A cross-operating condition fault diagnosis method based on uncertainty estimation and denoising

Abstract

The present invention belongs to the technical field of industrial system fault diagnosis, and specifically relates to a cross-operating condition fault diagnosis method based on uncertainty estimation and denoising. The specific steps are as follows: First, respectively obtain labeled source domain data and unlabeled target domain data, build a feature extractor based on a one-dimensional convolutional neural network and a Transformer layer, and extract features from fault signals; second, perform self-supervised learning on the target domain data to obtain pseudo labels; third, build a pseudo label redefinition selection module to dynamically select high-confidence pseudo labels and reduce noisy pseudo labels; fourth, build a class prototype alignment strategy based on Dirichlet uncertainty estimation to achieve accurate recognition of target domain samples; fifth, load the source domain data and high-confidence target domain data into a dual-classifier adversarial network model for training to obtain a trained neural network model; sixth, use the trained neural network model to realize fault diagnosis of rotating machinery systems.

Description

Technical Field

[0001] The present invention belongs to the technical field of industrial system fault diagnosis, and in particular relates to a cross-operating-condition fault diagnosis method based on uncertainty estimation and denoising. Background Art

[0002] With the continuous advancement of industrial development, the wave of industrial intelligence is sweeping across the globe. Rotating machinery, as an indispensable component in industrial applications, often operates in unstable and harsh working environments, making it prone to various failures. Therefore, the health monitoring and fault diagnosis of rotating machinery has become a research hotspot in industrial intelligent operation and maintenance.

[0003] In recent years, deep learning technology has dominated the field of rotating machinery fault diagnosis thanks to its superior feature extraction capabilities. However, in real-world industrial applications, rotating machinery operates for extended periods under various operating conditions (such as speed and load variations), resulting in significant discrepancies in data from different operating conditions. This severely hinders the practical application of deep learning models. To address these challenges and improve diagnostic performance, unsupervised domain adaptation fault diagnosis methods have emerged and garnered widespread attention.

[0004] Unsupervised domain adaptation techniques for cross-domain fault diagnosis are primarily based on statistical metrics, domain adversarial learning, and pseudo-label learning. While traditional adaptive methods have demonstrated promising performance, they primarily measure similarities between samples through differences or extract domain-invariant features through adversarial learning. However, this does not guarantee accurate and diverse classification of target samples, as ambiguous target samples may be misclassified due to irrational decision boundaries, leading to inaccurate alignment of class distributions during subsequent learning. This issue is primarily due to the fact that previous domain adaptation methods did not fully consider the accuracy of target samples and the impact of label noise in pseudo-labels on model generalization performance. Summary of the Invention

[0005] In view of the above problems, the present invention proposes a cross-operating condition fault diagnosis method based on uncertainty estimation denoising to solve the problem of poor generalization caused by the previous methods that do not fully consider the accuracy of target domain samples and noisy pseudo labels.

[0006] The technical solutions of the present invention are as follows:

[0007] A cross-operating-condition fault diagnosis method based on uncertainty estimation and denoising includes the following steps:

[0008] Step 1: Obtain source domain data and target domain data, build a feature extractor based on a one-dimensional convolutional neural network and a Transformer layer, load the source domain and target domain data into the feature extractor to extract features of the fault signal;

[0009] Step 2: Perform self-supervised clustering on the target domain data to obtain pseudo labels for the target domain samples;

[0010] Step 3: Build a pseudo-label redefinition selection module to dynamically select high-confidence pseudo-labels and reduce noisy data and labels;

[0011] Step 4: Construct a class prototype alignment strategy based on Dirichlet uncertainty estimation to achieve accurate recognition of target domain samples;

[0012] Step 5: Load the source domain data and high-confidence target domain data into the dual-classifier adversarial network model for training to obtain a trained neural network model;

[0013] Step 6: Use the trained neural network model to diagnose mechanical faults.

[0014] Among them, step 1 is specifically implemented according to the following steps:

[0015] Obtain source domain data separately and target domain data ,in, Represented as the first data, is its corresponding label, Represented as the first data, and Represents the number of samples in the source domain and the target domain respectively; constructs a feature extractor based on a one-dimensional convolutional neural network and a Transformer layer , the source domain and target domain data are simultaneously input into the feature extractor to effectively capture the global and local features of the fault signal; after the feature extractor, two classifiers are connected in parallel and Conduct adversarial optimization training.

[0016] Step 2 is implemented as follows:

[0017] In the initial training phase, the model quickly fits the source domain data. Subsequently, the target domain data is weighted using the two softmax function outputs of the dual-classifier adversarial network to obtain the centroid of each class. The distance between the target sample and the centroid is quantified using the cosine distance function, and the corresponding pseudo-label is obtained using the nearest centroid strategy.

[0018] In step 3, please follow the steps below:

[0019] Construct a pseudo-label redefinition selection module to redefine the pseudo-labels obtained in step 2 by aggregating neighbor knowledge, and update the temporary pseudo-labels of all target samples in the iterative process. By updating the iterations to perceive the density of the surrounding pseudo-labels and obtain a more accurate uncertainty estimate, the confidence score of each target domain sample is obtained. In the early stage of training, simple samples in the target domain are more likely to be assigned high confidence, so by setting a dynamic threshold , dynamically select high-confidence target domain samples from easy to difficult, specifically expressed as:

[0020]

[0021] Where, Represents the low threshold, Represents the high threshold, Represents the current iteration number, Expressed as the total number of iterations.

[0022] Step 4 is implemented as follows:

[0023] Construct a class prototype alignment strategy based on Dirichlet uncertainty estimation, use variational Dirichlet distribution to model the class probability distribution, and after obtaining the Dirichlet distribution, introduce subjective logic to assign a belief quality to each class. , and assign an uncertainty estimate , and further introduce this uncertainty into the class prototype alignment. Specifically, each training batch The class prototype is calculated as follows:

[0024]

[0025] Where, is the domain indicator, represents the source domain, represents the target domain, Indicates the samples from the source domain or the target domain, Indicates the Samples from the source domain or target domain pass through the feature extractor The feature representation calculated later, is the indicator function, It is The sample uncertainty value, is the uncertainty threshold, represents the true label in the source domain and the reliable pseudo label obtained in the target domain through step 3, Indicates that if The first in the category The sample uncertainty value is less than Then its value is 1, otherwise it is 0;

[0026] Through a dynamic dictionary To store the class prototypes from the previous batch and update them in a moving average manner, specifically expressed as:

[0027]

[0028] Where, Is a trade-off parameter; after obtaining the class prototypes of the source domain and the target domain, the loss of alignment between the samples in the execution domain and the class prototype is calculated. and inter-domain prototype alignment loss , specifically expressed as:

[0029]

[0030]

[0031] Where, Indicates the first The number of class samples, Indicates the source domain The class prototype of the class, Indicates the target domain The class prototype of the class.

[0032] Class prototype alignment loss based on Dirichlet uncertainty estimation Combining source domain samples to source domain class prototype alignment loss , target domain sample to target domain class prototype alignment loss and inter-domain prototype alignment loss , specifically expressed as follows:

[0033] .

[0034] In step 5, please follow the steps below:

[0035] The source domain data and the denoised target domain data are loaded into the dual-classifier adversarial network model for adversarial training, specifically:

[0036] Step 5.1 uses the labeled source domain data and reliable samples and pseudo labels obtained in the target domain to train the entire network:

[0037]

[0038] Where, is a trade-off parameter, 、 and Represents feature extractors and two classifiers 、 Parameters, represents the source domain sample set, represents the label set corresponding to the source domain samples, represents a high confidence sample set of the target domain, The pseudo label set corresponding to the high confidence samples of the target domain. Cross entropy loss For supervised training, classification loss of source domain samples The calculation is as follows:

[0039]

[0040] In order to utilize unlabeled target domain samples, the self-supervised learning mechanism is used to strengthen the dual classification adversarial learning, and the weighted cross entropy loss of self-supervised learning is used. The calculation is as follows:

[0041]

[0042] Where, is the weighted cross entropy loss function, which is formulated as follows:

[0043]

[0044]

[0045] Where, represents the standard information entropy, represents the softmax function, represents the first Reliable pseudo labels for samples.

[0046] Step 5.2 Fix the feature extractor , update both classifiers simultaneously and , in order to minimize the classification loss of the two classifiers in the source domain and maximize the judgment of the target sample:

[0047]

[0048] Where, Indicates difference loss, usually Norm to measure, Represents the classifier for the target sample The difference in discrimination;

[0049] Step 5.3 Combine the class prototype alignment loss based on Dirichlet uncertainty estimation between the two domains , thus obtaining the final optimization goal, which is expressed as:

[0050]

[0051] Where, is a trade-off parameter.

[0052] The beneficial effects of the present invention are:

[0053] (1) This paper proposes a cross-operating fault diagnosis method based on uncertainty estimation and denoising. By constructing a feature extractor based on a one-dimensional convolutional neural network and a Trasformer layer, the global and local features of the fault signal can be effectively extracted.

[0054] (2) This paper proposes an uncertainty-based pseudo-label redefinition selection module. This strategy selects high-confidence samples as clean samples, reduces the impact of noisy pseudo-labels on the generalization performance of the model, and ensures the separability of features.

[0055] (3) This paper proposes a class prototype alignment strategy based on Dirichlet uncertainty estimation. By performing intra-domain sample-to-class prototype alignment and inter-domain class prototype alignment, the fault recognition accuracy and generalization ability of target domain samples are effectively improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 Schematic diagram of the overall framework, i.e., the abstract;

[0057] Figure 2 Algorithm principle diagram;

[0058] Figure 3 Cross-domain fault diagnosis tasks Pseudo label accuracy curve of ;

[0059] Figure 4 Cross-domain fault diagnosis tasks Pseudo label accuracy curve of ;

[0060] Figure 5 Cross-domain fault diagnosis tasks 、 and Below are the feature visualizations of different methods;

[0061] Figure 6 Cross-domain fault diagnosis tasks The following is a comparison of the feature visualization diagrams of DASAN and the present invention;

[0062] Figure 7 Cross-domain fault diagnosis tasks Below is a visualization diagram of the features of the present invention;

[0063] Figure 8 Cross-domain fault diagnosis tasks Below is a visualization diagram of the features of the present invention;

[0064] Figure 9 Cross-domain fault diagnosis tasks Below is a visualization diagram of the features of the present invention; DETAILED DESCRIPTION

[0065] The cross-operating-condition fault diagnosis method based on uncertainty estimation and denoising proposed in this invention consists of a feature extraction module and a pseudo-label redefinition selection module. The specific flow chart is as follows: Figure 1 As shown. The function of the feature extractor is to extract fault features from the original signals of the source domain and the target domain. The prototype alignment term is introduced after the feature extractor module to further align the distribution and improve the accuracy of the target samples. The pseudo-label redefinition module aims to select high-confidence target domain samples and dynamically reduce noise labels, ultimately achieving accurate identification of fault types under variable working conditions. In order to enable those skilled in the art to better understand the scheme of the present invention, exemplary implementation methods or embodiments of the present invention will be described below in conjunction with the accompanying drawings.

[0066] 1. Model Building

[0067] 1.1 Feature Extractor

[0068] The feature extractor aims to extract high-level feature representations of the machinery health status from data in the source and target domains. Convolutional neural networks are widely used as feature extractors for machinery fault diagnosis because they have significant advantages in capturing local features. However, due to their lack of ability to capture global features and obvious contextual relationships in fault signals, it is necessary to fully consider both global and local features of fault signals. This paper designs a feature extractor based on a one-dimensional convolutional neural network and a Transformer. , in order to take into account both the global and local characteristics of the fault signal, and to improve the diagnostic performance of the model under variable working conditions. Figure 1 The principle diagram of the algorithm proposed in this invention is shown. The first convolution layer receives the one-dimensional original vibration signal As the model input. After the fourth convolution module, two Transformer layers are embedded to better capture global features. The feature extraction process of the feature extractor can be expressed as:

[0069] ,

[0070] Where, represents the extracted features, is the batch input matrix.

[0071] 1.2 Self-supervised learning of target domain samples

[0072] In the initial stage of training, the model quickly fits the source domain data. Subsequently, the two softmax functions of the dual classifier adversarial network weight the target domain data to obtain the centroid of each class. , specifically expressed as:

[0073]

[0074] Where, It is represented as a feature extractor and F is a classifier, where Indicates the corresponding first The target sample is quantified by the cosine distance function with the distance between itself and the cluster center of each fault mode, and the corresponding pseudo label is obtained by the nearest centroid strategy, which is specifically expressed as:

[0075] .

[0076] 1.3 Uncertainty-based Pseudo-label Redefinition Selection Module

[0077] Construct a pseudo-label redefinition selection module to redefine the pseudo-label obtained in step 2 by aggregating neighbor knowledge. The specific expression is:

[0078]

[0079] Where, Represents a temporary pseudo label A one-dimensional vector of temporary pseudo labels Pseudo labels obtained through self-supervised learning Initialize and update the temporary pseudo labels of all target samples during the iteration , specifically expressed as:

[0080]

[0081] By updating To iterate To perceive the density of the surrounding pseudo labels and obtain a more accurate uncertainty estimate, the confidence score is obtained after the uncertainty estimate is completed. , specifically expressed as:

[0082]

[0083] For the sample , only its The nearest neighbors sharing the same label are denoted as , will be right Make a contribution.

[0084] In the early stages of training, simple samples in the target domain are more likely to be assigned high confidence and learned by the model, so a dynamic threshold is set , select high-confidence target domain samples from easy to difficult, specifically expressed as:

[0085]

[0086] Where, Represents the low threshold, Represents the high threshold, Represents the current iteration number, Expressed as the total number of iterations.

[0087] In order to utilize unlabeled samples, the self-supervised learning mechanism is combined to strengthen the dual classification adversarial learning, and the weighted classification loss of the self-supervised learning is It is expressed as follows:

[0088]

[0089] Where, is the weighted cross entropy loss function, which is calculated as follows:

[0090] ,

[0091]

[0092] Where, represents the standard information entropy, Represents the softmax function.

[0093] 1.4 Class Prototype Alignment Strategy Based on Dirichlet Uncertainty Estimation

[0094] A class prototype alignment strategy based on Dirichlet uncertainty estimation is constructed, and the class probability distribution is modeled using variational Dirichlet distribution. Specifically, for Classification problem, assuming that the estimated classification probability is ,in The Dirichlet distribution is Concentration parameters Parameterized, the probability density function of the distribution is specifically expressed as:

[0095]

[0096] Where, yes dimensional polynomial beta function, yes dimensional simplex, defined as:

[0097]

[0098] After obtaining the Dirichlet distribution, in order to quantify its uncertainty, subjective logic is introduced to associate the parameters of the Dirichlet distribution with the belief quality and uncertainty. Classification problem, subjective logic assigns belief quality to each class based on Dirichlet distribution , and assign an overall uncertainty mass to all categories ,Right now .

[0099] Given Dirichlet distribution, subjective logic and evidence classification, divide the evidence into and the concentration parameter of the Dirichlet distribution Evidence refers to the indicator of support for classification obtained from the input, which is closely related to the expected concentration parameter of the Dirichlet distribution. Specifically, let Indicates the Evidence of this kind, is the Dirichlet distribution concentration parameters, and The relationship between , then belief and uncertainty is calculated as follows:

[0100]

[0101] Where, Represents the Dirichlet intensity. Since low uncertainty samples have better generalization ability, this uncertainty is introduced into class prototype alignment. Therefore, the present invention proposes an uncertainty class prototype alignment strategy based on Dirichlet uncertainty estimation for aligning class-level features. Specifically, each training batch The class prototype is calculated as follows:

[0102]

[0103] Where, is the domain indicator, represents the source domain, represents the target domain, Indicates the samples from the source domain or the target domain, Indicates the Samples from the source domain or target domain pass through the feature extractor The feature representation calculated later, is the indicator function, It is The sample uncertainty value, is the uncertainty threshold, represents the true label in the source domain and the reliable pseudo label obtained in the target domain through step 3, Indicates that if The first in the category The sample uncertainty value is less than Then its value is 1, otherwise it is 0;

[0104] By building a dynamic dictionary To store the class prototypes from the previous batch and update them in a moving average manner, specifically expressed as:

[0105]

[0106] Where, Is a trade-off parameter. By calculating the class prototype in the two domains and performing the domain sample to class prototype alignment loss and inter-domain class prototype alignment loss, In order to improve the model's ability to recognize target domain samples, the specific loss function is expressed as:

[0107]

[0108]

[0109] Where, Indicates the first The number of class samples, Indicates the source domain The class prototype of the class, Indicates the target domain The class prototype of the class.

[0110] Class prototype alignment loss based on Dirichlet uncertainty estimation Combining source domain samples to source domain class prototype alignment loss , target domain sample to target domain class prototype alignment loss and inter-domain prototype alignment loss , specifically expressed as follows:

[0111]

[0112] 1.5 Optimization process

[0113] The source domain data and the denoised target domain data are loaded into the dual-classifier adversarial network model for adversarial training, specifically:

[0114] Train the entire network using labeled source domain data and obtaining high confidence samples in the target domain:

[0115]

[0116] Where, is a trade-off parameter, 、 and Represents feature extractors and two classifiers 、 Parameters, represents the source domain sample set, represents the label set corresponding to the source domain samples, represents a high confidence sample set of the target domain, The pseudo label set corresponding to the high confidence samples of the target domain. Cross entropy loss For supervised training, classification loss of source domain samples The calculation is as follows:

[0117]

[0118] Fixed feature extractor , update both classifiers simultaneously and , in order to minimize the classification loss of the two classifiers in the source domain and maximize the judgment of the target sample:

[0119]

[0120] Where, Indicates difference loss, usually Norm to measure, Represents the classifier for the target sample The difference in discrimination;

[0121] Combining a class prototype alignment loss based on Dirichlet uncertainty estimation between two domains , thus obtaining the final optimization goal:

[0122]

[0123] Where, is a trade-off parameter.

[0124] 2. Experimental Verification

[0125] To demonstrate the superiority of the proposed method, we conducted a detailed experimental validation study. Seven models, including Correlation Alignment (CORAL), Domain Adversarial Neural Network (DANN), Deep Convolutional Transfer Learning Network (DCTLN), Joint Maximum Mean Discrepancy (JAN), Multi-Kernel Maximum Mean Discrepancy (MKMMD), Maximum Classifier Discrepancy (MCD), and Deep Adversarial Subdomain Adaptation Network (DASAN), were selected for comparison with the proposed method to obtain more convincing results. To ensure a fair comparison, the feature extractors, health classifiers, and optimizers used in these models remained consistent with those used in the proposed method. The network parameters are detailed in Table 1.

[0126] Table 1

[0127]

[0128] 2.1 Data Description

[0129] The present invention uses a bearing dataset from Paderborn University (PU) in Germany for experimental verification. The dataset contains real damage caused by both artificial damage and accelerated life. The test bench consists of a motor and a bearing seat with four 6203 bearings. By controlling the three parameters of speed, load torque, and radial force of the bearing, the bearing vibration signals under four working conditions are collected. The signal sampling frequency is 64kHz. The working condition parameters are shown in Table 2. In this embodiment, the bearing vibration signal is collected under four working conditions. Indicates that the migration task is migrated from working condition 0 to working condition 1.

[0130] Table 2

[0131]

[0132] This paper uses bearings with real damage induced by accelerated life testing to study transfer learning tasks under different conditions, more closely resonating with actual industrial operating scenarios. The real damage types are categorized into three types: electromachining damage, electroengraving damage, and drilling damage. Twelve fault types are combined based on bearing type, fault location, fault severity, and damage type. The damage characteristics are summarized in Table 3. Adding a health status, thirteen categories of fault diagnosis are achieved.

[0133] Table 3

[0134]

[0135] First, the proposed method was compared with several domain adaptation methods, including DANN, MKMMD, CORAL, DCTLN, MCD, and DASAN. Each model was evaluated in 10 independent trials to obtain reliable results. The experimental results are shown in Table 4. These results reveal the following: 1) The proposed model achieved an average diagnostic accuracy of 93.19%, 6.01% higher than MCD. This demonstrates that the proposed method effectively mitigates the distribution differences between working conditions and improves diagnostic accuracy.

[0136] Table 4

[0137]

[0138] In order to verify the effectiveness of the proposed uncertainty-based pseudo-label redefinition and dynamic selection method, a cross-domain fault diagnosis task and The method proposed in this invention is compared with the prediction results of self-supervised clustering and softmax function. Figure 3 and Figure 4 As shown in the pseudo-label selection accuracy graph, the proposed algorithm reaches the maximum accuracy faster and exhibits excellent stability. This result provides strong evidence for the denoising effect of the proposed method and it more effectively promotes cross-domain alignment.

[0139] In order to further demonstrate the effectiveness of the proposed model and provide intuitive understanding, in the cross-domain fault diagnosis task 、 and The t-SNE technique is used to visualize the feature distribution learned by the model. Figure 5 The feature distribution visualization of DANN, MCD, DASAN and the method of the present invention is shown. It can be seen that compared with other methods, the present invention achieves effective alignment of the source domain and target misclassification levels, shows fewer misclassifications, and demonstrates better diagnostic accuracy and generalization performance. Figure 6 A zoomed-in visualization of the proposed method and DASAN is shown. The results demonstrate that the proposed method, based on a class prototype alignment strategy based on Dirichlet uncertainty estimation, selects samples with low uncertainty and representative class characteristics as class prototypes. Compared to DASAN, the proposed method better aligns target and source samples at the class level and establishes clearer class boundaries, while DASAN exhibits significant aliasing. Figure 7 、 Figure 8 and Figure 9It is shown that the proposed method better aligns the distribution of target samples with source samples at the category level by reducing the noise in the target domain, resulting in clearer classification boundaries and higher diagnostic accuracy.

[0140] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It should be understood that the present invention is not limited to the implementation schemes described herein. The purpose of describing these implementation schemes is to help those skilled in the art to practice the present invention. Any person skilled in the art will readily make further improvements and perfections without departing from the spirit and scope of the present invention. Therefore, the present invention is limited only by the content and scope of the claims, which are intended to cover all alternatives and equivalents included in the spirit and scope of the present invention as defined by the appended claims.

Claims

1. A cross-operating-condition fault diagnosis method based on uncertainty estimation and denoising, characterized in that: The specific implementation steps are as follows: Step 1: Obtain source domain data and target domain data respectively, build a feature extractor based on a one-dimensional convolutional neural network and a Transformer layer, load the source domain and target domain data into the feature extractor to extract features of the fault signal; Step 2: Perform self-supervised learning clustering on the target domain data to obtain pseudo labels for the target domain samples; Step 3: Build a pseudo-label redefinition selection module to dynamically select high-confidence pseudo-labels and reduce noisy data and labels; Step 4: Construct a prototype alignment strategy based on Dirichlet uncertainty estimation; Step 5: Load the source domain data and high-confidence target domain data into the dual-classifier adversarial network model for training. During training, the prototype alignment loss based on Dirichlet uncertainty estimation is combined to obtain a trained neural network model. Step 6: Use the trained neural network model to perform fault diagnosis of rotating machinery systems.

2. The cross-operating-condition fault diagnosis method based on uncertainty estimation and denoising according to claim 1 is characterized in that: In step 1, obtain the source domain data separately and target domain data ,in, Feature extractor based on one-dimensional convolutional neural network and Transformer layer , the source domain and target domain data are simultaneously input into the feature extractor to effectively capture the global and local features of the fault signal; after the feature extractor, two classifiers are connected in parallel and Conduct adversarial optimization training.

3. The cross-operating-condition fault diagnosis method based on uncertainty estimation and denoising according to claim 1 is characterized in that: In step 2, at the initial stage of training, the model quickly fits the source domain data; then, the target domain data is weighted using the two softmax function outputs of the dual-classifier adversarial network to obtain the centroid of each class, and the distance between the target sample and the centroid is quantified using the cosine distance function and the corresponding pseudo label is obtained using the nearest centroid strategy.

4. The cross-operating-condition fault diagnosis method based on uncertainty estimation and denoising according to claim 1 is characterized in that: In step 3, a pseudo-label redefinition selection module is constructed to redefine the pseudo-labels obtained in step 2 by aggregating neighbor knowledge, and to update the temporary pseudo-labels of all target samples in the iterative process. The density of the surrounding pseudo-labels is perceived through the update iteration and a more accurate uncertainty estimate is obtained, thereby obtaining the confidence score of each target domain sample. In the early stage of training, simple samples in the target domain are more likely to be learned by the model with high confidence, so a dynamic threshold is set. , dynamically select high-confidence target domain samples from easy to difficult, specifically expressed as: Expressed as the total number of iterations.

5. The cross-operating-condition fault diagnosis method based on uncertainty estimation and denoising according to claim 1 is characterized in that: In step 4, a class prototype alignment strategy based on Dirichlet uncertainty estimation is constructed, and the class probability distribution is modeled using variational Dirichlet distribution. After obtaining the Dirichlet distribution, subjective logic and evidence classification are introduced to assign a belief quality to each class. , and assign an uncertainty estimate , and further introduce this uncertainty into class prototype alignment. Specifically, each training batch The class prototype is calculated as follows: Then it is 0; The update is performed in the following way, which is specifically expressed as follows:

6. The cross-operating-condition fault diagnosis method based on uncertainty estimation and denoising according to claim 1 is characterized in that: In step 5, the source domain data and the denoised target domain data are loaded into the dual-classifier adversarial network model for adversarial training, specifically: Step 5.1 trains the entire network using labeled source domain samples and high confidence samples in the target domain: In order to utilize unlabeled target domain samples, the self-supervised learning mechanism is used to strengthen the dual classification adversarial learning, and the weighted cross entropy loss of self-supervised learning is used. The calculation is as follows: The classification loss of the detector in the source domain and the maximum judgment of the target sample are expressed as: Step 5.3 Combine the class prototype alignment loss based on Dirichlet uncertainty estimation between the two domains Thus, the final optimization goal is obtained, which is expressed as: .

Images