Remaining useful life prediction method based on small sample enhancement and interpretable migration contribution

Abstract

The application discloses a residual life prediction method based on small sample enhancement and interpretable migration contribution, and relates to the technical field of predictive maintenance of rotating machinery. The method comprises the following steps: inputting a source domain time sequence into a preset data enhancement network for data enhancement to obtain a plurality of sub-source domain time sequences; determining a plurality of sub-source domain feature vectors and a predicted life of a rotating equipment based on time-frequency domain features corresponding to the plurality of sub-source domain time sequences, an interpretable estimation network and a life prediction network; determining a target domain feature vector based on time-frequency domain features corresponding to a target domain time sequence and the interpretable estimation network; constructing a life prediction loss and a maximum mean difference loss between the plurality of sub-source domain time sequences and the target domain time sequence according to the predicted life, the plurality of sub-source domain feature vectors and the target domain feature vector, and performing iterative training. The application can solve the problem of inconsistent distribution between a source domain and a target domain in migration learning.

Description

Technical Field

[0001] This application relates to the field of predictive maintenance technology for rotating machinery, and in particular to a method for predicting remaining life based on small-sample augmentation and interpretable migration contributions. Background Technology

[0002] In modern industry, rotating equipment (such as electric motors, pumps, and fans) is a core component, and its operating status directly affects the reliability and service life of the equipment. Effectively predicting equipment lifespan has become a key technology in equipment maintenance and predictive management. In practical industrial applications, due to differences in equipment models, operating environments, and monitoring systems, obtaining a large number of high-quality labeled samples is costly and difficult; therefore, transfer learning is typically employed.

[0003] Currently, existing transfer learning methods typically assume that the data distribution between the source and target domains is consistent. However, due to differences in equipment type, operating conditions, load, and environmental factors, the data distribution between the source and target domains often differs significantly. Existing transfer learning methods cannot effectively address the issue of inconsistent distribution between the source and target domains, resulting in poor transfer performance and low accuracy in equipment lifetime prediction. Summary of the Invention

[0004] In view of this, this application provides a remaining lifetime prediction method based on small sample augmentation and interpretable transfer contribution. The main purpose is to effectively solve the problem of inconsistent distribution between the source domain and the target domain in transfer learning, thereby improving the transfer effect and ensuring the accuracy of lifetime prediction for rotating equipment.

[0005] According to a first aspect of this application, a method for predicting remaining lifetime based on small-sample augmentation and interpretable migration contributions is provided, the method comprising: Obtain the source domain time series and target domain time series of the rotating device, as well as an initial lifetime prediction model, wherein the initial lifetime prediction model includes an interpretable estimation network and a lifetime prediction network; The source domain time series is input into a preset data augmentation network for data augmentation to obtain multiple sub-source domain time series; Based on the time-frequency domain features corresponding to the multiple sub-source domain time series, the interpretable estimation network, and the lifetime prediction network, multiple sub-source domain feature vectors and the predicted lifetime of the rotating equipment are determined. The interpretable estimation network adaptively allocates weights according to the contribution of each feature in the time-frequency domain features corresponding to each sub-source domain time series. Based on the time-frequency domain features corresponding to the target domain time series and the interpretable estimation network, the target domain feature vector is determined; Based on the predicted lifetime, the feature vectors of the multiple sub-source domains, and the feature vector of the target domain, a lifetime prediction loss and a maximum mean difference loss between the time series of the multiple sub-source domains and the time series of the target domain are constructed respectively. Based on the lifetime prediction loss and multiple maximum mean difference losses, the initial lifetime prediction model is iteratively trained to obtain a preset lifetime prediction model.

[0006] According to a second aspect of this application, a device for predicting remaining lifetime based on small-sample augmentation and interpretable migration contributions is provided, the device comprising: An acquisition unit is used to acquire the source domain time series and target domain time series of the rotating device, as well as an initial lifetime prediction model, wherein the initial lifetime prediction model includes an interpretable estimation network and a lifetime prediction network. An enhancement unit is used to input the source domain time series into a preset data enhancement network for data enhancement, thereby obtaining multiple sub-source domain time series; The determining unit is used to determine multiple sub-source domain feature vectors and the predicted lifetime of the rotating device based on the time-frequency domain features corresponding to the multiple sub-source domain time series, the interpretable estimation network, and the lifetime prediction network, wherein the interpretable estimation network adaptively allocates weights according to the contribution of each feature in the time-frequency domain features corresponding to each sub-source domain time series. The determining unit is further configured to determine the target domain feature vector based on the time-frequency domain features corresponding to the target domain time series and the interpretable estimation network; The construction unit is used to construct lifetime prediction loss and maximum mean difference loss between the time series of the multiple sub-source domains and the time series of the target domain, respectively, based on the predicted lifetime, the feature vectors of the multiple sub-source domains and the feature vector of the target domain. An iterative unit is used to iteratively train the initial lifetime prediction model based on the lifetime prediction loss and multiple maximum mean difference losses to obtain a preset lifetime prediction model.

[0007] According to a third aspect of this application, a storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the above-described method for predicting remaining lifetime based on small-sample augmentation and interpretable migration contributions.

[0008] According to a fourth aspect of this application, an electronic device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor, when executing the program, implements the above-described method for predicting remaining lifetime based on small sample augmentation and interpretable migration contribution.

[0009] By employing the aforementioned technical solutions, this application provides a remaining lifetime prediction method based on small-sample augmentation and interpretable transfer contributions. By calculating the maximum mean difference loss between multiple sub-source domain time series and the target domain time series, and constructing a loss function, the lifetime prediction model is iteratively trained. This effectively solves the problem of inconsistent distribution between the source and target domains in transfer learning, thereby improving transfer performance and ensuring the accuracy of lifetime prediction for rotating equipment. Simultaneously, this application enhances the source domain data through a data augmentation network, forming multiple sub-source domain time series. This improves data quality and usability while preserving key signal information, thus addressing the data scarcity problem under small-sample conditions. Furthermore, by introducing an interpretable estimation network, this application adaptively adjusts the contribution weights of various features in lifetime prediction, enhancing the interpretability of the prediction model's decision-making process and increasing user confidence in the model's prediction results.

[0010] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0011] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 The diagram shows a flowchart of a remaining lifetime prediction method based on small sample augmentation and interpretable migration contribution provided in an embodiment of this application. Figure 2 This paper illustrates the overall process of migration prediction provided in an embodiment of this application. Figure 3 The diagram shows a structural schematic of a remaining lifetime prediction device based on small sample enhancement and interpretable migration contribution provided in an embodiment of this application. Detailed Implementation

[0012] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0013] Due to differences in equipment type, operating conditions, load, and environmental factors, the data distribution between the source domain and the target domain often differs significantly. Existing transfer learning methods cannot effectively address the inconsistency between the source and target domain distributions, resulting in poor transfer performance and low accuracy in equipment lifespan prediction.

[0014] To address the aforementioned problems, embodiments of the present invention provide a remaining lifetime prediction method based on small-sample augmentation and interpretable migration contributions, such as... Figure 1 As shown, the method includes: Step 101: Obtain the source domain time series and target domain time series of the rotating equipment, as well as the initial lifetime prediction model.

[0015] Rotating equipment includes rotating machinery such as electric motors, pumps, and fans. The source domain time series and target domain time series can specifically be vibration signal time series. The initial lifetime prediction model includes an interpretable estimation network and a lifetime prediction network.

[0016] In this embodiment of the invention, the source domain time series comes from the training dataset, and the target domain time series comes from the validation dataset. The source domain training dataset has sufficient labeled samples, meaning that a large amount of data has been accumulated in similar rotating equipment, operating conditions, or monitoring environments. The target domain validation dataset and test set come from new rotating equipment and operating conditions, and the available labeled samples are very limited. Specifically, the source domain time series and the target domain time series can be vibration signal time series, which can be collected using sensors.

[0017] Due to differences in equipment type, operating conditions, load, and environmental factors, the feature distributions of the source and target domains often differ significantly. Traditional transfer learning methods cannot effectively address this distribution inconsistency, resulting in poor transfer performance. To effectively resolve the issue of inconsistent source and target domain distributions in transfer learning, this invention constructs an initial lifetime prediction model. By calculating the maximum mean difference loss between multiple sub-source domain time series and the target domain time series, this initial lifetime prediction model is iteratively trained, effectively resolving the issue of inconsistent source and target domain distributions in transfer learning. This initial lifetime prediction model includes an interpretable estimation network and a lifetime prediction network. The interpretable estimation network, by introducing a spatial multi-head attention mechanism, can adaptively adjust the contribution of each sub-source domain and each feature in lifetime prediction, enhancing the interpretability of the model's decision-making process and increasing user confidence in the model's prediction results. The lifetime prediction network can specifically be a one- to three-layer fully connected network.

[0018] This invention is primarily applicable to scenarios where a lifespan prediction model for rotating equipment is trained based on a small sample size. The executing entity of this invention is a device or equipment capable of training a lifespan prediction model for rotating equipment based on a small sample size, specifically, it can be located on a server side.

[0019] Step 102: Input the source domain time series into a preset data augmentation network for data augmentation to obtain multiple sub-source domain time series.

[0020] The preset data augmentation network includes an encoder and a decoder.

[0021] In this embodiment of the invention, when performing data augmentation, the source domain time series is input into a preset data augmentation network for processing to obtain multiple candidate time series; a target time series is determined from the multiple candidate time series; and different mathematical calculations are performed on the target time series to obtain the multiple sub-source domain time series.

[0022] Specifically, based on the time window corresponding to the encoder, the source domain time series is sequentially input into the encoder and decoder of the preset data augmentation network for processing to obtain multiple candidate time series.

[0023] For example, if the time window is set to 10, after the source domain time series is processed by the encoder and decoder, 10 candidate time series can be generated. Then, one candidate time series is randomly selected from the 10 candidate time series as the target time series. Then, the target time series is processed by taking the derivative, differentiating, integrating and other processes to obtain multiple sub-source domain time series.

[0024] It should be noted that the embodiments of the present invention can generate at least two sub-source domain time series. The embodiments of the present invention do not impose a specific limit on the number of sub-source domain time series generated, which can be set according to actual business needs. Furthermore, the training process of the encoder and decoder is a separate training process from the training process of the lifetime prediction model.

[0025] Step 103: Based on the time-frequency domain features corresponding to the multiple sub-source domain time series, the interpretable estimation network, and the lifetime prediction network, determine the multiple sub-source domain feature vectors and the predicted lifetime of the rotating equipment.

[0026] The interpretable estimation network adaptively allocates weights based on the contribution of each feature in the time-frequency domain features corresponding to each sub-source domain time series. The interpretable estimation network includes a gated recurrent unit (GRU) network and a multi-head attention mechanism network. The lifetime prediction network is specifically a regression model, such as one to three fully connected layers.

[0027] In this embodiment of the invention, after generating multiple sub-source domain time series, time-frequency domain features corresponding to the multiple sub-source domain time series are extracted respectively. Then, the time-frequency domain features corresponding to the multiple sub-source domain time series are input into an interpretable estimation network for feature extraction to obtain source domain context vectors and multiple sub-source domain feature vectors. Then, the source domain context vectors are input into a lifetime prediction network for lifetime prediction to obtain the predicted lifetime of the rotating equipment.

[0028] When extracting the source domain context vector and multiple sub-source domain feature vectors, the time-frequency domain features corresponding to the time series of multiple sub-source domains are input into a gated recurrent unit for feature extraction to obtain multiple sub-source domain degradation features. The multiple sub-source domain degradation features are input into the attention mechanism network for linear mapping and adaptive weight allocation to obtain a total value feature matrix and the contribution weight of each feature in the time-frequency domain features corresponding to the time series of multiple sub-source domains. Based on the contribution weight of each feature in the time-frequency domain features corresponding to the time series of multiple sub-source domains, the total value feature matrix is ​​weighted and summed to obtain the source domain context vector. The total value feature matrix is ​​sliced ​​and pooled according to the sub-source domains to obtain the multiple sub-source domain feature vectors.

[0029] Specifically, the time-frequency domain features corresponding to each sub-source domain time series are linearly projected to obtain their respective Value representations. Then, the Value vectors of each sub-source domain are concatenated according to their token order to form a unified total numerical feature matrix. This total numerical feature matrix is ​​then used simultaneously in two parallel branches. One branch multiplies the matrix by the contribution weights of each feature in the time-frequency domain corresponding to the multiple sub-source domain time series to obtain the source domain context vector. Simultaneously, the total numerical feature matrix is ​​sliced ​​and pooled according to the sub-source domains to obtain the feature vectors of each sub-source domain for subsequent maximum mean difference loss calculation.

[0030] In some embodiments, based on the contribution weights of each feature in the time-frequency domain features corresponding to the multiple sub-source domain time series, the contribution weights are summed with each sub-source domain as a statistical unit to obtain the total contribution corresponding to the multiple sub-source domain time series. This contribution is used to dynamically adjust the initial weight coefficients of the maximum mean difference loss of each sub-source domain.

[0031] This invention introduces a spatial multi-head attention mechanism to adaptively adjust the contribution of each sub-source domain and each feature in lifetime prediction, thereby enhancing the interpretability of the model decision-making process and increasing users' confidence in the model's prediction results.

[0032] Furthermore, the source domain context vector of the interpretable estimation network output is input into the regression model for lifetime prediction (RUL prediction). By combining Bayesian regression, the regression model can not only output the predicted lifetime, but also the confidence interval, thereby quantifying the uncertainty of the prediction and providing a reliable basis for subsequent maintenance decisions.

[0033] Step 104: Determine the target domain feature vector based on the time-frequency domain features corresponding to the target domain time series and the interpretable estimation network.

[0034] In this embodiment of the invention, following the data processing procedure of the interpretable estimation network in step 103 above, the time-frequency domain features corresponding to the target domain time series are similarly input into the gated recurrent unit for feature extraction to obtain the target domain degradation features. Then, the target domain degradation features are input into the attention mechanism network for processing to obtain the target domain feature vector. This target domain feature vector is used to calculate the maximum mean difference loss between the target domain feature vector and the feature vectors of multiple sub-source domains.

[0035] Step 105: Based on the predicted lifetime, the feature vectors of the multiple sub-source domains, and the feature vector of the target domain, construct the lifetime prediction loss and the maximum mean difference loss between the time series of the multiple sub-source domains and the time series of the target domain, respectively.

[0036] In this embodiment of the invention, when constructing the total loss function, the maximum mean difference loss between the time series of multiple sub-source domains and the time series of the target domain is calculated based on the feature vectors of multiple sub-source domains and the feature vector of the target domain. Simultaneously, the lifespan loss of the rotating equipment is calculated based on the actual lifespan and predicted lifespan of the rotating equipment.

[0037] Step 106: Based on the lifetime prediction loss and multiple maximum mean difference losses, iteratively train the initial lifetime prediction model to obtain a preset lifetime prediction model.

[0038] In this embodiment of the invention, during iterative training, initial weight coefficients corresponding to the plurality of maximum mean difference losses are determined; based on the plurality of initial weight coefficients, the plurality of maximum mean difference losses are weighted and summed to obtain the total maximum mean difference loss; the lifetime prediction loss and the total maximum mean difference loss are summed to obtain the total loss; based on the total loss, the initial lifetime prediction model and the plurality of initial weight coefficients are iteratively updated until a preset condition is met, and the preset lifetime prediction model is output. During iterative updates, the plurality of initial weight coefficients are dynamically adjusted according to the total contribution of the plurality of sub-source domain time series.

[0039] This invention employs Maximum Mean Difference (MMD) loss to minimize the distribution difference between the source and target domain time series, thereby optimizing the adaptability of the rotating equipment between the source and target domains. Simultaneously, this invention addresses the differences between sub-source domains by dynamically adjusting the contribution weight coefficients of different sub-source domains, avoiding negative transfer and thus improving the model's cross-domain adaptability. The overall training process of the preset lifetime prediction model is as follows: Figure 2 As shown.

[0040] After training the preset lifetime prediction model, the target domain time series data from the prediction set are sequentially input into a preset data augmentation network, an interpretable estimation network, and a lifetime prediction network for processing, outputting the predicted lifetime. By combining Bayesian regression, the prediction model can not only output the predicted lifetime but also the confidence interval, thereby quantifying the uncertainty of the prediction and providing a reliable basis for subsequent maintenance decisions.

[0041] This invention provides a remaining lifetime prediction method based on small-sample augmentation and interpretable transfer contributions. By augmenting the source domain data of rotating equipment, it can fully mine the potential information in existing data even when data is scarce, thereby significantly improving the lifetime prediction accuracy under small-sample conditions. This effect is of great significance in equipment maintenance scenarios where it is difficult to collect large amounts of labeled data. This invention utilizes the dynamic attention mechanism in the interpretable estimation network to clearly identify time steps and features that significantly influence lifetime prediction, explain key factors in the model decision-making process, and improve model transparency. For operators and maintenance engineers in industrial settings, this helps them understand the basis of the model's predictions, increasing their trust in the prediction results and enabling them to make more accurate maintenance decisions. Furthermore, this invention effectively solves the negative transfer problem in multi-source domain adaptation by adaptively adjusting the contribution weights of sub-source domains. The negative transfer problem often arises from excessive semantic differences between the source and target domains, causing some features of the source domain to adversely affect the prediction of the target domain. This invention achieves a more balanced knowledge transfer among multiple sub-source domains through dynamic weighting, improving the stability of multi-source domain data fusion and ensuring the accuracy of lifetime prediction in the target domain. Furthermore, by comprehensively utilizing small-sample augmentation, interpretable transfer contributions, and multi-source domain adaptation techniques, the embodiments of this invention significantly improve the accuracy, interpretability, and adaptability of predicting the remaining service life of rotating equipment. These technical effects not only enable the model to operate stably in small-sample and cross-domain environments but also improve the transparency and accuracy of decision-making, providing strong support for predictive maintenance of industrial equipment. Therefore, the embodiments of this invention have broad practical value and market prospects in industrial applications.

[0042] Furthermore, as Figure 1 and Figure 2 The specific implementation of the method shown in this embodiment provides a remaining lifetime prediction device based on small sample augmentation and interpretable migration contribution, such as... Figure 3 As shown, the device includes: an acquisition unit 31, an enhancement unit 32, a determination unit 33, a construction unit 34, and an iteration unit 35.

[0043] The acquisition unit 31 can be used to acquire the source domain time series and target domain time series of the rotating device, as well as the initial lifetime prediction model, wherein the initial lifetime prediction model includes an interpretable estimation network and a lifetime prediction network.

[0044] The enhancement unit 32 can be used to input the source domain time series into a preset data enhancement network for data enhancement to obtain multiple sub-source domain time series.

[0045] The determining unit 33 can be used to determine multiple sub-source domain feature vectors and the predicted lifetime of the rotating device based on the time-frequency domain features corresponding to the multiple sub-source domain time series, the interpretable estimation network, and the lifetime prediction network, wherein the interpretable estimation network adaptively allocates weights according to the contribution of each feature in the time-frequency domain features corresponding to each sub-source domain time series.

[0046] The determining unit 33 can also be used to determine the target domain feature vector based on the time-frequency domain features corresponding to the target domain time series and the interpretable estimation network.

[0047] The construction unit 34 can be used to construct lifetime prediction loss and maximum mean difference loss between the multiple sub-source domain feature vectors and the target domain feature vector, respectively, based on the predicted lifetime, the multiple sub-source domain feature vectors and the target domain feature vector.

[0048] The iterative unit 35 can be used to iteratively train the initial lifetime prediction model based on the lifetime prediction loss and multiple maximum mean difference losses to obtain a preset lifetime prediction model.

[0049] In some embodiments, the determining unit 33 includes an extraction module and a prediction module.

[0050] The extraction module can be used to input the time-frequency domain features corresponding to the multiple sub-source domain time series into the interpretable estimation network for feature extraction, so as to obtain the source domain context vector and the multiple sub-source domain feature vectors.

[0051] The prediction module can be used to input the source domain context vector into the lifetime prediction network to predict the lifetime and obtain the predicted lifetime of the rotating equipment.

[0052] In some embodiments, the interpretable estimation network includes a gated recurrent unit and an attention mechanism network. The extraction module can be specifically configured to input the time-frequency domain features corresponding to the multiple sub-source domain time series into the gated recurrent unit for feature extraction, obtaining multiple sub-source domain degradation features; input the multiple sub-source domain degradation features into the attention mechanism network for linear mapping and adaptive weight allocation, obtaining a total value feature matrix and the contribution weight of each feature in the time-frequency domain features corresponding to the multiple sub-source domain time series; perform a weighted summation on the total value feature matrix based on the contribution weight of each feature in the time-frequency domain features corresponding to the multiple sub-source domain time series, obtaining the source domain context vector; and slice and pool the total value feature matrix according to the sub-source domains to obtain the multiple sub-source domain feature vectors.

[0053] In some embodiments, the determining unit 33 further includes a summing module.

[0054] The summation module can be used to sum the contribution weights of each feature in the time-frequency domain features corresponding to the multiple sub-source domain time series, with each sub-source domain as a statistical unit, to obtain the total contribution corresponding to the multiple sub-source domain time series.

[0055] In some embodiments, the iterative unit 35 may be specifically used to determine the initial weight coefficients corresponding to the plurality of maximum mean difference losses; based on the plurality of initial weight coefficients, perform a weighted summation of the plurality of maximum mean difference losses to obtain the total maximum mean difference loss; sum the lifetime prediction loss and the total maximum mean difference loss to obtain the total loss; based on the total loss, iteratively update the initial lifetime prediction model and the plurality of initial weight coefficients until a preset condition is met, and output the preset lifetime prediction model, wherein, during the iterative update, the plurality of initial weight coefficients are dynamically adjusted according to the total contribution corresponding to the plurality of sub-source domain time series.

[0056] In some embodiments, the enhancement unit 32 includes a processing module, a determination module, and a calculation module.

[0057] The processing module can be used to input the source domain time series into a preset data augmentation network for processing to obtain multiple candidate time series.

[0058] The determining module can be used to determine the target time series from the plurality of candidate time series.

[0059] The calculation module can be used to perform different mathematical calculations on the target time series to obtain the multiple sub-source domain time series.

[0060] In some embodiments, the processing module may be specifically used to input the source domain time series into the encoder and the decoder sequentially for processing based on the time window corresponding to the encoder, so as to obtain the plurality of candidate time series.

[0061] It should be noted that other corresponding descriptions of the functional units involved in the remaining lifetime prediction device based on small sample enhancement and interpretable migration contribution provided in this embodiment of the invention can be found in the following references. Figure 1 The corresponding descriptions in [the document] will not be repeated here.

[0062] Based on the above, Figure 1 Accordingly, this embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the above-described method. Figure 1 The remaining lifetime prediction method shown is based on small sample augmentation and interpretable migration contribution.

[0063] Based on this understanding, the technical solution of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause an electronic device (such as personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of this application.

[0064] Based on the above, Figure 1 The method shown, and Figure 3 To achieve the above objectives, the present application also provides an electronic device, specifically a personal computer, tablet computer, server, or other network device, as shown in the virtual device embodiment. This device includes a storage medium and a processor; the storage medium stores a computer program; the processor executes the computer program to achieve the above-described objectives. Figure 1 The remaining lifetime prediction method shown is based on small sample augmentation and interpretable migration contribution.

[0065] Optionally, the aforementioned physical devices may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.

[0066] Those skilled in the art will understand that the physical device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0067] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.

[0068] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platform, or it can be implemented by hardware.

[0069] This invention enhances the source domain data of rotating equipment, enabling the full extraction of potential information from existing data even in data-scarce environments. This significantly improves the lifetime prediction accuracy under small sample conditions, a crucial benefit in equipment maintenance scenarios where collecting large amounts of labeled data is difficult. Furthermore, by employing a dynamic attention mechanism in an interpretable estimation network, this invention clearly identifies time steps and features that significantly impact lifetime prediction, explaining key factors in the model's decision-making process and enhancing model transparency. This helps operators and maintenance engineers in industrial settings understand the basis of the model's predictions, increasing their confidence in the results and enabling more accurate maintenance decisions. Moreover, by adaptively adjusting the contribution weights of sub-source domains, this invention effectively addresses the negative transfer problem in multi-source domain adaptation. Negative transfer often arises from significant semantic differences between the source and target domains, causing some source domain features to negatively impact target domain predictions. This invention achieves a more balanced knowledge transfer across multiple sub-source domains through dynamic weighting, improving the stability of multi-source domain data fusion and ensuring the accuracy of lifetime predictions in the target domain. Furthermore, by comprehensively utilizing small-sample augmentation, interpretable transfer contributions, and multi-source domain adaptation techniques, the embodiments of this invention significantly improve the accuracy, interpretability, and adaptability of predicting the remaining service life of rotating equipment. These technical effects not only enable the model to operate stably in small-sample and cross-domain environments but also improve the transparency and accuracy of decision-making, providing strong support for predictive maintenance of industrial equipment. Therefore, the embodiments of this invention have broad practical value and market prospects in industrial applications.

[0070] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application. Those skilled in the art will understand that the modules in the apparatus of the embodiment can be distributed within the apparatus of the embodiment as described, or can be modified to be located in one or more apparatuses different from this embodiment. The modules of the above-described embodiment can be combined into one module, or further divided into multiple sub-modules.

[0071] The serial numbers in this application are for descriptive purposes only and do not represent the superiority or inferiority of any particular implementation scenario. The above disclosures are merely a few specific implementation scenarios of this application; however, this application is not limited thereto, and any variations conceived by those skilled in the art should fall within the protection scope of this application.

Claims

1. A method for predicting remaining lifetime based on small-sample augmentation and interpretable migration contributions, characterized in that, include: Obtain the source domain time series and target domain time series of the rotating device, as well as an initial lifetime prediction model, wherein the initial lifetime prediction model includes an interpretable estimation network and a lifetime prediction network; The source domain time series is input into a preset data augmentation network for data augmentation to obtain multiple sub-source domain time series; Based on the time-frequency domain features corresponding to the multiple sub-source domain time series, the interpretable estimation network, and the lifetime prediction network, multiple sub-source domain feature vectors and the predicted lifetime of the rotating equipment are determined. The interpretable estimation network adaptively allocates weights according to the contribution of each feature in the time-frequency domain features corresponding to each sub-source domain time series. Based on the time-frequency domain features corresponding to the target domain time series and the interpretable estimation network, the target domain feature vector is determined; Based on the predicted lifetime, the feature vectors of the multiple sub-source domains, and the feature vector of the target domain, a lifetime prediction loss and a maximum mean difference loss between the time series of the multiple sub-source domains and the time series of the target domain are constructed respectively. Based on the lifetime prediction loss and multiple maximum mean difference losses, the initial lifetime prediction model is iteratively trained to obtain a preset lifetime prediction model.

2. The method according to claim 1, characterized in that, The step of determining multiple sub-source domain feature vectors and the predicted lifetime of the rotating equipment based on the time-frequency domain features corresponding to the time series of the multiple sub-source domains, the interpretable estimation network, and the lifetime prediction network includes: The time-frequency domain features corresponding to the time series of the multiple sub-source domains are input into the interpretable estimation network for feature extraction, resulting in source domain context vectors and feature vectors of the multiple sub-source domains. The source domain context vector is input into the lifetime prediction network to predict the lifetime of the rotating equipment.

3. The method according to claim 2, characterized in that, The interpretable estimation network includes a gated recurrent unit and an attention mechanism network. The step of inputting the time-frequency domain features corresponding to the time series of the multiple sub-source domains into the interpretable estimation network for feature extraction, to obtain the source domain context vector and the feature vectors of the multiple sub-source domains, includes: The time-frequency domain features corresponding to the time series of the multiple sub-source domains are input into the gated recurrent unit for feature extraction to obtain multiple sub-source domain degradation features. The degradation features of the multiple sub-source domains are input into the attention mechanism network for linear mapping and adaptive weight allocation to obtain the total value feature matrix and the contribution weight of each feature in the time-frequency domain features corresponding to the time series of the multiple sub-source domains. Based on the contribution weights of each feature in the time-frequency domain features corresponding to the multiple sub-source domain time series, the total value feature matrix is ​​weighted and summed to obtain the source domain context vector; The total value feature matrix is ​​sliced ​​and pooled according to the sub-source domain to obtain the multiple sub-source domain feature vectors.

4. The method according to claim 3, characterized in that, The method further includes: Based on the contribution weights of each feature in the time-frequency domain features corresponding to the multiple sub-source domain time series, the contribution weights are summed with each sub-source domain as a statistical unit to obtain the total contribution corresponding to the multiple sub-source domain time series.

5. The method according to claim 4, characterized in that, The step of iteratively training the initial lifetime prediction model based on the lifetime prediction loss and multiple maximum mean difference losses to obtain a preset lifetime prediction model includes: Determine the initial weighting coefficients corresponding to the multiple maximum mean difference losses; Based on the multiple initial weight coefficients, the multiple maximum mean difference losses are weighted and summed to obtain the total maximum mean difference loss; The total loss is obtained by summing the lifetime prediction loss and the total maximum mean difference loss. Based on the total loss, the initial lifetime prediction model and the multiple initial weight coefficients are iteratively updated until a preset condition is met, at which point the preset lifetime prediction model is output. During the iterative update, the multiple initial weight coefficients are dynamically adjusted according to the total contribution of the multiple sub-source domain time series.

6. The method according to any one of claims 1-5, characterized in that, The process involves inputting the source domain time series into a preset data augmentation network for data augmentation to obtain multiple sub-source domain time series, including: The source domain time series is input into a preset data augmentation network for processing to obtain multiple candidate time series; The target time series is determined from the plurality of candidate time series; Different mathematical calculations are performed on the target time series to obtain the multiple sub-source domain time series.

7. The method according to claim 6, characterized in that, The preset data augmentation network includes an encoder and a decoder. The source domain time series is input into the preset data augmentation network for processing to obtain multiple candidate time series, including: Based on the time window corresponding to the encoder, the source domain time series is sequentially input into the encoder and the decoder for processing to obtain the multiple candidate time series.

8. A remaining lifetime prediction device based on small sample augmentation and interpretable migration contribution, characterized in that, include: An acquisition unit is used to acquire the source domain time series and target domain time series of the rotating device, as well as an initial lifetime prediction model, wherein the initial lifetime prediction model includes an interpretable estimation network and a lifetime prediction network. An enhancement unit is used to input the source domain time series into a preset data enhancement network for data enhancement, thereby obtaining multiple sub-source domain time series; The determining unit is used to determine multiple sub-source domain feature vectors and the predicted lifetime of the rotating device based on the time-frequency domain features corresponding to the multiple sub-source domain time series, the interpretable estimation network, and the lifetime prediction network, wherein the interpretable estimation network adaptively allocates weights according to the contribution of each feature in the time-frequency domain features corresponding to each sub-source domain time series. The determining unit is further configured to determine the target domain feature vector based on the time-frequency domain features corresponding to the target domain time series and the interpretable estimation network; The construction unit is used to construct lifetime prediction loss and maximum mean difference loss between the time series of the multiple sub-source domains and the time series of the target domain, respectively, based on the predicted lifetime, the feature vectors of the multiple sub-source domains and the feature vector of the target domain. An iterative unit is used to iteratively train the initial lifetime prediction model based on the lifetime prediction loss and multiple maximum mean difference losses to obtain a preset lifetime prediction model.

9. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

10. An electronic device comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

Images