Method and system for diagnosing unknown faults of high-speed train transmission system under variable working conditions

Abstract

The invention belongs to the technical field of intelligent operation and maintenance of high-speed train transmission systems, and discloses an unknown fault diagnosis method for a high-speed train transmission system under variable working conditions, which comprises the following steps of: defining a health state of the transmission system as a domain attribute and a state category attribute, and constructing input data according to a triple; domain features and state type features are decoupled through an attribute decoupler; the two types of features are spliced and input into an automatic encoder to generate reconstruction features, and the reconstruction loss is minimized; and then decoupling features are randomly disorganized and recombined, unknown fault features are reconstructed, and the unknown fault features are separated by a decoupler to enhance the unknown fault diagnosis capability. In the online diagnosis stage, the attribute decoupler and the classifier work cooperatively, and accurate diagnosis of known and unknown faults is achieved. According to the method, the dependence of an existing method on consistent fault types in operation conditions is overcome, accurate diagnosis of unknown faults under variable working conditions is realized, and an important engineering value is provided for safe and stable operation of a high-speed train transmission system.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault diagnosis, and more specifically, relates to a method and system for diagnosing unknown faults in a high-speed train transmission system under variable operating conditions. Background Art

[0002] As the core power transmission unit of rail transit equipment, the operating status of high-speed train transmission systems directly determines the safety and reliability of the trains. In actual operation, the system is constantly exposed to variable operating conditions such as sudden speed changes, alternating loads, and fluctuations in ambient temperature and humidity. This can lead to a combination of performance degradation and sudden failures in mechanical components, significantly increasing the difficulty of fault diagnosis. Accurately diagnosing faults under complex and changing operating conditions has become a key challenge to ensure safe train operation and reduce maintenance costs.

[0003] Current mainstream transfer learning methods can improve cross-operational diagnostic capabilities to a certain extent through domain adaptation. However, their performance relies heavily on the complete consistency of fault categories across different operating conditions and requires the manual design of complex domain difference metric functions, which limits their flexibility and applicability in real-world scenarios. Meanwhile, while few-shot learning methods can accomplish diagnostic tasks using a small number of samples from a specific target operating condition, their core assumption remains closed set, making them ineffective at identifying new, unknown fault types that arise during operation. Existing methods generally implicitly assume that fault categories remain consistent across operating conditions. However, this assumption is significantly inconsistent with real-world engineering scenarios. Different operating conditions often induce novel failure modes. For example, under sustained high temperatures, gears may fail due to thermal deformation, while under high humidity, bearings are susceptible to corrosive pitting. These fault types, triggered by specific operating conditions, often exceed the scope of existing knowledge bases, making them difficult for traditional methods to address.

[0004] Therefore, in the absence of prior information about unknown faults and the presence of significant differences in operating conditions, accurately diagnosing transmission system faults has become a core technical bottleneck restricting the development of intelligent rail transit operations and maintenance. This challenge requires not only diagnostic methods that can adapt to differences in data distribution but also robustness against novel fault modes. Summary of the Invention

[0005] To address the shortcomings of existing technologies and meet practical needs, this paper proposes a method and system for diagnosing unknown faults in high-speed train transmission systems under variable operating conditions. This method aims to accurately diagnose unknown fault types in transmission systems operating under complex and variable conditions. This paper proposes a multi-domain collaborative attribute feature decoupling and reconstruction framework, which overcomes the limitations of traditional methods for diagnosing unknown fault types and significantly improves the robustness and adaptability of diagnosis.

[0006] To achieve the above objectives, the present invention provides a high-speed train transmission system unknown fault diagnosis method based on a multi-domain collaborative attribute feature decoupling and reconstruction framework under variable operating conditions, comprising:

[0007] S1. Define the health status of the train transmission system as consisting of two independent attributes: domain attributes and status category attributes. Collect known health status data under various operating conditions and construct training data based on triples, covering different combinations of domain attributes and status category attributes.

[0008] S2. Use the residual shrinkage network as a feature extractor to effectively filter out noise and extract features of high-quality training data triplets;

[0009] S3. Design an attribute decoupler to decouple the output features of the residual shrinkage network into domain features and state type features. Furthermore, construct separate classifiers to calculate the domain classification loss and state classification loss, the sum of which constitutes the total classification loss.

[0010] S4. Concatenate the decoupled domain features and state type features and input them into the autoencoder to obtain reconstructed features. Use the reconstruction loss to minimize the difference between the reconstructed features and the original features, ensuring information integrity and reducing the discriminative loss caused by decoupling.

[0011] S5. Reconstruct the unknown fault features by randomly shuffling and recombining the decoupled domain attribute features and state category attribute features, and inputting them into the autoencoder. The reconstructed unknown fault features are decoupled again by the decoupler, and the unknown fault decoupling loss is calculated. The reconstruction loss is added to the unknown fault decoupling loss to form the reconstruction-decoupling loss.

[0012] S6. Using the total attribute classification loss and reconstruction-decoupling loss as optimization targets, we employ iterative optimization techniques to jointly train the residual shrinkage network feature extractor, attribute decoupler, autoencoder, and classifier to ensure that the model converges to the optimal state.

[0013] During the online fault diagnosis phase, real-time monitoring data from the train transmission system is collected as a test set and fed into a trained residual shrinkage network feature extractor. The train transmission health status is determined by the collaborative work of the attribute decoupler and the classifier, enabling accurate diagnosis of both known and unknown faults.

[0014] Furthermore, the health status of the train transmission system is represented by the domain attribute set and state category attribute collections The set of all possible health states is expressed as Collect known health status data under various operating conditions as training data sets and record them as As A subset of , and designed as a triple The domain attributes Indicates different operating conditions. Different fault categories; to ensure the comprehensiveness and accuracy of diagnosis Must include All domain attributes in All state category attributes in ; the combination set of input triples is defined as

[0015] Furthermore, the attribute decoupler is designed to decouple the domain attribute features and the state type attribute features. and state class attribute decoupler At the same time, the classifier designs the corresponding domain attribute classifier and state category attribute classifier

[0016] Furthermore, the decoupled domain attribute features and state category attribute characteristics Splice into combined features and input into the autoencoder In the above example, we generate reconstruction features Its mathematical expression is

[0017] Furthermore, the decoupled features are randomly shuffled to generate new domain attribute decoupled features Decoupling features from state category attributes The shuffle operation is represented as By reconstructing unknown fault characteristics Decouple again and identify domain and state category attributes, and calculate the unknown fault decoupling loss Defined as

[0018] Furthermore, the test set is denoted as That is, the test phase includes all known and unknown fault types in terms of domain attributes and state category attributes; the fault type prediction is completed by the state category decoupler and the state category classifier, and the operating condition prediction is achieved by the domain decoupler and the domain classifier.

[0019] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art.

[0020] Existing intelligent diagnosis methods under variable operating conditions generally face a key bottleneck: they usually require the dataset to cover fault categories in all domains, even if some methods do not have strict requirements for labeled data. For example, in mechanical fault diagnosis under different operating conditions, the dataset for each condition must contain all possible fault types. However, this assumption is often difficult to hold in industrial practice. When machines operate for a long time in harsh or variable environments, the fault manifestations vary significantly - certain conditions may trigger specific faults, while other conditions may lead to completely different failure modes. As the complexity of the operating environment increases, the distribution of fault categories becomes increasingly unpredictable, and it is almost impossible to construct a complete and consistent dataset for each operating condition. Therefore, a large number of unknown fault categories often appear in cross-domain diagnosis, which poses a severe challenge to the adaptability of traditional methods. To address this problem, the present invention proposes a method for unknown fault diagnosis of high-speed train transmission systems under variable operating conditions based on a multi-domain collaborative attribute feature decoupling and reconstruction framework. This framework achieves efficient feature separation by decoupling multi-domain fault states into two independent parts: domain attributes and state category attributes. Domain attributes focus on capturing the unique characteristics of operating conditions, while state category attributes focus on extracting the inherent features of fault types, thereby ensuring diagnostic accuracy and cross-domain adaptability. The framework introduces a reconstruction-decomposition mechanism, which uses the decoupled domain attributes and state category attributes to reconstruct the original data in a subspace to ensure information integrity. Subsequently, by randomly shuffling and recombining these attribute features, synthetic unknown fault features are generated, reducing the model's dependence on scarce fault data. This improves the performance of unknown fault diagnosis. In summary, the present invention, with its unique attribute decoupling and feature recombination technology, provides an efficient and reliable solution for unknown fault diagnosis under variable operating conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] To make the technical details of this technical solution clearer and easier to understand, the following briefly describes the basic details of the relevant drawings of this technical solution. It should be noted that the drawings shown here are only schematic diagrams of some typical embodiments of this technology. Those skilled in the relevant art can deduce the supporting drawings required for other implementation methods based on the illustrations without any creative work.

[0022] Figure 1 Schematic diagram of a multi-domain collaborative attribute feature decoupling and reconstruction framework provided by the present invention;

[0023] Figure 2 yes Figure 1 Schematic diagram of the reconstruction-decoupling mechanism involved in the multi-domain collaborative attribute feature decoupling and reconstruction framework; DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solutions and advantages of the present invention clearer and more thorough, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be emphasized that the specific embodiments described herein are only used to illustrate the present invention and do not limit its scope. In addition, the technical features involved in each embodiment can be flexibly combined to give full play to their synergistic effects, provided that they do not conflict with each other.

[0025] See also Figure 1 and Figure 2 The present invention provides a high-speed train transmission system unknown fault diagnosis method based on a multi-domain collaborative attribute feature decoupling and reconstruction framework. Figure 1 As shown, the present invention provides a framework for decoupling and reconstructing multi-domain collaborative attribute features, and the specific steps are as follows:

[0026] Step 1: The health status of the train transmission system is defined as consisting of two independent attributes: domain attribute (characterizing operating conditions) and state category attributes (Characterize the fault type). By collecting known health status data under various operating conditions, a training data set is constructed and a triplet design is used. Ensure that the data covers a combination of different domain attributes and state category attributes.

[0027] Specifically, the health status of the train transmission system is represented by the domain attribute set and state category attribute collections The set of all possible health states is expressed as Collect known health status data under various operating conditions as training data sets and record them as As A subset of , and designed as a triple The domain attributes Indicates different operating conditions. Different fault categories. For example, It can reflect the running status of the train at different speeds. This triple design significantly improves the diversity and representativeness of data by integrating multi-domain and multi-category information. Its input combination set is defined as To ensure the comprehensiveness and accuracy of the diagnosis Must include All domain attributes in All state category attributes in .

[0028] Step 2: Use the residual shrinkage network as a feature extractor to extract features from the triplet training data And effectively filter out noise interference in the train transmission system monitoring data.

[0029] Specifically, the residual shrinkage network uses its unique soft threshold function to reduce the noise of the input data. Through a carefully designed threshold filtering mechanism, the noise components hidden in the feature map are gradually approached to zero. The feature extraction process is expressed as:

[0030]

[0031] Where x i is the input data, ξ and δ are the learned weight and bias parameters respectively, and δ is the ReLU activation function.

[0032] Step 3: Design domain attribute decoupler and state class attribute decoupler Separate the operating condition information from the fault type information in the feature. Input attribute decoupler to generate independent domain attribute features and state type attribute characteristics Then, the decoupled attribute features are input into the domain attribute classifier and state category attribute classifier Compute domain classification loss and state category classification loss And the total attribute classification loss formed by adding the two

[0033] Specifically, the extracted features Directly input the decoupler, the calculation process is as follows:

[0034]

[0035]

[0036] In the formula and yes Domain Property Decoupler and state class attribute decoupler The decoupled features are then fed into the domain attribute classifier and state category attribute classifier Forming domain classification loss and state classification loss is defined as:

[0037]

[0038]

[0039] In the formula and is a domain attribute classifier and state category attribute classifier Parameters, and is the true domain attribute and category attribute label, I(·) is the cross entropy loss function. The total classification loss Expressed as:

[0040]

[0041] Step 4, Figure 2 As shown, the design of reconstruction-decoupling mechanism will decouple the domain attribute characteristics and state category attribute characteristics Splice into combined features and input into the autoencoder Generate reconstruction features By reconstruction loss Minimize the difference between the reconstructed features and the original features to ensure information integrity and make up for the possible loss of discriminability caused by decoupling.

[0042] Specifically, the combined features after splicing are input into the autoencoder Generate reconstruction features

[0043]

[0044] Then, the reconstruction loss function is defined as:

[0045]

[0046] By minimizing the difference between the reconstructed features and the original coupled features, the final output is ensured to closely match the input combination.

[0047] Step 5: Design a feature reorganization strategy by randomly disrupting and reorganizing the decoupled domain attribute features and state category attribute characteristics Input to the autoencoder Reconstructing and synthesizing unknown fault characteristics This improves the model’s generalization ability for unknown faults. Next, the unknown fault features are reconstructed. Perform decoupling again and calculate the decoupling loss of unknown faults and decouple losses from unknown faults Add together to form the reconstruction-decoupling loss

[0048] Specifically, the decoupled features are randomly shuffled to generate new domain attribute decoupled features Decoupling features from category attributes The shuffle operation is expressed as:

[0049]

[0050] After splicing, input the autoencoder Generate reconstruction features:

[0051]

[0052] By reconstructing the features Decouple again and identify domain and category attributes to further ensure the combinability and reversibility of decoupling features and calculate the unknown fault decoupling loss Defined as:

[0053]

[0054] The overall reconstruction-decoupling loss is defined as

[0055]

[0056] Step 6: Classify the loss based on total attributes and reconstruction-decoupling loss As an optimization goal Iterative optimization technology is used to jointly train the residual shrinkage network feature extractor, attribute decoupler and autoencoder to ensure that the model converges to the optimal state.

[0057] Specifically, the optimization goal is:

[0058]

[0059] Step 7. In the online fault diagnosis phase, real-time monitoring data of the train transmission system is collected to form a test set. in That is, the test phase includes all possible attribute combinations, a wider range than the training set. This is fed into the trained residual shrinkage network feature extractor. Fault type prediction is accomplished through the collaborative work of the state category attribute decoupler and the state category attribute classifier. Operating condition prediction is accomplished through the collaboration of the domain attribute decoupler and the domain attribute classifier, enabling accurate diagnosis of both known and unknown faults.

[0060] To highlight the advantages of our method for unknown fault diagnosis, we conducted comparative experiments with several advanced methods, including deep convolutional neural networks, domain adversarial neural networks (DANs) and deep subdomain adaptation networks (DSANs) in transfer learning, and matching networks and relational networks in few-shot learning. Table 1 presents the comparative results of these methods in terms of unknown fault diagnosis accuracy, demonstrating that our method significantly outperforms the other five methods in terms of diagnostic accuracy.

[0061] Table 1

[0062] It will be easily understood by those skilled in the art that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for diagnosing unknown faults in a high-speed train transmission system under variable operating conditions, characterized in that: include: S1. The health status of the train transmission system is defined as consisting of two independent attributes: domain attribute and status category attribute; Collect known health status data under various operating conditions and construct training data based on triples, which must cover combinations of different domain attributes and status category attributes; S2. Use the residual shrinkage network as a feature extractor to effectively filter out noise and extract features of high-quality training data triplets; S3. Design an attribute decoupler to decouple the output features of the residual shrinkage network from domain features and state type features; And construct each classifier to calculate the domain classification loss and state classification loss, the sum of the two constitutes the total classification loss; S4. Concatenate the decoupled domain features and state type features and input them into the autoencoder to obtain reconstructed features. Use the reconstruction loss to minimize the difference between the reconstructed features and the original features, ensuring information integrity and reducing the discriminative loss caused by decoupling. S5. Randomly shuffle and reorganize the decoupled domain attribute features and state category attribute features, and input them into the autoencoder to reconstruct the unknown fault features; The reconstructed unknown fault features are decoupled again by the decoupler to calculate the unknown fault decoupling loss; the reconstruction loss and the unknown fault decoupling loss are added to form the reconstruction-decoupling loss; S6. Using the total attribute classification loss and reconstruction-decoupling loss as optimization targets, we employ iterative optimization techniques to jointly train the residual shrinkage network feature extractor, attribute decoupler, autoencoder, and classifier to ensure that the model converges to the optimal state. During the online fault diagnosis phase, real-time monitoring data from the train transmission system is collected as a test set and fed into a trained residual shrinkage network feature extractor. The train transmission health status is determined by the collaborative work of the attribute decoupler and the classifier, enabling accurate diagnosis of both known and unknown faults.

2. The method for diagnosing unknown faults in a train transmission system under variable operating conditions according to claim 1, characterized in that: The health status of the train transmission system is represented by the domain attribute set Y d and state category attribute set Y c The set of all possible health states is expressed as Collect known health status data under various operating conditions as training data sets and record them as As A subset of , and designed as a triple The domain attributes Indicates different operating conditions. Different fault categories; to ensure the comprehensiveness and accuracy of diagnosis Must contain Y d All domain attributes in Y c The combination set of all state category attributes in the input triples is defined as 3. The method for diagnosing unknown faults of a high-speed train transmission system under variable operating conditions according to claim 1 is characterized in that: Attribute decoupler φ is designed for the decoupling of domain attribute features and state type attribute features. d (·) and the state-category-attribute decoupler φ c (·), and the classifier designed the corresponding domain attribute classifier and state category attribute classifier 4. The method for diagnosing unknown faults in a high-speed train transmission system under variable operating conditions according to claim 1 is characterized in that: The decoupled domain attribute features and state category attribute characteristics Splice into combined features and input into the autoencoder In the above example, we generate reconstruction features Its mathematical expression is 5. The method for diagnosing unknown faults of a high-speed train transmission system under variable operating conditions according to claim 1 is characterized in that: Randomly shuffle the decoupled features to generate new domain attribute decoupled features Decoupling features from state category attributes The shuffle operation is represented as By reconstructing unknown fault characteristics Decouple again and identify domain and state category attributes, and calculate the unknown fault decoupling loss Defined as 6. The method for diagnosing unknown faults in a high-speed train transmission system under variable operating conditions according to claim 1, characterized in that: The test set is denoted as That is, the test phase includes all known and unknown fault types in terms of domain attributes and state category attributes; the fault type prediction is completed by the state category decoupler and the state category classifier, and the operating condition prediction is achieved by the domain decoupler and the domain classifier.

7. A high-speed train transmission system unknown fault diagnosis system under variable working conditions, characterized by: include: In the dataset definition and design unit, the health status of the train transmission system is defined as consisting of two independent attributes: domain attribute and status category attribute; Collect known health status data under various operating conditions and construct training data based on triples, which must cover combinations of different domain attributes and status category attributes; The model building unit uses a residual shrinkage network as a feature extractor to effectively filter out noise and extract features of high-quality training data triplets; Design an attribute decoupler to decouple the output features of the residual shrinkage network into domain features and state type features. Separate classifiers are constructed to calculate the domain classification loss and state classification loss, the sum of which constitutes the total classification loss. The decoupled domain features and state type features are concatenated and input into an autoencoder to obtain reconstructed features. The reconstruction loss is used to minimize the difference between the reconstructed features and the original features, ensuring information integrity and mitigating the discriminative loss caused by decoupling. The decoupled domain attribute features and state category attribute features are randomly shuffled and reorganized, and then input into the autoencoder to reconstruct unknown fault features. The reconstructed unknown fault features are decoupled again by the decoupler to calculate the unknown fault decoupling loss; the reconstruction loss and the unknown fault decoupling loss are added to form the reconstruction-decoupling loss; The model optimization unit uses the total attribute classification loss and reconstruction-decoupling loss as optimization targets and adopts iterative optimization technology to jointly train the residual shrinkage network feature extractor, attribute decoupler, autoencoder and classifier to ensure that the model converges to the optimal state; The online diagnosis unit collects real-time monitoring data of the train transmission system as a test set and inputs it into the trained residual shrinkage network feature extractor; the train transmission health status is completed by the collaborative work of the attribute decoupler and classifier, thereby achieving accurate diagnosis of known and unknown faults.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that When the program is executed by the processor, the method for diagnosing unknown faults of a high-speed train transmission system under variable working conditions as described in any one of claims 1 to 6 is implemented.

Images