] S64: Optimize the network parameters of the supervised learning model based on the loss value to obtain a fault diagnosis model.

[0051] In another exemplary embodiment of this application, such as Figure 3 As shown, a fault diagnosis model application method based on fault relationship constraints for data-scarce scenarios is provided. The application method includes: A1: Obtain the measurement status of the device to be diagnosed.

[0052] A2: Input the status of the measurement point to be diagnosed into the fault diagnosis model to obtain the corresponding fault diagnosis result; the fault diagnosis model is a model obtained based on the fault diagnosis model determination method based on fault relationship constraints for data-scarce scenarios described above.

[0053] A3: Based on the fault diagnosis results, the fault diagnosis model is continuously optimized using incremental learning or periodic retraining.

[0054] In practical applications, such as Figure 4 As shown, a fault diagnosis method based on fault relationship constraints for cases without sample conditions is provided. The method includes the following steps: Step 1: This step addresses the lack of structured data in the system by integrating FMECA analysis results and other relevant technical materials (such as design documents and maintenance manuals) to create a unified model and structured representation of fault-related information.

[0055] Specifically, multi-source information such as fault causes, fault modes, and measurement points (detection methods or local effects) are extracted from the system and standardized to eliminate naming inconsistencies and semantic redundancy. Based on this, the above information is mapped as nodes, and "causal relationships" and "influence relationships" are used as edges to construct a knowledge graph structure for fault diagnosis.

[0056] Knowledge graphs can be represented as sets of triples: ; in, For knowledge graphs; The head entity representing the atlas; Represents the relationship between entities; The tail entity represents the graph; this graph clarifies the relationships between concepts in the fault domain, providing a clear structured constraint and prior knowledge framework for subsequent relation modeling, reasoning analysis, and synthetic data generation.

[0057] Step Two: This step extracts key information from the constructed knowledge graph, filtering out entities and relationship types directly related to fault diagnosis. The focus is on extracting causal relationship edges between "fault cause - measurement point," specifically triples that satisfy the relationship type "cause," "affect," or semantic equivalence. For cases with multiple relationship types, a unified relationship mapping rule can be used. For example, semantically, "cause," "trigger," and "cause" can be uniformly categorized as "cause" to ensure data consistency.

[0058] Based on this, the entities in the knowledge graph are clearly classified and uniformly named. Abnormal items are removed, synonymous items are merged, and the causes of failures are combined with the test points to form a set. ; ; in, A set of causes for the failure; For the set of measurement points; N This represents the total number of causes of failure. M This represents the total number of measurement points.

[0059] Based on this, a relationship matrix between fault causes and measuring points is constructed to describe the impact patterns of different faults on measuring points: ; ; in, This is a matrix showing the relationship between the causes of the fault and the measuring points; For the first i The cause of the fault is related to the first j The relationship between the measurement points.

[0060] Through the above process, the complex relational structure in the original knowledge graph is transformed into a unified matrix representation, realizing the transformation from "semantic relational modeling" to "numerical computational modeling." This relational matrix not only retains the core mechanism information of fault propagation, but also has good scalability and computability, and can be directly used for subsequent sample generation, information entropy calculation, and fault diagnosis model training.

[0061] Step 3: In the absence of or with only a small amount of real-world operational data, this step, based on the fault cause-measurement point relationship matrix constructed in step two, generates training samples that satisfy the mechanism constraints. These samples support the subsequent construction of the fault diagnosis model. Because only a very small amount of historical data exists, or even none at all, unlike traditional sample construction methods that rely on historical data distribution, the data generated in this step is not used to approximate the true probability distribution, but rather to characterize the "feasible state space" that the system may exhibit under fault mechanism constraints. This ensures the logical consistency and physical interpretability of the generated samples. Figure 5 As shown, the specific process is as follows: First, the response patterns of standard measurement points corresponding to a single fault are extracted based on the relationship matrix. For any fault cause... The corresponding state vector of the measuring point is defined as: ; This vector indicates that only a fault has occurred. The ideal state of each measuring point at that time, based on the standard design parameters, can be used as a noise-free standard fault sample.

[0062] Based on this, in order to simulate the concurrent multi-fault situation in reality, it is necessary to model the complex situation. Specifically, several fault situations are randomly selected from the existing fault set. In the process of multi-fault combination, candidate measurement point state vectors are first generated based on the fault-measurement point relationship matrix. Then, a set of measurement point constraints is introduced, and the candidate samples are constrained to perform consistency correction. They are then projected to the feasible solution space that satisfies the physical and logical constraints, thereby avoiding the generation of abnormal state combinations that do not conform to the system mechanism.

[0063] ; ; in, This is a combination of multiple fault data formed by simple logical combinations; The constraints formed by the graph include eliminating all cases that do not conform to reality (such as the simultaneous occurrence of mutually exclusive measurement points). Cause of the malfunction The corresponding measurement point state vector; To constrain the data; This is the corrected sample.

[0064] Furthermore, since false alarms and missed detections often occur in reality, perturbations need to be added to the simulated data to further expand the sample space. To make the simulated perturbations more closely resemble the actual environment, uncertainty modeling of measurement points based on information entropy is introduced here. First, the probability of each measurement point occurring in all failure modes is calculated: ; in, For the first j The probability of occurrence of a measurement point in all failure modes; For the first i The cause of the fault is related to the first j The relationship between the measurement points; This represents the total number of causes of the failure.

[0065] Based on this, the information entropy of the measuring point is calculated, which is used to characterize its uncertainty: ; in, For the first j Information entropy of each measurement point.

[0066] Based on this, the information entropy is normalized and used to adjust the disturbance probability of the measuring point, thereby simulating actual uncertainty factors such as measurement error and sensor fluctuation: ; ; in, The normalized information entropy; The maximum information entropy; Based on the probability of perturbation; For the first j The probability of disturbance at each measuring point.

[0067] Subsequently, the standard samples were perturbed to generate noisy samples, thereby enhancing the model's adaptability to complex real-world conditions. The final result is a hybrid dataset containing both noise-free samples and samples with varying noise intensities.

[0068] The aforementioned data construction is based on the absence of original fault data. If a small amount of real data exists, directly fitting a probability distribution based on this data can easily introduce significant statistical bias. Therefore, this application does not rely on real data for distribution modeling, but instead introduces it as correction information into the generated data system. By weighted fusion of real and generated data, the model gradually approximates the actual distribution while maintaining mechanistic consistency, thereby improving the reliability of the diagnostic results.

[0069] Step Four: Based on the data obtained in step three, this step constructs a fault diagnosis model to achieve reverse mapping from measurement points to fault causes. Since multiple faults occur concurrently in actual systems, and there is simulated data of this type, this step uses a multi-label learning method to train the fault diagnosis model. The specific process is as follows: First, the fault modes corresponding to the data are converted into fault label vectors. For the first... A sample, defined Indicate the cause of the malfunction The corresponding measurement point state vector has a dimension of A binary vector of 0 and 1; corresponding output Indicate the cause of the malfunction The corresponding fault label is a dimension of A binary vector of 0 and 1, to obtain Samples in the form of...

[0070] Subsequently, the generated data was divided into training and testing sets to construct a supervised learning model. During the model learning process, the real data was weighted to retain the model's judgments on correct mechanisms while enhancing the model's ability to fit the real distribution. ; in, For supervised learning models; For the first k The weight of each measurement point; Cause of the malfunction The corresponding measurement point state vector; Cause of the malfunction The corresponding fault label; The loss function; This is the predicted result.

[0071] After model training, the model performance is evaluated using a test set. By comparing the consistency between the predicted results and the true labels, the accuracy and stability of the model in multiple fault diagnosis scenarios are assessed. Finally, in the practical application stage, the state vector of the measurement point to be diagnosed is input into the model to output the corresponding fault prediction result.

[0072] As real-world data accumulates, the model needs to gradually transition to a learning mode dominated by real-world data to reduce distribution bias introduced by generated data. Furthermore, for model updates, incremental learning or periodic retraining can be employed. Incremental learning uses only newly added data to locally update model parameters; periodic retraining retrains the model based on the updated complete dataset. Ultimately, this allows the fault diagnosis model to evolve from "mechanism-based initialization" to "continuous optimization based on real-world data," enabling the system to possess adaptive capabilities and long-term application value.

[0073] In summary, this application has the following advantages compared with the prior art: This application unifies and structures the knowledge related to multi-source faults in the system, constructs a computable representation that reflects the correlation between faults, introduces mechanistic constraints for sample generation and correction, and combines multi-fault modeling and uncertainty characterization mechanisms to further achieve continuous optimization and evolution of the model by integrating a small amount of real data.

[0074] Through the above methods, this application can generate effective training samples that conform to system mechanism constraints under conditions of lack or only a small amount of historical data, improve the interpretability, reliability and adaptability to complex working conditions of the fault diagnosis model, and support the gradual improvement of model performance as data accumulates.

[0075] Compared with existing technologies, this application effectively solves the problems of difficulty in utilizing prior knowledge under small sample or zero sample conditions, lack of consistency in generated data mechanisms, insufficient modeling of multiple faults and uncertainties, and lack of continuous evolution capability of models, thereby improving the engineering practicality of fault diagnosis methods for complex systems.

[0076] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 6As shown, the computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores multi-source information. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a fault diagnosis model determination method based on fault relationship constraints for data-scarce scenarios or an application method of a fault diagnosis model based on fault relationship constraints for data-scarce scenarios.

[0077] Those skilled in the art will understand that Figure 6 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0078] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0079] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0080] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Moreover, the collection, use and processing of the relevant data are carried out in compliance with the relevant data protection laws and policies of the country where the location is located, and with the authorization granted by the owner of the corresponding device.

[0081] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0082] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0083] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0084] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for determining a fault diagnosis model based on fault relationship constraints for data-scarce scenarios, characterized in that, The determination method includes: Acquire multi-source information; the multi-source information includes: fault cause, fault mode, probability of fault occurrence, and measurement points; After standardizing the multi-source information, it is mapped to nodes, and a knowledge graph is constructed with causal and influence relationships as edges; the causal relationship is the relationship determined based on the fault cause and the measurement point; the influence relationship is the relationship determined based on the fault mode and the measurement. Key information is extracted from the knowledge graph to obtain a set of fault causes and a set of measurement points; Based on the set of fault causes and the set of measurement points, construct a relationship matrix between fault causes and measurement points; Based on the aforementioned relation matrix, a hybrid dataset is constructed; The supervised learning model is trained based on the hybrid dataset to obtain a fault diagnosis model.

2. The method for determining a fault diagnosis model based on fault relationship constraints for data-scarce scenarios according to claim 1, characterized in that, The expression for the relationship matrix between the fault cause and the measuring point is: ； ； in, This is a matrix showing the relationship between the causes of the fault and the measuring points; For the first i The cause of the fault is related to the first j The relationship between the measurement points; This represents the total number of causes of failure. This represents the total number of measurement points.

3. The method for determining a fault diagnosis model based on fault relationship constraints for data-scarce scenarios according to claim 1, characterized in that, Based on the aforementioned relation matrix, a hybrid dataset is constructed, specifically including: Based on the relationship matrix, extract the standard measurement point response pattern corresponding to a single fault; Based on the response pattern, several fault causes are randomly selected from the probability of fault occurrence to generate a candidate measurement point state vector. The state vectors of the candidate measurement points are subjected to constraint consistency correction and projected onto the feasible solution space that satisfies physical and logical constraints to obtain multiple fault combinations; Based on the aforementioned multiple fault combinations, the probability of occurrence of each measuring point in all fault modes is calculated. Calculate the information entropy of each measuring point based on its occurrence probability in all fault modes; The information entropy of the measurement points is normalized to obtain the normalized information entropy; Based on the normalized information entropy, adjust the perturbation probability of the measuring point; Based on the perturbation probability, the multi-fault combination is perturbed to generate noisy samples; Determine if real data exists; If so, the real data, the multi-fault combination, and the noisy sample are weighted and fused to obtain a hybrid dataset; If not, the multi-fault combination is weighted and fused with the noisy sample to obtain a hybrid dataset.

4. The method for determining a fault diagnosis model based on fault relationship constraints for data-scarce scenarios according to claim 3, characterized in that, The formula for calculating the probability of occurrence of each measuring point in all fault modes is as follows: ； in, For the first j The probability of occurrence of a measurement point in all failure modes; For the first i The cause of the fault is related to the first j The relationship between the measurement points; This represents the total number of causes of failure. The formula for calculating the entropy of the measuring point information is: ； in, For the first j Information entropy of each measurement point; The normalization expression is: ； in, The normalized information entropy; The maximum information entropy; The adjustment formula for the disturbance probability of the measuring point is: ； in, Based on the probability of perturbation; For the first j The probability of disturbance at each measuring point.

5. The method for determining a fault diagnosis model based on fault relationship constraints for data-scarce scenarios according to claim 1, characterized in that, The supervised learning model is trained based on the aforementioned hybrid dataset to obtain a fault diagnosis model, specifically including: Based on the hybrid dataset, the fault modes corresponding to the measurement point status are converted into fault label vectors. The state of the measurement point is input into the supervised learning model to obtain the prediction result; Based on the prediction results, the fault label vector corresponding to the measurement point status, and the determined loss function, the loss value is determined; The network parameters of the supervised learning model are optimized based on the loss value to obtain a fault diagnosis model.

6. The method for determining a fault diagnosis model based on fault relationship constraints for data-scarce scenarios according to claim 5, characterized in that, The expression for the fault diagnosis model is: ； in, For supervised learning models; For the first k The weight of each measurement point; Cause of the malfunction The corresponding measurement point state vector; Cause of the malfunction Corresponding fault labels; The loss function; This is the predicted result.

7. A fault diagnosis model application method based on fault relationship constraints for data-scarce scenarios, characterized in that, The application method includes: Obtain the measurement status of the test to be diagnosed; The status of the measurement point to be diagnosed is input into the fault diagnosis model to obtain the corresponding fault diagnosis result; the fault diagnosis model is a model obtained based on the fault diagnosis model determination method based on fault relationship constraints for data-scarce scenarios as described in any one of claims 1-6. Based on the fault diagnosis results, the fault diagnosis model is continuously optimized using incremental learning or periodic retraining.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that the processor executes the computer program to implement the fault diagnosis model determination method based on fault relationship constraints for data-scarce scenarios as described in any one of claims 1-6 or the fault diagnosis model application method based on fault relationship constraints for data-scarce scenarios as described in claim 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the fault diagnosis model determination method based on fault relationship constraints for data-scarce scenarios as described in any one of claims 1-6, or the fault diagnosis model application method based on fault relationship constraints for data-scarce scenarios as described in claim 7.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the fault diagnosis model determination method based on fault relationship constraints for data-scarce scenarios as described in any one of claims 1-6, or the fault diagnosis model application method based on fault relationship constraints for data-scarce scenarios as described in claim 7.

Images