Method, system and device for predicting residual life of bearing of gear box of motor train unit under condition of incomplete data

Abstract

This invention discloses a method, system, and device for predicting the remaining life of EMU gearbox bearings under incomplete data conditions. It extracts degradation features from multi-source domain full-life degradation data and target domain non-full-life degradation data; designs a multi-domain local adversarial learning strategy to align the degradation stages and distribution characteristics of the source domain full-life degradation features and the target domain non-full-life degradation features, accurately mining the common degradation information of the two types of degradation features, and thus constructing a high-precision and highly generalizable predictor; using the total training loss function as the optimization objective, a well-trained multi-domain local adversarial learning remaining life prediction model is obtained, thereby achieving accurate prediction of the bearing's remaining life. This invention can achieve effective feature alignment between source domain full-life degradation data and target domain non-full-life degradation data with different degradation stages and distribution characteristics, thereby improving the accuracy and generalization of EMU gearbox bearing remaining life prediction under incomplete data conditions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bearing life prediction, and in particular to a method, system and device for predicting the residual life of a high-speed train gearbox bearing under incomplete data conditions. BACKGROUND

[0002] As a key component of a rotating mechanical system, the running state of a high-speed train gearbox bearing directly affects the safety and reliability of the train and the economy of maintenance. Carrying out accurate residual life prediction is an important prerequisite for predictive maintenance. However, in actual service scenarios, it is difficult to obtain degradation data covering the entire life cycle of the high-speed train gearbox bearing during operation, and most of the degradation data is non-full-life degradation data, which leads to the failure of existing residual life prediction methods based on full-life degradation data. Therefore, it is necessary to use full-life degradation data (such as test bench data or historical data on the line) for model training. However, due to the differences between the accelerated degradation test conditions of the test bench and the actual line operation environment, as well as the differences in working conditions of the line historical full-life degradation data of different vehicle types and operation environments, the test bench full-life degradation data, the line historical full-life degradation data and the line real-time degradation data present significant inconsistency in the degradation stage and distribution characteristics, which makes it easy to cause negative transfer and reduce the prediction accuracy and generalization ability of the model by directly aligning the features using existing transfer learning methods, thereby affecting the reliability of the intelligent operation and maintenance system.

[0003] Therefore, the existing residual life prediction method has the following technical limitations: 1) The existing method generally relies on complete full-life degradation data for model training, and cannot be compatible with full-life degradation data and non-full-life degradation data with different degradation stages and distribution characteristics, which limits the accuracy and generalization of residual life prediction, and seriously affects its applicability in real service scenarios; 2) The transfer learning method used only focuses on solving the feature alignment problem caused by the difference in distribution characteristics, and fails to effectively handle the inconsistency in the degradation stage caused by the difference in data integrity, ignoring some valuable degradation information.

[0004] Under this background, it is urgent to develop a residual life prediction method for high-speed train gearbox bearings under incomplete data conditions, which can fully exploit the public degradation information in non-full-life degradation data and full-life degradation data, thereby achieving high-precision and strong-generalization residual life prediction of railway vehicle gearbox bearings. SUMMARY

[0005] The present application provides a residual life prediction method, system and device for high-speed train gearbox bearings under incomplete data conditions to solve the above technical problems existing in the existing residual life prediction method.

[0006] According to the first aspect, one embodiment provides a method for predicting the remaining life of gearbox bearings in high-speed trains under conditions of incomplete data, the method comprising:

[0007] The vibration signals of the gearbox bearings of EMUs under multiple operating conditions throughout their lifespan and non-lifespan were collected and preprocessed. Based on the lifespan degradation, the data were divided into multi-source domain datasets and target domain datasets.

[0008] Based on the feature extraction network, degradation features are extracted from the full-lifetime degradation data of each source domain and the non-full-lifetime degradation data of the target domain, so as to obtain the full-lifetime degradation features of each source domain and the non-full-lifetime degradation features of the target domain.

[0009] We design a multi-domain local adversarial learning strategy. By introducing a weighted domain discriminator and an adversarial domain discriminator, we can align the degradation stages and distribution characteristics of the full-lifetime degradation features of each source domain and the non-full-lifetime degradation features of the target domain, and mine the common degradation information that combines the source domain and the target domain.

[0010] A predictor is constructed using supervised learning by leveraging the full-lifetime degradation features of each source domain that integrate common degradation information and the corresponding real labels of remaining lifetime.

[0011] Using the total training loss function as the optimization objective, the network parameters of the feature extraction network, weight domain discriminator, adversarial domain discriminator, and predictor are adjusted until a well-trained multi-domain local adversarial learning remaining lifetime prediction model is obtained.

[0012] The trained multi-domain local adversarial learning remaining lifetime prediction model is used to test the non-full lifetime degradation data of the target domain to obtain the remaining lifetime results.

[0013] Furthermore, vibration signals of the EMU gearbox bearings throughout their entire lifespan and those outside their entire lifespan under multiple operating conditions were collected and preprocessed. Based on the lifespan degradation, the data was divided into a multi-source domain dataset and a target domain dataset, specifically including:

[0014] Vibration signals of the gearbox bearings of the EMU under multiple operating conditions throughout / incomplete life were collected at fixed sampling intervals, and the vibration signal obtained each time was used as a sample.

[0015] Z-Score standardization was used to process each sample to eliminate differences in statistical characteristics across different time scales;

[0016] Several preprocessed sample segments are used to construct a full-lifetime / non-full-lifetime degradation dataset, and then divided into multi-source domain datasets based on the full-lifetime degradation status. and target domain dataset D T , where N S Indicates the number of source domains;

[0017] The multi-source domain dataset contains test bench full-lifetime degradation data and line historical full-lifetime degradation data under different operating conditions, covering the complete degradation process from the initial early fault to complete failure. Let the j-th source domain full-lifetime degradation dataset be denoted as... This represents the i-th source domain sample. For its corresponding true remaining lifespan label This represents the total number of samples in the source domain.

[0018] The target domain dataset includes real-time line degradation data, covering only the degradation process from the onset of an early fault to the current moment, and lacks degradation labels. This target domain non-full-lifetime degradation dataset is denoted as... Let n represent the i-th target domain sample. T For the target domain sample.

[0019] Furthermore, based on the feature extraction network, degradation features are extracted from the full-lifetime degradation data of each source domain and the non-full-lifetime degradation data of the target domain, obtaining the full-lifetime degradation features of each source domain and the non-full-lifetime degradation features of the target domain, specifically including:

[0020] Select N S Each Transformer encoder, with identical structure but independent parameters, serves as a feature extractor. Each Transformer encoder includes an input embedding layer, position encoding, a multi-head self-attention mechanism layer, a feedforward neural network, and layer normalization, enabling the extraction of multi-scale temporal features.

[0021] For each source domain j∈{1,2,…,N S}, N S The j-th Transformer encoder is input in parallel with the corresponding source domain full-lifetime degradation data and target domain non-full-lifetime degradation data to automatically learn the degradation patterns of the source and target domain data, thereby obtaining the degradation features of the two types of data respectively. The specific feature extraction process is as follows:

[0022]

[0023] Among them, Transformer j (·) represents the j-th Transformer encoder. and These are the degradation features extracted from the source and target domains respectively under the j-th Transformer encoder.

[0024] Furthermore, the alignment of the degradation stages specifically includes:

[0025] The matching degree between the full-lifetime degradation features of each source domain and the non-full-lifetime degradation features of the target domain at the degradation stage is evaluated using a weighted domain discriminator, and transferable weights are generated to selectively identify the full-lifetime degradation features of the source domain that are most relevant to the current degradation stage of the non-full-lifetime degradation features of the target domain. The specific steps include:

[0026] Targeting the non-full-lifetime degradation characteristics of the target domain and N S Based on the lifetime degradation characteristics of each source domain, construct N S A weighted domain discriminator;

[0027] During training, each source domain lifetime degradation feature is labeled as a positive class, and the target domain non-lifetime degradation feature is labeled as a negative class. The corresponding weight domain discriminator is then trained under supervision using a binary cross-entropy loss function. The loss function for the j-th weight domain discriminator is defined as follows:

[0028]

[0029] Among them, D j (·) represents the discriminator of the j-th weight domain; E(·) represents the expected value of the distribution function; Represents the lifetime degradation characteristics of the j-th source domain. It follows its corresponding distribution p(S); Represents the non-lifetime degradation characteristics of the target domain. It follows its corresponding distribution p(T);

[0030] In each training batch, the constructed weighted domain discriminator is used to calculate the confidence scores of each source domain's full-lifetime degradation feature and the target domain's non-full-lifetime degradation feature belonging to their respective domains. These confidence scores are then converted into transferable weights, providing a quantitative basis for degradation stage alignment. Specifically, this is expressed as follows:

[0031]

[0032]

[0033] in, For the lifetime degradation characteristics of the j-th source domain The confidence level of being identified as the source domain; Non-full-lifetime degradation characteristics of the target domain The confidence level of being identified as the target domain; n Y The target domain samples; ε is a constant; transferable weights It is always between (0,1). The closer it is to 1, the closer the degradation stage of the source domain's full-lifetime degradation feature and the target domain's non-full-lifetime degradation feature are. At this time, the source domain's full-lifetime degradation feature has a high transfer value. Conversely, it indicates that there is a large difference between the two degradation stages.

[0034] For each source domain j∈{1,2,…,N S}, N S The number of source domains is represented. The transferable weights are L2 regularized, and the regularization term is defined as follows:

[0035]

[0036] Here, ⊙ represents element-wise multiplication. This represents the total number of samples in the source domain.

[0037] Based on the L2 regularized transferable weights, targeted screening of the lifetime degradation features of each source domain is performed. Source domain lifetime degradation features that are at the same degradation stage as the non-lifetime degradation features of the target domain are strengthened, while source domain lifetime degradation features that are at different degradation stages from the non-lifetime degradation features of the target domain are suppressed, thereby achieving alignment of degradation stages. The specific calculation process is as follows:

[0038]

[0039] in, Let be the transferable weight after L2 regularization for the j-th time.

[0040] Furthermore, the alignment of the distribution characteristics specifically includes:

[0041] An adversarial domain discriminator is introduced to map the full-lifetime degradation features of each source domain after degradation stage alignment to the non-full-lifetime degradation features of the target domain into the same feature space. This facilitates the effective transmission and fusion of the two types of degradation features under different distribution characteristics, thereby mining common degradation information that combines the features of each source domain and the target domain. The specific steps include:

[0042] Construct N S Each adversarial discriminator distinguishes between source domain degradation features and target domain degradation features aligned during the degradation stage, and calculates the discriminative loss of each adversarial discriminator based on the binary cross-entropy loss function.

[0043]

[0044] in, Let E(·) represent the j-th adversarial discriminator; E(·) represents the expected value of the distribution function. This represents the lifetime degradation characteristics of the source domain after alignment at the j-th degradation stage. It follows its corresponding distribution p(S); Represents the non-lifetime degradation characteristics of the target domain. It follows its corresponding distribution p(T);

[0045] Meanwhile, a gradient reversal layer is introduced between the feature extractor and the adversarial domain discriminator to achieve adversarial learning, which enables the feature extractor to learn the common degradation information of each source domain and target domain, so as to further achieve feature alignment between each source domain and target domain.

[0046] Furthermore, by utilizing the lifetime degradation features of each source domain that integrate common degradation information and the corresponding real labels of remaining lifetime, a predictor is constructed using supervised learning, specifically including:

[0047] Construct N S There are 1 predictor; for each source domain j∈{1,2,…,N} S}, N S The number of source domains is represented by the degraded features after fusing common degradation information, which are then input into the corresponding predictor to obtain the remaining lifetime prediction value. The mean squared error loss function is then used to calculate the deviation between the remaining lifetime prediction value and the true remaining lifetime label, which serves as the prediction loss for each predictor. The specific formula is as follows:

[0048]

[0049] in, and These represent the remaining lifetime prediction and the actual label value of the j-th predictor, respectively. This represents the total number of samples in the source domain.

[0050] Furthermore, using the total training loss function as the optimization objective, the network parameters of the feature extraction network, weight domain discriminator, adversarial domain discriminator, and predictor are adjusted until a well-trained multi-domain local adversarial learning remaining lifetime prediction model is obtained, specifically including:

[0051] Through N in the fusion model S The total training loss function is obtained by taking the discriminator loss in the weight domain, the L2 regularization loss, the adversarial discriminator loss, and the predictor loss.

[0052]

[0053] Where λ, β, and γ are the discriminator losses in the weight region, respectively. L2 regularization loss Adversarial domain discriminator loss hyperparameters, The predictor loss is used for each source domain j∈{1,2,…,N}. S}, N S Indicates the number of source domains;

[0054] With the goal of minimizing the total training loss function, the parameters of the remaining lifetime prediction model are iteratively adjusted using the stochastic gradient descent algorithm until the training error converges, thus obtaining a well-trained prediction model.

[0055] Furthermore, the trained multi-domain local adversarial learning remaining lifetime prediction model is tested on non-full lifetime degradation data in the target domain to obtain remaining lifetime results, specifically including:

[0056] The non-full-lifetime degradation data of the target domain is input into a trained multi-domain local adversarial learning remaining lifetime prediction model to obtain N. S The output of the predictor The mean value is taken as the final remaining lifetime result:

[0057]

[0058] in, Let j be the output of the j-th predictor, j∈{1,2,…,N} s}

[0059] According to the second aspect, one embodiment provides a system for predicting the remaining life of gearbox bearings in high-speed trains under conditions of incomplete data, the system comprising:

[0060] The data acquisition and preprocessing module is used to collect and preprocess the full-life vibration signals and non-full-life vibration signals of the EMU gearbox bearings under multiple operating conditions, and divides them into multi-source domain datasets and target domain datasets according to the full-life degradation.

[0061] The feature extraction module is used to extract degradation features from the full-lifetime degradation data of each source domain and the non-full-lifetime degradation data of the target domain based on the feature extraction network, so as to obtain the full-lifetime degradation features of each source domain and the non-full-lifetime degradation features of the target domain.

[0062] The multi-domain local adversarial learning module is used to design multi-domain local adversarial learning strategies. By introducing a weight domain discriminator and an adversarial domain discriminator, it achieves the alignment of degradation stages and distribution characteristics of the full-lifetime degradation features of each source domain and the non-full-lifetime degradation features of the target domain, and mines the common degradation information that combines the source domain and the target domain.

[0063] The predictor building module is used to construct a predictor by using the full-lifetime degradation features of each source domain that integrate common degradation information and the corresponding real labels of remaining lifetime in a supervised learning manner.

[0064] The model training module is used to adjust the network parameters of the feature extraction network, weight domain discriminator, adversarial domain discriminator and predictor with the total training loss function as the optimization objective, until a trained multi-domain local adversarial learning remaining lifetime prediction model is obtained.

[0065] The remaining lifetime prediction module is used to test the non-full lifetime degradation data of the target domain using a trained multi-domain local adversarial learning remaining lifetime prediction model to obtain the remaining lifetime result.

[0066] According to a third aspect, one embodiment provides an electronic device, the device comprising: a processor and a memory;

[0067] The memory is used to store one or more program instructions;

[0068] The processor is configured to run one or more program instructions to perform the steps of a method for predicting the remaining life of a train gearbox bearing under incomplete data conditions as described in any of the preceding claims.

[0069] According to a fourth aspect, one embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of a method for predicting the remaining life of a train gearbox bearing under incomplete data conditions as described in any of the preceding claims.

[0070] This invention provides a method, system, and device for predicting the remaining life of gearbox bearings in high-speed trains under conditions of incomplete data, which has the following beneficial effects:

[0071] (1) This invention proposes a method for predicting the remaining life of gearbox bearings in high-speed trains under conditions of incomplete data. Existing remaining life prediction methods rely on full-life degradation data. This invention achieves high-precision and strong generalization ability remaining life prediction by deeply mining and fully activating the potential value of non-full-life degradation data and effectively combining the complementary advantages of non-full-life degradation data and full-life degradation data.

[0072] (2) This invention proposes a multi-domain local adversarial learning method. Existing feature alignment methods mainly address situations where there are differences in distribution characteristics, and cannot effectively solve the problem of inconsistent degradation stages. This invention combines degradation stage alignment and distribution characteristic alignment to achieve accurate matching of full-lifetime degradation features and non-full-lifetime degradation features in the local degradation stage, as well as effective transmission and fusion of source domain degradation features and target domain degradation features. This allows for the precise mining of common degradation information between source domain full-lifetime degradation data and target domain non-full-lifetime degradation data, and the capture of common degradation patterns. Attached Figure Description

[0073] Figure 1 A flowchart of a method for predicting the remaining life of a high-speed train gearbox bearing under conditions of incomplete data, provided as an embodiment of the present invention;

[0074] Figure 2A flowchart of a multi-domain local adversarial learning algorithm in a method for predicting the remaining life of a high-speed train gearbox bearing under incomplete data conditions, provided in an embodiment of the present invention;

[0075] Figure 3 This is a schematic diagram illustrating the alignment and distribution characteristics of the j-th source domain full-life degradation data and the target domain non-full-life degradation data during the degradation stage in a method for predicting the remaining life of a train gearbox bearing under incomplete data conditions, provided in an embodiment of the present invention.

[0076] Figure 4 A schematic diagram of the Transformer encoder network structure in a method for predicting the remaining life of a gearbox bearing in a high-speed train under incomplete data conditions, provided in an embodiment of the present invention.

[0077] Figure 5 A schematic diagram of a system for predicting the remaining life of a train gearbox bearing under conditions of incomplete data, provided in an embodiment of the present invention;

[0078] Figure 6 This is a schematic diagram of the structure of an electronic device provided in one embodiment of the present invention. Detailed Implementation

[0079] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings. Similar elements in different embodiments are referred to by associated similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of the invention. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, certain operations related to the present invention are not shown or described in the specification. This is to avoid obscuring the core parts of the invention with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.

[0080] Furthermore, the features, operations, or characteristics described in the specification can be combined in any suitable manner to form various embodiments. At the same time, the steps or actions in the method description can be rearranged or adjusted in a manner obvious to those skilled in the art. Therefore, the various orders in the specification and drawings are only for the clear description of a particular embodiment and do not imply a necessary order, unless otherwise stated that a particular order must be followed.

[0081] As a core component of the rotating machinery of a high-speed train, the operating status of gearbox bearings directly affects the safety and maintenance costs of the vehicle. Therefore, effective prediction of the remaining life of gearbox bearings in high-speed trains is crucial for achieving predictive maintenance. However, for gearbox bearings in actual high-speed trains, the degradation data is not full-life data, meaning it only covers part of the degradation stages and is insufficient to support the prediction of remaining life. Therefore, it is necessary to introduce full-life degradation data (such as test bench data or historical track data) into the model training. However, due to differences in train models and operating conditions, the degradation stages of non-full-life degradation data and full-life degradation data are not consistent, and their distribution characteristics differ significantly. If existing transfer learning methods are still used for feature alignment, it is easy to induce negative transfer, which will reduce the prediction effect and affect the reliability of intelligent operation and maintenance.

[0082] To address the above problems, the first embodiment of this invention provides a method for predicting the remaining life of EMU gearbox bearings under incomplete data conditions. This method aims to solve the problem that there are significant differences in degradation stages and distribution characteristics between source domain full-life degradation data and target domain non-full-life degradation data. It mainly consists of six steps, including data acquisition and preprocessing, feature extraction, multi-domain local adversarial learning, predictor construction, model training, and remaining life prediction. The following is a detailed explanation... Figure 1 Please provide a detailed explanation.

[0083] S1. Considering the time-varying operating conditions of the gearbox bearings in high-speed trains, full-life vibration signals and non-full-life vibration signals under multiple operating conditions are collected and preprocessed to form full-life / non-full-life degradation datasets. The full-life degradation data of the gearbox bearings includes full-life degradation data from the test bench under different operating conditions and historical full-life degradation data from the railway line, serving as a multi-source domain dataset; the non-full-life degradation data of the gearbox bearings is real-time degradation data from the railway line, serving as a target domain dataset.

[0084] Specifically, step S1 includes the following steps:

[0085] S1.1: Collect vibration signals of the EMU gearbox bearings under multiple operating conditions at fixed sampling intervals, covering both full-life and non-full-life conditions. Each vibration signal is collected as a sample and stored as a CSV file according to the sampling time. The file names use consecutive numerical sequences to indicate the sampling order, i.e., 1.csv, 2.csv, ..., n.csv, where n is the total number of samples.

[0086] S1.2: Since the gearbox bearings of high-speed trains are in a non-stationary state for a long time during actual service, Z-Score standardization is used to process each sample to eliminate the differences in statistical characteristics at different time scales.

[0087] S1.3: The samples processed in S1.2 are used to construct a full-lifetime / non-full-lifetime degradation dataset, and then divided into multi-source domain datasets based on the full-lifetime degradation status. and target domain dataset D t N s Indicates the number of source domains;

[0088] The multi-source domain dataset includes test bench lifecycle degradation data under different operating conditions and historical line lifecycle degradation data, covering the complete degradation process from early failure to complete failure. Let the j-th source domain lifecycle degradation dataset be denoted as... This represents the i-th source domain sample. This corresponds to the actual remaining lifespan label.

[0089] The target domain dataset consists of real-time line degradation data, covering only the degradation process from the onset of an early fault to the current moment, and lacks degradation labels. Let the target domain non-full-lifetime degradation dataset be denoted as... Let i represent the i-th target domain sample.

[0090] S2, based on the Transformer encoder network, extracts degradation features from the full-lifetime degradation data of each source domain and the non-full-lifetime degradation data of the target domain, and obtains the full-lifetime degradation features of each source domain and the non-full-lifetime degradation features of the target domain.

[0091] Specifically, step S2 includes the following steps:

[0092] S2.1: In order to extract sufficient feature information from the degradation data of the gearbox bearings of the high-speed train, N is selected. s We use identical but parameter-independent Transformer encoders as feature extractors. For each source domain j∈{1,2,…,N} S The corresponding source domain full-lifetime degradation data and target domain non-full-lifetime degradation data are input into the j-th Transformer encoder in parallel to automatically learn the degradation patterns of the source and target domain data, thereby obtaining the degradation features of the two types of data respectively. The specific feature extraction process can be represented as follows:

[0093]

[0094] Among them, Transformer J (·) represents the j-th Transformer encoder. and These are the degenerate features extracted from the source and target domains respectively under the j-th Transformer encoder. Each Transformer encoder includes an input embedding layer, position encoding, a multi-head self-attention mechanism layer, a feedforward neural network, and layer normalization, possessing the ability to extract multi-scale temporal features, providing rich feature information for subsequent feature alignment and remaining lifetime prediction.

[0095] S3 designs a multi-domain local adversarial learning strategy, which achieves accurate matching and effective transfer of the full-lifetime degradation features of each source domain to the non-full-lifetime degradation features of the target domain through degradation stage alignment and distribution feature alignment.

[0096] The degradation stage alignment module evaluates the degree of matching between the full-lifetime degradation features of each source domain and the non-full-lifetime degradation features of the target domain at the degradation stage through a weighted domain discriminator, and generates transferable weights to selectively filter out the full-lifetime degradation features of the source domain that are most relevant to the current degradation stage of the non-full-lifetime degradation features of the target domain.

[0097] The distribution feature alignment module introduces an adversarial domain discriminator, which maps the full-lifetime degradation features of each source domain and the non-full-lifetime degradation features of the target domain after the degradation stage alignment to the same feature space. This promotes the effective transmission and fusion of the two types of degradation features under different distribution characteristics, thereby mining the common degradation information that combines the source domain and the target domain.

[0098] Specifically, step S3 includes the following steps:

[0099] S3.1: Targeting the non-full-lifetime degradation characteristics of the target domain and those from N s The source domain lifetime degradation characteristics of each are used to construct N. s A weighted domain discriminator consists of two fully connected layers. During training, each source domain lifetime degradation feature is labeled as a positive class, and the target domain non-lifetime degradation feature is labeled as a negative class. The corresponding weighted domain discriminator is then trained under supervised supervision using a binary cross-entropy loss function. Specifically, the loss function for the j-th weighted domain discriminator is defined as follows:

[0100]

[0101] Among them, D j (·) represents the discriminator of the j-th weight domain; E(·) represents the expected value of the distribution function; Represents the lifetime degradation characteristics of the j-th source domain. It follows its corresponding distribution p(S); Represents the non-lifetime degradation characteristics of the target domain. It follows its corresponding distribution p(T).

[0102] S3.2: To further improve the transfer effectiveness of source domain full-lifetime degradation features to target domain non-full-lifetime degradation features, in each training batch, the confidence scores of each source domain full-lifetime degradation feature and target domain non-full-lifetime degradation feature belonging to their respective domains are calculated using the weight domain discriminator constructed in S3.1. The confidence scores of the two types of degradation features are then converted into transferable weights, providing a quantitative basis for degradation stage alignment. Specifically, this can be expressed as:

[0103]

[0104] in, The confidence level at which the lifetime degradation feature of the j-th source domain is determined as the source domain; The confidence level of determining the non-full-lifetime degradation feature of the target domain as a target domain (relative to the j-th weighted domain discriminator). ε = 10 -6 Transferable weights It is always between (0,1). The closer it is to 1, the closer the degradation stage of the source domain's full-lifetime degradation feature and the target domain's non-full-lifetime degradation feature are. At this time, the source domain's full-lifetime degradation feature has a high transfer value. Conversely, there is a large difference in the degradation stages of the two.

[0105] S3.3: For each source domain j∈{1,2,…,N S For the transferable weights, L2 regularization is applied. This regularization term is defined as follows:

[0106]

[0107] Here, ⊙ represents element-wise multiplication.

[0108] S3.4: Based on the L2 regularized transferable weights, targeted screening of the lifetime degradation features of each source domain is performed. Source domain lifetime degradation features at the same degradation stage as the non-lifetime degradation features of the target domain are strengthened, while source domain lifetime degradation features at different degradation stages are suppressed, thereby achieving alignment of degradation stages. The specific calculation process is as follows:

[0109]

[0110] in, Let be the transferable weight after L2 regularization for the j-th time.

[0111] S3.5: To reduce the difference in distribution characteristics between the full-lifetime degradation features of the source domains and the non-full-lifetime degradation features of the target domain after the degradation stages are aligned, N is constructed. S An adversarial discriminator, consisting of two fully connected layers, discriminates degenerate features from the source and target domains respectively, and calculates the discriminative loss of each adversarial discriminator based on the binary cross-entropy loss function.

[0112]

[0113] in, Let E(·) represent the j-th adversarial discriminator; E(·) represents the expected value of the distribution function. This represents the lifetime degradation characteristics of the source domain after alignment at the j-th degradation stage. It follows its corresponding distribution p(S); Represents the non-lifetime degradation characteristics of the target domain. It follows its corresponding distribution p(T).

[0114] Meanwhile, a gradient inversion layer is introduced between the feature extractor and the adversarial domain discriminator to achieve adversarial learning, which enables the feature extractor to learn the common degradation information of each source domain and the target domain, so as to further achieve feature alignment between each source domain and the target domain.

[0115] S4 utilizes the lifetime degradation features of each source domain that integrate common degradation information, and combines them with the corresponding real labels of remaining lifetime to construct a predictor using supervised learning.

[0116] Specifically, step S4 includes the following steps:

[0117] S4.1: Construct N S A predictor consists of two fully connected layers. For each source domain, the degradation features, after fusing common degradation information, are input into the corresponding predictor to obtain the remaining lifetime prediction. Then, the mean squared error loss function is used to calculate the deviation between the remaining lifetime prediction and the true remaining lifetime label, which serves as the prediction loss for each predictor.

[0118]

[0119] in, and Let represent the remaining lifetime prediction value and the actual label value of the j-th predictor, respectively.

[0120] S5. After executing steps S2 to S4, a multi-domain local adversarial learning remaining lifetime prediction model is obtained. During training, the total training loss function is calculated, and the parameters of the multi-domain local adversarial learning remaining lifetime prediction model are iteratively adjusted using an optimizer to obtain a trained prediction model.

[0121] Specifically, step S5 includes the following steps:

[0122] S5.1: After executing steps S2 to S4 above, a multi-domain local adversarial learning remaining lifetime prediction model is obtained. This model is then fused with N... SThe total training loss function is obtained by taking the discriminator loss in the weight domain, the L2 regularization loss, the adversarial discriminator loss, and the predictor loss.

[0123]

[0124] Where λ, β, and γ are the hyperparameters of the weighted domain discriminator loss, L2 regularization loss, and adversarial domain discriminator loss, respectively.

[0125] S5.2: With minimizing the total training loss function as the optimization objective, the parameters of the remaining lifetime prediction model are iteratively adjusted using the stochastic gradient descent algorithm until the training error converges, thus obtaining a well-trained prediction model.

[0126] S6. The trained multi-domain local adversarial learning remaining lifetime prediction model is used to test the non-full lifetime degradation data of the target domain to obtain the remaining lifetime results.

[0127] Specifically, step S6 includes the following steps:

[0128] S6.1: Input the non-full-lifetime degradation data of the target domain into the trained multi-domain local adversarial learning remaining lifetime prediction model to obtain N. S The output of the predictor The mean value is taken as the final remaining lifetime result:

[0129]

[0130] Based on the above, such as Figure 2 As shown, the multi-domain local adversarial learning strategy in this embodiment of the invention achieves accurate matching and effective transfer of full-lifetime degradation features of each source domain and non-full-lifetime degradation features of the target domain through degradation stage alignment and distribution characteristic alignment. Specifically, the degradation stage alignment module evaluates the degree of matching between the full-lifetime degradation features of each source domain and the non-full-lifetime degradation features of the target domain at the degradation stage using a weighted domain discriminator, and generates transferable weights to selectively identify the full-lifetime degradation features of the source domain most relevant to the current degradation stage of the non-full-lifetime degradation features of the target domain. The distribution characteristic alignment module introduces an adversarial domain discriminator to map the full-lifetime degradation features of each source domain and the non-full-lifetime degradation features of the target domain after degradation stage alignment to the same feature space, promoting the effective transfer and fusion of the two types of degradation features under different distribution characteristics, thereby mining common degradation information that combines the characteristics of each source domain and the target domain.

[0131] like Figure 3As shown in the embodiments of the present invention, after aligning the degradation stages, source domain full-lifetime degradation features that are at similar degradation stages to the non-full-lifetime degradation features of the target domain are assigned large weights, while source domain full-lifetime degradation features that are not similar in degradation stages are assigned small weights. This effectively alleviates the global feature mismatch problem and enhances the application value of source domain full-lifetime degradation features in lifetime prediction. After aligning the distribution characteristics, the peak values ​​of the probability density curves of source domain full-lifetime degradation features and target domain non-full-lifetime degradation features tend to be consistent and have a large overlap area, indicating that their distribution characteristics tend to be consistent.

[0132] like Figure 4 As shown, the Transformer encoder network in this embodiment of the invention includes an input embedding layer, position encoding, a multi-head self-attention mechanism layer, a feedforward neural network, and layer normalization. The full-lifetime degradation data of each source domain and the non-full-lifetime degradation data of the target domain are input into the network in parallel. The extracted degradation features provide rich feature information for subsequent feature alignment and remaining lifetime prediction.

[0133] Corresponding to the aforementioned method for predicting the remaining life of EMU gearbox bearings under conditions of incomplete data, this invention also discloses a system for predicting the remaining life of EMU gearbox bearings under conditions of incomplete data, such as... Figure 5 As shown, it specifically includes:

[0134] The data acquisition and preprocessing module is used to collect and preprocess the full-life vibration signals and non-full-life vibration signals of the EMU gearbox bearings under multiple operating conditions, and divides them into multi-source domain datasets and target domain datasets according to the full-life degradation.

[0135] The feature extraction module is used to extract degradation features from the full-lifetime degradation data of each source domain and the non-full-lifetime degradation data of the target domain based on the feature extraction network, so as to obtain the full-lifetime degradation features of each source domain and the non-full-lifetime degradation features of the target domain.

[0136] The multi-domain local adversarial learning module is used to design multi-domain local adversarial learning strategies. By introducing a weight domain discriminator and an adversarial domain discriminator, it achieves the alignment of degradation stages and distribution characteristics of the full-lifetime degradation features of each source domain and the non-full-lifetime degradation features of the target domain, and mines the common degradation information that combines the source domain and the target domain.

[0137] The predictor building module is used to construct a predictor by using the full-lifetime degradation features of each source domain that integrate common degradation information and the corresponding real labels of remaining lifetime in a supervised learning manner.

[0138] The model training module is used to adjust the network parameters of the feature extraction network, weight domain discriminator, adversarial domain discriminator and predictor with the total training loss function as the optimization objective, until a trained multi-domain local adversarial learning remaining lifetime prediction model is obtained.

[0139] The remaining lifetime prediction module is used to test the non-full lifetime degradation data of the target domain using a trained multi-domain local adversarial learning remaining lifetime prediction model to obtain the remaining lifetime result.

[0140] It should be noted that for a detailed description of the system for predicting the remaining life of a train gearbox bearing under incomplete data conditions provided in this embodiment of the invention, please refer to the relevant description of the method for predicting the remaining life of a train gearbox bearing under incomplete data conditions provided in this embodiment of the invention, which will not be repeated here.

[0141] In addition, embodiments of the present invention also provide an electronic device, such as... Figure 6 As shown, the device includes: a processor and a memory; the memory is used to store one or more program instructions; the processor is used to run one or more program instructions to perform the steps of a method for predicting the remaining life of a train gearbox bearing under incomplete data conditions as described in any of the preceding claims.

[0142] It should be noted that for a detailed description of an electronic device provided in the embodiments of the present invention, please refer to the relevant description of a method for predicting the remaining life of a train gearbox bearing under incomplete data conditions provided in the embodiments of this application, which will not be repeated here.

[0143] In addition, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method for predicting the remaining life of a train gearbox bearing under incomplete data conditions as described in any of the preceding embodiments.

[0144] It should be noted that for a detailed description of the computer-readable storage medium provided in the embodiments of the present invention, please refer to the relevant description of the method for predicting the remaining life of a train gearbox bearing under incomplete data conditions provided in the embodiments of this application, which will not be repeated here.

[0145] Those skilled in the art will understand that all or part of the functions of the various methods in the above embodiments can be implemented by hardware or by computer programs. When all or part of the functions in the above embodiments are implemented by computer programs, the program can be stored in a computer-readable storage medium, which may include: read-only memory, random access memory, disk, optical disk, hard disk, etc., and the program is executed by a computer to achieve the above functions. For example, the program can be stored in the memory of a device, and when the program in the memory is executed by the processor, all or part of the above functions can be achieved. In addition, when all or part of the functions in the above embodiments are implemented by computer programs, the program can also be stored in a server, another computer, disk, optical disk, flash drive, or external hard drive, etc., and can be downloaded or copied to the memory of a local device, or the system of the local device can be updated. When the program in the memory is executed by the processor, all or part of the functions in the above embodiments can be achieved.

[0146] The above examples illustrate the present invention only to aid in understanding it and are not intended to limit the scope of the invention. Those skilled in the art can make various simple deductions, modifications, or substitutions based on the principles of this invention.

Claims

1. A method for predicting the remaining life of gearbox bearings in high-speed trains under conditions of incomplete data, characterized in that, The method includes: The vibration signals of the gearbox bearings of EMUs under multiple operating conditions throughout their lifespan and non-lifespan were collected and preprocessed. Based on the lifespan degradation, the data were divided into multi-source domain datasets and target domain datasets. Based on the feature extraction network, degradation features are extracted from the full-lifetime degradation data of each source domain and the non-full-lifetime degradation data of the target domain, so as to obtain the full-lifetime degradation features of each source domain and the non-full-lifetime degradation features of the target domain. We design a multi-domain local adversarial learning strategy. By introducing a weighted domain discriminator and an adversarial domain discriminator, we can align the degradation stages and distribution characteristics of the full-lifetime degradation features of each source domain and the non-full-lifetime degradation features of the target domain, and mine the common degradation information that combines the source domain and the target domain. Specifically, a weighted domain discriminator is used to evaluate the matching degree between the full-lifetime degradation features of each source domain and the non-full-lifetime degradation features of the target domain at different degradation stages, and transferable weights are generated to selectively identify the full-lifetime degradation features of the source domain that are most relevant to the current degradation stage of the non-full-lifetime degradation features of the target domain. The specific steps include: Targeting the non-full-lifetime degradation characteristics of the target domain and Each source domain's lifetime degradation characteristics are used to construct... A weighted domain discriminator; During training, each source domain lifetime degradation feature is labeled as a positive class, and the target domain non-lifetime degradation feature is labeled as a negative class. The corresponding weight domain discriminator is then trained under supervision using a binary cross-entropy loss function. The loss function for the j-th weight domain discriminator is defined as follows: in, This represents the discriminator of the j-th weight domain; This represents the expected value of the distribution function; Represents the lifetime degradation characteristics of the j-th source domain. Follows its corresponding distribution ; Represents the non-lifetime degradation characteristics of the target domain. Follows its corresponding distribution ; In each training batch, the constructed weighted domain discriminator is used to calculate the confidence scores of each source domain's full-lifetime degradation feature and the target domain's non-full-lifetime degradation feature belonging to their respective domains. These confidence scores are then converted into transferable weights, providing a quantitative basis for degradation stage alignment. Specifically, this is expressed as follows: in, For the lifetime degradation characteristics of the j-th source domain The confidence level of being identified as the source domain; Non-full-lifetime degradation characteristics of the target domain The confidence level of being identified as the target domain; For the target domain samples; Constant; transferable weights It is always between (0,1). The closer it is to 1, the closer the degradation stage of the source domain's full-lifetime degradation feature and the target domain's non-full-lifetime degradation feature are. At this time, the source domain's full-lifetime degradation feature has a high transfer value. Conversely, it indicates that there is a large difference between the two degradation stages. For each source domain , The number of source domains is represented. The transferable weights are L2 regularized, and the regularization term is defined as follows: in, This indicates element-wise multiplication. This represents the total number of samples in the source domain. Based on the L2 regularized transferable weights, targeted screening of the lifetime degradation features of each source domain is performed. Source domain lifetime degradation features that are at the same degradation stage as the non-lifetime degradation features of the target domain are strengthened, while source domain lifetime degradation features that are at different degradation stages from the non-lifetime degradation features of the target domain are suppressed, thereby achieving alignment of degradation stages. The specific calculation process is as follows: in, Let be the transferable weight after the j-th L2 regularization; The method involves introducing an adversarial domain discriminator to map the full-lifetime degradation features of each source domain after degradation stage alignment to the non-full-lifetime degradation features of the target domain into the same feature space. This facilitates the effective transmission and fusion of the two types of degradation features under different distribution characteristics, thereby mining common degradation information that combines the features of each source domain and the target domain. The specific steps include: Build Each adversarial discriminator distinguishes between source domain degradation features and target domain degradation features aligned during the degradation stage, and calculates the discriminative loss of each adversarial discriminator based on the binary cross-entropy loss function. in, This represents the j-th adversarial domain discriminator; This represents the expected value of the distribution function; This represents the lifetime degradation characteristics of the source domain after alignment at the j-th degradation stage. Follows its corresponding distribution ; Represents the non-lifetime degradation characteristics of the target domain. Follows its corresponding distribution ; Meanwhile, a gradient reversal layer is introduced between the feature extractor and the adversarial domain discriminator to achieve adversarial learning, which enables the feature extractor to learn the common degradation information of each source domain and the target domain, so as to further achieve feature alignment between each source domain and the target domain. A predictor is constructed using supervised learning by leveraging the full-lifetime degradation features of each source domain that integrate common degradation information and the corresponding real labels of remaining lifetime. Using the total training loss function as the optimization objective, the network parameters of the feature extraction network, weight domain discriminator, adversarial domain discriminator, and predictor are adjusted until a well-trained multi-domain local adversarial learning remaining lifetime prediction model is obtained. The trained multi-domain local adversarial learning remaining lifetime prediction model is used to test the non-full lifetime degradation data of the target domain to obtain the remaining lifetime results.

2. The method for predicting the remaining life of a train gearbox bearing under incomplete data conditions as described in claim 1, characterized in that, Vibration signals of the gearbox bearings of high-speed trains under multiple operating conditions throughout their entire lifespan and those outside the entire lifespan were collected and preprocessed. Based on the lifespan degradation, the data was divided into a multi-source domain dataset and a target domain dataset, specifically including: Vibration signals of the gearbox bearings of the EMU under multiple operating conditions throughout / incomplete life were collected at fixed sampling intervals, and the vibration signal obtained each time was used as a sample. Z-Score standardization was used to process each sample to eliminate differences in statistical characteristics across different time scales; Several preprocessed sample segments are used to construct a full-lifetime / non-full-lifetime degradation dataset, and then divided into multi-source domain datasets based on the full-lifetime degradation status. and target domain dataset ,in Indicates the number of source domains; The multi-source domain dataset contains test bench full-lifetime degradation data and line historical full-lifetime degradation data under different operating conditions, covering the complete degradation process from the initial early fault to complete failure. Let the j-th source domain full-lifetime degradation dataset be denoted as... , This represents the i-th source domain sample. For its corresponding true remaining lifespan label This represents the total number of samples in the source domain. The target domain dataset includes real-time line degradation data, covering only the degradation process from the onset of an early fault to the current moment, and lacks degradation labels. This target domain non-full-lifetime degradation dataset is denoted as... , This represents the i-th target domain sample. For the target domain sample.

3. The method for predicting the remaining life of a train gearbox bearing under incomplete data conditions as described in claim 1, characterized in that, Based on a feature extraction network, degradation features are extracted from the full-lifetime degradation data of each source domain and the non-full-lifetime degradation data of the target domain, obtaining the full-lifetime degradation features of each source domain and the non-full-lifetime degradation features of the target domain, specifically including: Select Each Transformer encoder, with identical structure but independent parameters, serves as a feature extractor. Each Transformer encoder includes an input embedding layer, position encoding, a multi-head self-attention mechanism layer, a feedforward neural network, and layer normalization, enabling the extraction of multi-scale temporal features. For each source domain , The j-th Transformer encoder is input in parallel with the corresponding source domain full-lifetime degradation data and target domain non-full-lifetime degradation data to automatically learn the degradation patterns of the source and target domain data, thereby obtaining the degradation features of the two types of data respectively. The specific feature extraction process is as follows: in, This represents the j-th Transformer encoder. and These are the degradation features extracted from the source and target domains respectively under the j-th Transformer encoder.

4. The method for predicting the remaining life of a train gearbox bearing under incomplete data conditions as described in claim 1, characterized in that, A predictor is constructed using supervised learning, leveraging the lifetime degradation features of each source domain that incorporate common degradation information and their corresponding remaining lifetime true labels. Specifically, this includes: Build One predictor; for each source domain , The number of source domains is represented by the degraded features after fusing common degradation information, which are then input into the corresponding predictor to obtain the remaining lifetime prediction value. The mean squared error loss function is then used to calculate the deviation between the remaining lifetime prediction value and the true remaining lifetime label, which serves as the prediction loss for each predictor. The specific formula is as follows: in, and These represent the remaining lifetime prediction and the actual label value of the j-th predictor, respectively. This represents the total number of samples in the source domain.

5. The method for predicting the remaining life of a train gearbox bearing under incomplete data conditions as described in claim 1, characterized in that, Using the total training loss function as the optimization objective, the network parameters of the feature extraction network, weight domain discriminator, adversarial domain discriminator, and predictor are adjusted until a well-trained multi-domain local adversarial learning remaining lifetime prediction model is obtained. Specifically, this includes: Through the fusion model The total training loss function is obtained by taking the discriminator loss in the weight domain, the L2 regularization loss, the adversarial discriminator loss, and the predictor loss. in, , and These are the discriminator losses in the weighted region. L2 regularization loss Adversarial domain discriminator loss hyperparameters, For the predictor loss; for each source domain , Indicates the number of source domains; With the goal of minimizing the total training loss function, the parameters of the remaining lifetime prediction model are iteratively adjusted using the stochastic gradient descent algorithm until the training error converges, thus obtaining a well-trained prediction model.

6. The method for predicting the remaining life of a train gearbox bearing under conditions of incomplete data, as described in claim 1, is characterized in that... A trained multi-domain local adversarial learning model for predicting remaining lifetime is used to test non-full lifetime degradation data in the target domain to obtain remaining lifetime results, specifically including: Inputting the non-full-lifetime degradation data of the target domain into a trained multi-domain local adversarial learning remaining lifetime prediction model yields... The output of the predictor The average value is then taken as the final remaining lifetime result: in, For the first The output of each predictor .

7. A system for predicting the remaining life of gearbox bearings in high-speed trains under conditions of incomplete data, characterized in that, The system includes: The data acquisition and preprocessing module is used to collect and preprocess the full-life vibration signals and non-full-life vibration signals of the EMU gearbox bearings under multiple operating conditions, and divides them into multi-source domain datasets and target domain datasets according to the full-life degradation. The feature extraction module is used to extract degradation features from the full-lifetime degradation data of each source domain and the non-full-lifetime degradation data of the target domain based on the feature extraction network, so as to obtain the full-lifetime degradation features of each source domain and the non-full-lifetime degradation features of the target domain. The multi-domain local adversarial learning module is used to design multi-domain local adversarial learning strategies. By introducing a weight domain discriminator and an adversarial domain discriminator, it achieves the alignment of degradation stages and distribution characteristics of the full-lifetime degradation features of each source domain and the non-full-lifetime degradation features of the target domain, and mines the common degradation information that combines the source domain and the target domain. Specifically, a weighted domain discriminator is used to evaluate the matching degree between the full-lifetime degradation features of each source domain and the non-full-lifetime degradation features of the target domain at different degradation stages, and transferable weights are generated to selectively identify the full-lifetime degradation features of the source domain that are most relevant to the current degradation stage of the non-full-lifetime degradation features of the target domain. The specific steps include: Targeting the non-full-lifetime degradation characteristics of the target domain and Each source domain's lifetime degradation characteristics are used to construct... A weighted domain discriminator; During training, each source domain lifetime degradation feature is labeled as a positive class, and the target domain non-lifetime degradation feature is labeled as a negative class. The corresponding weight domain discriminator is then trained under supervision using a binary cross-entropy loss function. The loss function for the j-th weight domain discriminator is defined as follows: in, This represents the discriminator of the j-th weight domain; This represents the expected value of the distribution function; Represents the lifetime degradation characteristics of the j-th source domain. Follows its corresponding distribution ; Represents the non-lifetime degradation characteristics of the target domain. Follows its corresponding distribution ; In each training batch, the constructed weighted domain discriminator is used to calculate the confidence scores of each source domain's full-lifetime degradation feature and the target domain's non-full-lifetime degradation feature belonging to their respective domains. These confidence scores are then converted into transferable weights, providing a quantitative basis for degradation stage alignment. Specifically, this is expressed as follows: in, For the lifetime degradation characteristics of the j-th source domain The confidence level of being identified as the source domain; Non-full-lifetime degradation characteristics of the target domain The confidence level of being identified as the target domain; For the target domain samples; Constant; transferable weights It is always between (0,1). The closer it is to 1, the closer the degradation stage of the source domain's full-lifetime degradation feature and the target domain's non-full-lifetime degradation feature are. At this time, the source domain's full-lifetime degradation feature has a high transfer value. Conversely, it indicates that there is a large difference between the two degradation stages. For each source domain , The number of source domains is represented. The transferable weights are L2 regularized, and the regularization term is defined as follows: in, This indicates element-wise multiplication. This represents the total number of samples in the source domain. Based on the L2 regularized transferable weights, targeted screening of the lifetime degradation features of each source domain is performed. Source domain lifetime degradation features that are at the same degradation stage as the non-lifetime degradation features of the target domain are strengthened, while source domain lifetime degradation features that are at different degradation stages from the non-lifetime degradation features of the target domain are suppressed, thereby achieving alignment of degradation stages. The specific calculation process is as follows: in, Let be the transferable weight after the j-th L2 regularization; The method involves introducing an adversarial domain discriminator to map the full-lifetime degradation features of each source domain after degradation stage alignment to the non-full-lifetime degradation features of the target domain into the same feature space. This facilitates the effective transmission and fusion of the two types of degradation features under different distribution characteristics, thereby mining common degradation information that combines the features of each source domain and the target domain. The specific steps include: Build Each adversarial discriminator distinguishes between source domain degradation features and target domain degradation features aligned during the degradation stage, and calculates the discriminative loss of each adversarial discriminator based on the binary cross-entropy loss function. in, This represents the j-th adversarial domain discriminator; This represents the expected value of the distribution function; This represents the lifetime degradation characteristics of the source domain after alignment at the j-th degradation stage. Follows its corresponding distribution ; Represents the non-lifetime degradation characteristics of the target domain. Follows its corresponding distribution ; Meanwhile, a gradient reversal layer is introduced between the feature extractor and the adversarial domain discriminator to achieve adversarial learning, which enables the feature extractor to learn the common degradation information of each source domain and the target domain, so as to further achieve feature alignment between each source domain and the target domain. The predictor building module is used to construct a predictor by using the full-lifetime degradation features of each source domain that integrate common degradation information and the corresponding real labels of remaining lifetime in a supervised learning manner. The model training module is used to adjust the network parameters of the feature extraction network, weight domain discriminator, adversarial domain discriminator and predictor with the total training loss function as the optimization objective, until a trained multi-domain local adversarial learning remaining lifetime prediction model is obtained. The remaining lifetime prediction module is used to test the non-full lifetime degradation data of the target domain using a trained multi-domain local adversarial learning remaining lifetime prediction model to obtain the remaining lifetime result.

8. An electronic device, characterized in that, The device includes: a processor and a memory; The memory is used to store one or more program instructions; The processor is configured to run one or more program instructions to perform the steps of a method for predicting the remaining life of a train gearbox bearing under incomplete data conditions as described in any one of claims 1 to 6.

Images