A cross-multi-source domain industrial fault diagnosis method based on generalized zero-shot learning

Abstract

The application relates to a multi-source domain industrial fault diagnosis method based on generalized zero sample learning and belongs to the field of industrial fault diagnosis. The method realizes fault diagnosis of source domains and target domains through joint learning of multi-classifiers and domain alignment consistency, and constructs a latent hypersphere space with orthogonal constraints for connecting a feature space and a semantic attribute space, so that discriminative information is extracted and visible and unseen fault diagnosis is realized. The application realizes efficient and accurate diagnosis of visible and unseen faults in the case of missing unseen samples, and improves the generalization performance and practicability of a model.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of industrial fault diagnosis, and relates to a cross-multi-source domain industrial fault diagnosis method based on generalized zero-shot learning. BACKGROUND

[0002] Industrial fault diagnosis is one of the important technologies to ensure the stable, efficient and safe operation of industrial systems. With the increasing complexity of industrial equipment and systems, traditional model and knowledge-based diagnosis methods are gradually replaced by data-driven technologies. In recent years, machine learning methods, especially deep learning, transfer learning and zero-shot learning (ZSL) technology, have been widely applied in industrial fault diagnosis tasks. However, due to the fact that factories cannot allow equipment to run to a fault state to obtain fault samples for training diagnosis models, and it is very difficult to obtain enough labeled data, traditional supervised learning methods have great limitations in practical applications.

[0003] ZSL, as an advanced learning method, provides a new solution to the problem of data deficiency by predicting unseen samples through semantic attributes. Although ZSL methods perform well in the field of computer vision, their application in industrial scenarios still faces many challenges. The industrial field usually involves multiple working conditions and environmental noise, and existing ZSL methods have limitations in cross-domain generalization. Especially in the task of multi-source domain fault diagnosis and generalized zero-shot setting, how to realize effective knowledge transfer between different domains and ensure the generalization ability of the model has become the focus of current technology development.

[0004] Most of the existing industrial fault diagnosis methods adopt a supervised learning mode, and train the model through a large amount of historical data. For the diagnosis problem of few samples or unseen faults, transfer learning and zero-shot learning technology are gradually introduced. ZSL belongs to the category of machine learning, which predicts the zero-shot fault that has not been encountered in the training stage by using the prior knowledge (such as semantic attributes) of visible faults. Patent CN112763214B discloses a rolling bearing fault diagnosis method based on multi-label zero-shot learning, which opens up the field of fault diagnosis under zero-shot conditions, and effectively diagnoses the unseen class fault by migrating the visible class fault attributes to the unseen class. Patent CN114383845B discloses a bearing composite fault diagnosis method based on an embedded zero-shot learning model, which uses an embedded zero-shot learning model for composite fault diagnosis. However, under the setting of generalized zero-shot, these methods are prone to misclassifying unseen faults as visible faults. To solve the problem of diagnosing visible faults and unseen faults at the same time, patent CN113609569B discloses a discriminative generalized zero-shot learning fault diagnosis method, which divides the generalized zero-shot diagnosis task into supervised learning and zero-shot learning tasks by discriminating fault samples. However, the existing methods have not considered the distribution difference problem between the actual working condition source domain and the target domain, which limits the generalization ability under multiple source domains and different working conditions.

[0005] In summary, the existing technology mainly has the following shortcomings in the multi-source domain fault diagnosis task:

[0006] (1) Poor cross-domain adaptability: existing methods assume that the source domain and the target domain have similar distributions, which makes it difficult to adapt to the distribution difference in actual industrial applications.

[0007] (2) Single source domain dependence: Most fault diagnosis methods focus on a single source domain, and fail to effectively utilize multi-source domain data for diagnosis knowledge transfer and fusion, resulting in insufficient generalization ability of the model under different working conditions.

[0008] (3) Poor generalization in generalized scenarios: due to the lack of unseen fault samples, existing zero-shot fault diagnosis methods perform poorly in more practical generalized scenarios, especially in the classification task of unseen faults. SUMMARY

[0009] In view of this, the purpose of the present application is to provide a cross multi-source domain industrial fault diagnosis method based on generalized zero-shot learning, which solves the problem of insufficient generalization ability of existing industrial fault diagnosis technology under multi-source domain conditions, especially the limitation that it cannot effectively distinguish between visible faults and unseen faults in the generalized zero-shot fault diagnosis task. Existing methods usually rely on the same distribution assumption between source and target domains, and are limited to single-source domain scenarios, making it difficult to deal with distribution differences under different working conditions. The present application proposes an unsupervised framework that overcomes the limitations of existing technology in terms of insufficient ability to diagnose unseen faults under multi-source domain conditions by transferring knowledge from multiple source domains and mapping the latent space, thereby achieving accurate diagnosis of visible and unseen faults in the absence of unseen samples, and improving the generalization performance and practicality of the model.

[0010] To achieve the above purpose, the present application provides the following technical solutions:

[0011] A cross multi-source domain industrial fault diagnosis method based on generalized zero-shot learning is provided to achieve efficient and accurate classification of visible and unseen faults in industrial fault diagnosis tasks, especially in the case of differences in multi-source domain data distribution. Through joint learning of multi-classifier and domain alignment consistency, fault diagnosis of source and target domains is achieved, and a latent hypersphere space with orthogonal constraints is constructed to connect the feature space and semantic attribute space, thereby extracting discriminative information and achieving diagnosis of visible and unseen faults.

[0012] The method specifically includes the following steps:

[0013] S1: joint learning of multi-classifier and cross-domain alignment consistency;

[0014] An independent classifier is constructed for each source domain, and distribution consistency between source and target domains is achieved by aligning source domain features with target domain unlabeled sample features. The feature maximum mean difference method is used for domain alignment strategy to ensure that the feature distribution of the source and target domains and the inter-domain distribution between multiple domains are unified in the same feature space;

[0015] S2: fault prototype and semantic embedding learning;

[0016] A latent hypersphere space with orthogonal constraints is constructed for mapping learning between feature space and semantic attribute space; for different source domains of visible faults, learning is performed by obtaining corresponding sample labels in the training phase; through domain alignment technology, all samples belonging to a specific category are averaged to obtain the prototype representation of the category fault; in the context of generalized zero-shot fault diagnosis, a semantic similarity-based strategy is adopted to determine the semantic attribute description of the similar visible fault by calculating the Euclidean distance, thereby realizing effective reasoning of unseen faults; by learning the mapping containing information of unseen faults, the connection between feature space and latent space and attribute space and latent space is realized respectively;

[0017] S3: comprehensive fault diagnosis;

[0018] Based on visible faults, a model capable of cross-source domain classification is learned, and a classifier with strong generalization ability is trained for unlabeled target domains through domain alignment; this method respectively diagnoses the test samples containing visible faults and unseen faults in the feature space, latent space and fault attribute space, and simultaneously obtains the diagnosis results by introducing visible information in the multi-source domain learning process and adopting a comprehensive diagnosis strategy.

[0019] Further, in step S1, the multi-classifier learning specifically includes: for the kth source domain, wherein there are c fault samples and c fault categories, the classifier C k is continuously optimized by calculating the classification loss J; the formula of the loss function is as follows:

[0020]

[0021] wherein, y i is the label of the ith sample x i , S is the visible fault category; I S (y ij ) is the jth value in the one-hot encoding vector of y i , which can be regarded as an indicator function; if y i = j, then I S (y ij ) = 1; otherwise I S (y ij ) = 0; w j is the weight connecting the feature f(x i , ω) and the mth output neuron in the fault classifier C k , and the superscript T represents transposition; subsequently, the multi-classification learning loss L cls can be represented as:

[0022]

[0023] ​where K is the number of classifiers, is the loss function of the i-th classifier C k .

[0024] Further, in step S2, the fault prototype and semantic embedding learning specifically comprises: a fault prototype FP i s can be obtained by the following way:

[0025]

[0026] where m i is the number of samples of the i-th visible class fault in all source domains, X s is the sample space of the visible class;

[0027] For the j-th unseen class fault, the first l (1

[0028]

[0029] where Dis is the Euclidean distance function, is the j-th unseen class semantic attribute vector, A s is the visible class attribute matrix, and rank(t, Dis) represents the ascending order of t in Dis;

[0030] The j-th unseen class fault prototype FP j u can be expressed as:

[0031]

[0032] where FP s (r(i)) is the fault prototype of the i-th similar visible attribute.

[0033] Further, in step S3, the comprehensive diagnosis strategy specifically comprises: setting a weight distance set as wherein, is the weight (normalized) distance between the feature of the i-th sample in the feature space and the fault prototype of the j-th fault, is the weight distance between the feature of the i-th sample in the latent space and the fault prototype of the j-th fault, is the weight distance between the feature of the i-th sample in the attribute space and the fault prototype of the j-th fault, n te is the number of visible class fault test samples, and q is the number of unseen classes; the comprehensive label set can be represented by the corresponding label value in the minimum distance space; meanwhile, it is necessary to ensure that the label value is consistent with the classification result C[f(x i[Maintain consistency; sample x] i tags It can be represented as follows:

[0034]

[0035] in, θ ti,j This represents the t-th element of vector θ when i and j are determined; It is the predicted label of the i-th sample in the feature space. It is the predicted label of the i-th sample in the latent space. It is the predicted label of the i-th sample in the attribute space; the predicted label It is expressed as follows:

[0036]

[0037] in, and It is the normalized weighted distance between the features of the i-th sample in the feature space and the fault prototype of the j-th type of fault. It is the normalized weighted distance between the features of the i-th sample in the latent space and the fault prototype of the j-th type of fault. It is the normalized weighted distance between the features of the i-th sample in the attribute space and the fault prototype of the j-th type of fault; It represents the number of test samples for unseen fault types; · indicates FP uT H u and A uT , * represent FP respectively uT H u and A uT I, B and D T B, FP uT H u and A uT Let f represent the prototype of the unseen fault class, the latent space of the unseen fault class, and the attributes of the unseen fault class, respectively; I represents the identity matrix; B represents the mapping between the feature space and the latent space; and D represents the mapping between the attribute space and the latent space. k This represents the feature of the k-th test sample.

[0038] Furthermore, in step S3, the predicted label of the test sample The expression is:

[0039]

[0040] Among them, f i The features of the i-th test sample are n. te It is the number of test samples; the jth column vector of the matrix FP (i.e., FP of the jth type of fault); T is the weight (normalized) distance between the feature of the ith sample in the feature space and the fault prototype of the jth type of fault; Y is a set of fault types.

[0041] Further, in step S3, in the latent space, the predicted label can be obtained by the learned mapping B:

[0042]

[0043] wherein, is the predicted label in the latent space, f i is the feature of the ith test sample, n te is the number of test samples; H j is the latent space of the jth fault category; Y is a set of fault types.

[0044] Further, in step S3, in the fault attribute space, by the mappings B and D, the predicted label can be obtained as follows:

[0045]

[0046] wherein, is the predicted label in the fault attribute space, f i is the feature of the ith test sample, n te is the number of test samples; is the attribute matrix of the ith fault category; Y is a fault category label; Y is a set of fault types.

[0047] The present application has the following beneficial effects:

[0048] (1) Performance improvement: The unsupervised generalized zero-shot fault diagnosis framework proposed by the present application effectively improves the fault diagnosis performance under multiple source domains. Without relying on the same distribution assumption, accurate diagnosis of visible and unseen faults can be achieved under different working conditions (multiple source domains). By combining cross-domain alignment and semantic attribute space mapping, the performance and accuracy of fault diagnosis are significantly improved, especially in the case of facing larger data distribution differences, the efficient fault diagnosis capability can still be maintained.

[0049] (2) Wide applicability: The method of the present application has good generalization ability under different devices and various working conditions, effectively solving the problem of failure of traditional methods under different working conditions. The existing technology usually assumes that the training and test data distribution is the same, which limits the application scene of the model. However, the present application overcomes the distribution difference problem between different domains through alignment and knowledge transfer between multiple source domains, and is suitable for more diversified industrial scenes.

[0050] ​(3) Cost advantage brought by unsupervised learning: Existing technologies rely heavily on a large amount of labeled data, while in industrial scenarios, the cost of obtaining labeled data is extremely high, especially in the case of unseen faults. The present application uses unsupervised learning to reduce the dependence on manual labeling, reducing the cost of data labeling while ensuring diagnostic performance. This label-free diagnostic capability makes the method more economical and efficient in actual industrial applications.

[0051] (4) Improved accuracy and reliability of fault diagnosis: Through the construction of the orthogonal constrained latent hypersphere space, the present application can extract more discriminative representations between the feature space and the semantic attribute space. The discriminative nature of this representation ensures accurate identification of different fault categories and effectively avoids misclassification of unseen faults as visible faults, significantly improving the accuracy and reliability of diagnosis.

[0052] (5) Expanded application scope: The present application is not only suitable for single equipment or working conditions, but also can be applied in complex industrial scenarios with multiple source domains, adapting to different equipment, working conditions, and even different factories. Through cross-source domain learning, the present application can achieve wider adaptability, greatly expanding its application scope.

[0053] (6) Specific effects of achieving the purpose of the invention: Through unsupervised learning across multiple source domains and comprehensive diagnostic strategies, the present application realizes effective knowledge transfer, successfully applying diagnostic knowledge from multiple source domains to the target domain. Even in the case of scarce or no labeled data in the target domain, it can still complete efficient fault diagnosis tasks, meeting the actual needs of industry and demonstrating significant economic value and application prospects.

[0054] Other advantages, objects, and features of the present application will be set forth in part in the description which follows, and in part will become apparent to those skilled in the art upon examination of the following or can be learned by practice of the present application. The objects and other advantages of the present application can be realized and attained by the embodiments particularly pointed out in the written description and claims hereof. BRIEF DESCRIPTION OF DRAWINGS

[0055] In order to make the purpose, technical solutions and advantages of the present application clearer, the preferred detailed description of the present application will be made below in combination with the drawings, in which:

[0056] Fig. 1 Schematic diagram of the generalized zero-shot fault diagnosis framework of the present application;

[0057] Fig. 2 Schematic diagram of a closed-loop continuous stirred reactor;

[0058] Fig. 3 Fault semantic attribute matrix of a continuous stirred reactor. DETAILED DESCRIPTION

[0059] The present application is described herein with reference to particular non-limiting embodiments. Various modifications and changes can be made thereto by those skilled in the art without departing from the spirit and scope of the application as set forth in the claims. The disclosure is not to be limited to the specific embodiments described but is intended to cover any and all changes and modifications of the application unless otherwise specified. There are many examples of embodiments provided below, and features of the examples can be combined with each other as appropriate, without departing from the spirit of the application.

[0060] The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings:

[0061] The same or similar components in the drawings of the embodiments of the present application correspond to the same or similar components; in the description of the present application, it should be understood that if the terms "upper", "lower", "left", "right", "front", "back" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, therefore the terms describing the positional relationship in the drawings are only for illustrative purposes, and cannot be understood as a limitation of the present application, for those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0062] Referring to Figs. 1-3 , the present application provides a cross multi-source domain industrial fault diagnosis method based on generalized zero sample learning, and the specific process is as follows:

[0063] Step one: multi-classifier learning;

[0064] For the kth source domain, which contains fault samples and c fault categories, the classifier C k is continuously optimized by calculating the classification loss J. The formula of the loss function is as follows:

[0065]

[0066] Where y i is the label of the ith sample x i . I S (y ij ) is y ithe jth value in the one-hot encoding vector of y, which can be regarded as an indicator function. If y i = j, then I S (y ij ) = 1; otherwise I S (y ij ) = 0. w j is the weight connecting the feature f(x i , ω) and the mth output neuron in the fault classifier C k . Subsequently, the multi-classification learning loss can be represented as:

[0067]

[0068] Step 2: Fault Prototype and Semantic Embedding Learning;

[0069] The fault prototype FP i s of the ith visible class can be obtained by:

[0070]

[0071] where m i is the number of samples of the ith visible class fault in all source domains.

[0072] For the jth unseen class fault, the first l (1 < l ≤ c) similar visible attribute descriptions can be represented as:

[0073]

[0074] where Dis is the Euclidean distance function, and rank(t, Dis) represents the ascending order of t in Dis. The jth unseen class fault prototype can be represented as:

[0075]

[0076] Step 3: Fault Comprehensive Diagnosis;

[0077] The predicted label of the test sample can be represented as:

[0078]

[0079] where f i is the feature of the ith test sample, and n te is the number of test samples. represents the jth column vector of the matrix FP T (i.e., the FP of the jth class fault). is the weight (normalized) distance between the feature of the ith sample in the feature space and the fault prototype of the jth class fault.

[0080] In latent space, the predicted label can be obtained by the learned mapping B:

[0081]

[0082] Similarly, in the fault attribute space, through the mapping B and D, the predicted label can be obtained as follows:

[0083]

[0084] Based on the set By introducing the visible information in the multi-source domain learning process, a diagnosis strategy is proposed. Let the weight distance set be Synthetic label set It can be represented by the corresponding label value in the minimum distance space. At the same time, it is necessary to ensure that the label value is consistent with the classification result C[f(x i )] in multi-source domain learning. The label i of sample x can be represented as follows:

[0085]

[0086] Where, θ ti,j represents the t-th element of the vector θ when i and j are determined. The predicted label is represented as follows:

[0087]

[0088] Where, and is the number of test samples of unseen faults.· represents FP uT , H u and A uT , respectively, represent I, B and D uT in FP u , H uT and A T B.

[0089] Verification experiment:

[0090] The experimental verification is based on a continuous stirred reactor system. During the 20-hour data collection process, process data under different operating conditions were generated by operating T, including 4 working conditions (C1: T = 430.88K, C2: T = 440.88K, C3: T = 400.88K and C4: T = 460.88K), forming 4 different domains. In each case, 7 process variables with Gaussian noise were collected. In addition, the input C i , T i and TCi is dynamically changing, randomly changing its nominal value every 60 minutes. The sampling interval is set to 1 minute, and a fault is introduced after 200 minutes. Table 1 lists the 10 initial process fault modes.

[0091] Table 1 Early fault categories for the continuous stirred reactor system

[0092]

[0093] Table 2 Two different diagnostic tasks for the continuous stirred reactor process

[0094]

[0095]

[0096] Table 3 Comparison results in the continuous stirred reactor dataset (%)

[0097]

[0098] The training samples only include 6 visible faults from 3 labeled source domains and 1 unlabeled target domain, with 350 samples for each fault. The total number of training samples is 8400 (350x6x4). The test samples include all faults in the target domain, with 350 instances for each visible fault and 525 instances for each unseen fault. The total number of test samples is 4200 (350x6+525x4).

[0099] The evaluation metrics are the average accuracy Acc s and Acc u , the harmonic mean H1, and the overall accuracy H2:

[0100]

[0101] It can be seen from the experimental results in Tables 2 and 3 that the unsupervised generalized zero-shot fault diagnosis method proposed in the present application further improves the ability of generalized zero-shot fault diagnosis. Overall, this method can achieve satisfactory fault diagnosis performance, which is highly consistent with the goal of the generalized zero-shot learning task.

[0102] Finally, it should be pointed out that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit it. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the purpose and scope of the present application, which should be covered by the scope of the claims of the present application.

Claims

1. A cross-source domain industrial fault diagnosis method based on generalized zero-shot learning, characterized in that, The method specifically includes the following steps: S1: Joint learning of multi-classifier and cross-domain alignment consistency; An independent classifier is constructed for each source domain, and the distribution consistency between the source and target domains is achieved by aligning the features of the source domain with the features of the unlabeled samples in the target domain. The domain alignment strategy adopts the feature maximum mean difference method to ensure that the feature distributions of the source and target domains, as well as the inter-domain distributions among multiple domains, are unified in the same feature space. S2: Fault Prototype and Semantic Embedding Learning; A latent space with orthogonal constraints is constructed for mapping learning between the feature space and the semantic attribute space. For visible faults from different source domains, corresponding sample labels are acquired during the training phase for learning. Domain alignment techniques are used to average all samples belonging to a specific category to obtain a prototype representation of that category of faults. In the context of generalized zero-sample fault diagnosis, a semantic similarity-based strategy is adopted to determine the semantic attribute descriptions of visible faults similar to unseen faults by calculating Euclidean distance. By learning mappings containing information about unseen faults, connections between the feature space and the latent space, as well as between the attribute space and the latent space, are realized respectively. S3: Comprehensive fault diagnosis; Based on visible faults, a model capable of classification across multiple source domains is learned, and a classifier with strong generalization ability is trained for the unlabeled target domain through domain alignment. In the feature space, latent space, and fault attribute space, test samples containing visible faults and unseen faults are diagnosed respectively. At the same time, by introducing visible information in the multi-source domain learning process, a comprehensive diagnostic strategy is adopted to obtain the diagnostic results. The comprehensive diagnostic strategy specifically includes: Let the weighted distance set be... ,in, It is the first in the feature space i The characteristics of the sample and the first j Weighted distance between fault prototypes of the same type of fault. It is the first in the latent space i The characteristics of the sample and the first j Weighted distance between fault prototypes of the same type of fault. It is the first in the attribute space i The characteristics of the sample and the first j Weighted distance between fault prototypes of the same type of fault. It is the number of test samples. q This refers to the number of unseen categories; the comprehensive tag set. It is represented by the corresponding label value in the minimum distance space; at the same time, it must be ensured that this label value is consistent with the classification result in multi-source domain learning. Maintain consistency; samples tags It is expressed as follows: in, , Indicates when i and j Determine the time vector The t One element; It is the first in the feature space i Predicted label for each sample, It is the first in the latent space i Predicted label for each sample, It is the first in the attribute space i Predicted labels for each sample; predicted labels It is expressed as follows: in, ,and , It is the first in the feature space i The characteristics of the sample and the first j Normalized weighted distance between fault prototypes of the same fault class It is the first in the latent space i The characteristics of the sample and the first j Normalized weighted distance between fault prototypes of the same fault class It is the first in the attribute space i The characteristics of the sample and the first j Normalized weighted distance between fault prototypes of the same type of fault; This is the number of test samples for unseen fault types; express , and , * respectively represent , and In I , B and , , and These represent the prototype of the unseen fault class, the latent space of the unseen fault class, and the attributes of the unseen fault class, respectively. I Represents the identity matrix. B D represents the mapping between the feature space and the latent space; Indicates the first k Features of each test sample.

2. The cross-source domain industrial fault diagnosis method according to claim 1, characterized in that, In step S1, multi-classifier learning specifically includes: for the first... k A source domain, which contains One fault sample and c Fault categories, classifier By calculating classification loss J Continuous optimization; the formula for the loss function is as follows: in, It is the first i Sample The tag, S It is a visible fault category. ; yes The first one-hot encoded vector j One value; if ,but ;otherwise ; It is a connection feature and fault classifier The Middle m The weights of each output neuron are determined; subsequently, the multi-class learning loss is calculated. Represented as: in, K It is the number of source domains. It is the classifier The loss function.

3. The cross-source domain industrial fault diagnosis method according to claim 2, characterized in that, In step S2, the fault prototype and semantic embedding learning specifically includes: i A visible class of fault prototypes Obtained through the following methods: in, It is the first among all source domains i The number of samples for each visible fault class. It is the visible class sample space; For the j No previously observed type of fault, l Similar visible attribute descriptions Represented as: ， in, Dis It is a Euclidean distance function. It is the first j Unseen class semantic attribute vectors It is a visible class attribute matrix. express Dis middle t Sort in ascending order; No. j Unseen Fault Prototype Represented as: in, It is the first i A fault prototype with similar visible attributes.

4. The cross-source domain industrial fault diagnosis method according to claim 3, characterized in that, In step S3, the predicted label of the test sample The expression is: in, It is the first i Features of each test sample It is the number of test samples; Representation matrix The j The i-th column vector, i.e., the nth column vector j Class of faults FP ; It is the first in the feature space i The characteristics of the sample and the first j Weighted (normalized) distance between fault prototypes of the same fault class; Y It is a fault-related tag.

5. The cross-source domain industrial fault diagnosis method according to claim 3, characterized in that, In step S3, in the latent space, the predicted label is obtained through the learned mapping. B get: in, It is a predicted label in the latent space. It is the first i Features of each test sample It is the number of test samples; It is the first j The potential space for each fault category; Y It is a fault-related tag.

6. The cross-source domain industrial fault diagnosis method according to claim 3, characterized in that, In step S3, in the fault attribute space, through mapping B and D The following predicted labels were obtained: in, It is a predicted label in the fault attribute space. It is the first i Features of each test sample It is the number of test samples; It is the first i Attribute matrix of each fault class; Y It is a fault-related tag.

Images