A generalized zero-shot bearing composite fault diagnosis method based on multi-label learning

Through multi-label learning and conditional diffusion model, the single fault data is used to generate composite fault characteristics, which solves the problem that samples are difficult to obtain in composite fault diagnosis, and realizes efficient diagnosis under composite fault training samples, which is suitable for industrial scenarios.

CN118427672BActive Publication Date: 2024-11-08GUANGDONG POLYTECHNIC NORMAL UNIV
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202410892158.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-04
Publication Date
2024-11-08
Estimated Expiration
2044-07-04

AI Technical Summary

Technical Problem

In the prior art, in bearing fault diagnosis, the unpredictability of composite faults and single fault samples are difficult to deduce composite fault characteristics, resulting in difficulty in diagnosis of composite faults, and requires a large amount of label data and high computational complexity.

Method used

Using a generalized zero-sample method based on multi-label learning, a pre-trained composite fault diagnosis model is used to define vectors and Gaussian noise to generate training sets, and combining conditional diffusion models to generate composite fault data to achieve no composite fault training samples. diagnosis.

Benefits of technology

It reduces the difficulty of collecting composite fault samples, improves diagnostic efficiency and accuracy, reduces model complexity and calculation costs, and is suitable for the identification of composite faults in industrial scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118427672B_ABST
    Figure CN118427672B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of fault diagnosis, and in particular to a generalized zero-sample bearing composite fault diagnosis method based on multi-label learning, the characteristic of which is that it solves the problem that composite fault data is difficult to obtain nowadays. The method not only takes into account the unpredictability of the occurrence of single and composite faults of bearings due to the working environment and self-wear of the bearings in real industrial scenarios, but also alleviates the demand of artificial intelligence algorithms for composite fault samples, and only uses single fault samples to train the model, reducing the difficulty of collecting composite fault samples, and solving the problem that it is difficult to collect composite fault samples, resulting in the inability to use sufficient samples for the model to learn the characteristics of composite fault samples. This method can be embedded and integrated with existing industrial systems, enhance the intelligence of the system, effectively reduce the cost of fault diagnosis, and greatly improve the efficiency of fault diagnosis.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of fault diagnosis, and in particular to a generalized zero-sample bearing composite fault diagnosis method based on multi-label learning. Background Art

[0002] In the field of fault diagnosis, the unpredictability of compound faults and the fact that a single fault can mask the occurrence of compound faults make it extremely difficult to diagnose compound faults in industrial scenarios. At present, with the development of deep learning, a large number of fault diagnosis methods based on deep learning are used to automatically identify bearing faults. The network layer of the deep learning model can hierarchically learn features from the data for bearing fault diagnosis. These methods are usually based on supervised learning or semi-supervised learning, and require a large amount of labeled or unlabeled training data to train deep neural networks.

[0003] At present, diagnostic methods based on decoupling models and separated signals have become popular, taking advantage of the fact that single fault features and compound fault features are highly coupled, and single fault samples are relatively easy to obtain. However, these decoupling models require the construction of multiple classifiers and a large amount of labeled data to train the classifiers, which does not conform to the actual situation with few fault samples and greatly increases the complexity and computational cost of the model.

[0004] The method of using the information obtained on the known classes to identify and classify the unseen classes, constructing the semantic vector of the unseen classes through auxiliary prior knowledge, and training a learning model with the known class data set and the constructed unseen class semantics. Converting the recognition of unseen classes into a supervised learning problem that matches the semantic labels greatly reduces the demand for sample size and reduces the complexity of the model. The emergence of this idea can well solve the problem of no compound fault samples. It is a very creative idea to use easily available single fault samples to derive the characteristics of compound faults and convert compound fault diagnosis into a supervised learning problem. Summary of the invention

[0005] The purpose of the present invention is to overcome the above-mentioned deficiencies in the prior art and to provide a generalized zero-sample bearing composite fault diagnosis method based on multi-label learning.

[0006] In order to achieve the above-mentioned object of the invention, the present invention provides the following technical solutions:

[0007] A generalized zero-shot bearing composite fault diagnosis method based on multi-label learning includes the following steps:

[0008] Obtaining the fault signal to be analyzed;

[0009] Inputting the fault signal to be analyzed into a pre-trained composite fault diagnosis model;

[0010] The composite fault diagnosis model outputs a fault diagnosis result corresponding to the fault signal to be analyzed;

[0011] Wherein, the composite fault diagnosis model includes a semantic label definition module, a fault generation module and a multi-label classification module;

[0012] The semantic label definition module is used to store a single fault semantic label definition vector set and a compound fault semantic label definition vector set;

[0013] The fault generation module generates a single fault data set and a compound fault data set according to the single fault semantic label definition vector set and the compound fault semantic label definition vector set in combination with Gaussian noise, and uses them together as training sets for training a multi-label classification module;

[0014] The multi-label classification module is used to test single fault signal and compound fault signal samples that have no intersection with the training set based on the model parameters obtained through training, and outputs the single fault type / compound fault type with the highest similarity as the fault type corresponding to the fault signal to be analyzed.

[0015] As a preferred solution of the present invention, the pre-training of the composite fault diagnosis model includes the following steps:

[0016] S1: Construct sample dataset;

[0017] The sample data set includes a training set and a test set; the training set includes labeled single fault signal samples and compound fault signal samples of the fault generation module, and the test set includes untrained single fault signal samples and unlabeled compound fault signal samples;

[0018] S2: construct a single fault semantic label definition vector set and a compound fault semantic label definition vector set, and store them in the semantic label definition module;

[0019] S3: training the semantic label generation module through the training set and the single fault semantic label definition vector set, and after the training is completed, generating a single fault semantic data set and a composite fault semantic data set according to the single fault semantic label definition vector set and the composite fault semantic label definition vector set in combination with Gaussian noise;

[0020] S4: The multi-label classification module is trained according to the training set to obtain a trained model, which has the ability to identify compound faults. The classifier is then tested and verified according to the test set, and the classification accuracy of the single fault signal set and the compound fault signal set is output.

[0021] As a preferred embodiment of the present invention, the training set D s and test set Dt The following conditions are met:

[0022]

[0023] in, I (·) is used to calculate the two sample distributions p ( X s )and p ( X c )’s mutual information; X s and X c represent the single fault training set and the compound fault set respectively; represents a single-fault test set; L s and L c They represent single fault category label set and compound fault category label set respectively; Z s and Z c They represent single fault semantic label vector set and compound fault semantic label vector set respectively; Represents the mapping function between single fault semantics and compound fault semantics.

[0024] As a preferred embodiment of the present invention, S2 comprises the following steps:

[0025] S21: Constructing a single fault semantic label definition vector set through single fault signal samples , and then according to the single fault type contained in the composite fault signal sample, the semantic label definition vector set corresponding to the composite fault signal sample is obtained ;

[0026] S22: The single fault data corresponding to the kth single fault category composed of N consecutive data points in the single fault signal sample The signal input semantic label definition module obtains matrices of different levels from N data points through discrete wavelet transform , and the semantic label definition vector of each single fault category ;

[0027] S23: storing the semantic label definition vectors of each single fault category of different types as a single fault semantic label definition vector set ;

[0028] S24: Combine the correlation between single faults and compound faults to generate a semantic label definition for compound faults, which is expressed as:

[0029]

[0030] Among them, i, j are the single fault category numbers, and are the first and second faults in the i-th and j-th categories respectively. The dimension value corresponding to the data point;

[0031] S25: Output composite fault semantic label definition vector set , Defines the single fault category numbers that constitute the vector of composite fault semantic labels.

[0032] As a preferred embodiment of the present invention, S3 comprises the following steps:

[0033] The fault generation module is composed of a conditional diffusion model, which includes forward diffusion and reverse diffusion. Gaussian noise is superimposed on the fault image through forward diffusion, and then the noise addition operation of forward diffusion is offset through reverse diffusion. After the model training is completed, a noise is randomly sampled from the standard normal distribution, and then the inverse process is used to restore the noise to an image.

[0034] As a preferred solution of the present invention, the forward diffusion process of the fault generation module is:

[0035]

[0036] in, x t It is t The diffusion data of the step, x t-1 is the data from the previous step, ɛ t At every step t Noise sampled independently and identically from a standard normal distribution, α t is to control the amount of noise added in each step.

[0037] As a preferred solution of the present invention, the reverse diffusion process of the fault generation module predicts the noise loss function as follows: L 1:

[0038]

[0039] in, x t yes ɛ The result after encoding and adding noise to the fault image is obtained. t It's time. ɛ is the actual noise, ɛ θ is the prediction noise, Tθ ( y ) is to encode the fault label into the corresponding intermediate representation, It's expectation.

[0040] As a preferred solution of the present invention, the loss function of the multi-label classification module is L 2 is:

[0041]

[0042] in, C is the total number of fault categories, y o,c is a binary indicator (0 or 1) indicating the o Does the sample belong to c categories, p o,c The model predicts o The samples belong to c The probability of a class.

[0043] A generalized zero-sample bearing composite fault diagnosis device based on multi-label learning, comprising at least one processor and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute any of the methods described above.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] The present invention proposes a new method for diagnosing bearing composite faults based on multi-label learning without composite fault training samples. According to the characteristics of the vibration signal and the theoretical relationship between single faults and composite faults, the fault signal is subjected to discrete wavelet transform to obtain wavelet coefficients, and the wavelet coefficients are used to construct the composite fault data of the unseen class, which can not only well cover the time domain and frequency domain characteristics of each single fault class, but also improve the accuracy of the establishment of the image of the unseen class composite fault. The fault generation module is composed of a conditional diffusion model. Through the composite fault data of the semantic label definition module, the conditional diffusion model can generate more information about the composite fault image, increase the accuracy of the multi-label classification module for composite fault classification, and realize the diagnosis of composite faults in the absence of composite fault training samples, that is, without obtaining composite fault training samples. The characteristic of this method is to solve the problem that composite fault data is difficult to obtain nowadays. This method not only takes into account the unpredictability of the occurrence of single and compound faults of bearings due to the working environment and wear of the bearings in real industrial scenarios, but also alleviates the demand of artificial intelligence algorithms for compound fault samples. It only uses single fault samples to train the model, reduces the difficulty of collecting compound fault samples, and solves the problem of not being able to use sufficient samples for the model to learn the characteristics of compound fault samples due to the difficulty in collecting compound fault samples. This method can be embedded and integrated with existing industrial systems to enhance the intelligence of the system, effectively reduce the cost of fault diagnosis, and greatly improve the efficiency of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a flow chart of a generalized zero-sample bearing composite fault diagnosis method based on multi-label learning according to Embodiment 1 of the present invention;

[0047] Figure 2 It is a structural diagram of a composite fault diagnosis model in a generalized zero-sample bearing composite fault diagnosis method based on multi-label learning described in Example 2 of the present invention;

[0048] Figure 3 A training flow chart of a composite fault diagnosis model in a generalized zero-shot bearing composite fault diagnosis method based on multi-label learning as described in Example 2 of the present invention;

[0049] Figure 4 This is a structural schematic diagram of a bearing composite fault diagnosis device described in Example 3 of the present invention, which utilizes the generalized zero-sample bearing composite fault diagnosis method based on multi-label learning described in Example 1. DETAILED DESCRIPTION

[0050] The present invention is further described in detail below in conjunction with test examples and specific implementation methods. However, this should not be understood as the scope of the above subject matter of the present invention being limited to the following embodiments, and all technologies realized based on the content of the present invention belong to the scope of the present invention.

[0051] Example 1

[0052] like Figure 1 As shown, a generalized zero-shot bearing composite fault diagnosis method based on multi-label learning includes the following steps:

[0053] Obtaining the fault signal to be analyzed;

[0054] Inputting the fault signal to be analyzed into a pre-trained composite fault diagnosis model;

[0055] The composite fault diagnosis model outputs a fault diagnosis result corresponding to the fault signal to be analyzed;

[0056] The composite fault diagnosis model includes a semantic label definition module, a fault generation module and a multi-label classification module.

[0057] The semantic label definition module is used to store a single fault semantic label definition vector set and a compound fault semantic label definition vector set;

[0058] The fault generation module generates a single fault data set and a compound fault data set according to the single fault semantic label definition vector set and the compound fault semantic label definition vector set in combination with Gaussian noise, and uses them together as training sets for training a multi-label classification module;

[0059] The multi-label classification module is used to test single fault signal and compound fault signal samples that have no intersection with the training set based on the model parameters obtained through training, and outputs the single fault type / compound fault type with the highest similarity as the fault type corresponding to the fault signal to be analyzed.

[0060] Example 2

[0061] This embodiment is a pre-training method for the composite fault diagnosis model in the generalized zero-sample bearing composite fault diagnosis method based on multi-label learning described in Embodiment 1. The structure of the composite fault diagnosis model is as follows: Figure 2 As shown, the pre-training Figure 3 As shown, the following steps are included:

[0062] S1: Construct sample dataset;

[0063] The sample data set includes a training set D s and test set D tThe training set includes labeled single fault signal samples, and the test set includes labeled single fault signal samples and labeled compound fault signal samples.

[0064] The present invention includes three single fault types and one composite fault type, including inner ring fault (IF), outer ring fault (OF), ball fault (BF) and composite fault of three single faults (IF&OF&BF). Embodiment 2 of the present invention includes three single fault types and three composite fault types, including inner ring fault (IF), outer ring fault (OF), ball fault (BF), inner and outer ring fault (IOF), inner ring ball fault (IBF) and outer ring ball fault (OBF).

[0065] At the same time, the training set D s and test set D t The following conditions are met:

[0066]

[0067] in, I (·) is used to calculate the two sample distributions p ( X s )and p ( X c )’s mutual information; X s and X c represent the single fault training set and the compound fault set respectively; represents a single-fault test set; L s and L c They represent single fault category label set and compound fault category label set respectively; Z s and Z c They represent single fault semantic label vector set and compound fault semantic label vector set respectively; Represents the mapping function between single fault semantics and compound fault semantics.

[0068] S2: construct a single fault semantic label definition vector set and a compound fault semantic label definition vector set, and store them in the semantic label definition module.

[0069] S21: Constructing a single fault semantic label definition vector set through single fault signal samples , and then according to the single fault type contained in the composite fault signal sample, the semantic label definition vector set corresponding to the composite fault signal sample is obtained .

[0070] S22: The single fault data corresponding to the kth single fault category composed of N consecutive data points in the single fault signal sample The signal input semantic label definition module obtains matrices of different levels from N data points through discrete wavelet transform , and the semantic label definition vector of each single fault category .

[0071] S23: storing the semantic label definition vectors of each single fault category of different types as a single fault semantic label definition vector set .

[0072] S24: Combine the correlation between single faults and compound faults to generate a semantic label definition for compound faults, which is expressed as:

[0073]

[0074] Among them, i, j are the single fault category numbers, and are the first and second faults in the i-th and j-th categories respectively. The dimension value corresponding to each data point.

[0075] S25: Output composite fault semantic label definition vector set , Defines the single fault category numbers that constitute the vector of composite fault semantic labels.

[0076] S3: The fault generation module is composed of a conditional diffusion model, which includes forward diffusion and reverse diffusion. Gaussian noise is superimposed on the fault image through forward diffusion, and then the noise addition operation of forward diffusion is offset through reverse diffusion. After the model training is completed, a noise is randomly sampled from the standard normal distribution, and then the noise is restored to an image using the reverse process;

[0077] Among them, the forward diffusion process of the fault generation module is:

[0078] ,

[0079] in, x t It is t The diffusion data of the step, x t-1 is the data from the previous step, ɛ t At every step t Noise sampled independently and identically from a standard normal distribution, α tis to control the amount of noise added in each step.

[0080] The reverse diffusion process of the fault generation module predicts the noise loss function as follows: L 1:

[0081]

[0082] in, x t yes ɛ The result after encoding and adding noise to the fault image is obtained. t It's time. ɛ is the actual noise, ɛ θ is the prediction noise, T θ ( y ) is to encode the fault label into the corresponding intermediate representation, It's expectation.

[0083] Use the trained fault generation module to generate faults in the training set D s Perform feature expansion to enhance the multi-label classification model's ability to identify complex faults.

[0084] S4: Loss function of the multi-label classification module L 2 is:

[0085]

[0086] in, C is the total number of fault categories, y o,c is a binary indicator (0 or 1) indicating the o Does the sample belong to c categories, p o,c The model predicts o The samples belong to c If the loss meets the requirements, the multi-label classification module training is completed.

[0087] The present invention considers the occurrence of fault conditions in real scenarios, and only uses the training set, i.e., single fault samples, to train the diagnostic model, and uses the test set, i.e., single fault and compound fault samples, to test the model together, to verify the diagnostic ability of the model in the absence of compound fault training samples. The specific experimental examples are as follows:

[0088] The bearing faults of test case A include ball fault (BF), inner race fault (IF), outer race fault (OF) and a combination of the above three faults (IF&OF&BF). During the experiment, the motor speed was set to 1800r / min and the sampling frequency was set to 20480Hz.

[0089] The bearing faults of test case B include inner race fault (IF), outer race fault (OF), ball fault (BF), inner and outer race fault (IOF), inner race ball fault (IBF) and outer race ball fault (OBF). During the experiment, the motor speed was set to 1200r / min and the sampling frequency was set to 50kHz.

[0090] Through experiments on the tasks of the two embodiments, the specific division of the experimental data set is shown in Table 1. When the number of training fault samples remains unchanged, the classification results presented by different test fault samples are shown in Table 2. The average accuracy rates of the seen single faults of the two embodiments are 99.65% and 99.81% respectively, and the accuracy rates of the unseen class, i.e., compound faults, are 89.16% and 88.30% respectively.

[0091] Table 1

[0092]

[0093] Table 2

[0094] In order to highlight the effectiveness and superiority of the method described in this application, the following two comparison methods are constructed:

[0095] 1) Comparison method 1: using an embedded zero-shot model; 2) Comparison method 2: using a generative zero-shot model. According to the H-score of the generalized zero-shot, the experimental results are shown in Table 3. The H-scores of the comparison method 1 in the embodiment are 65.45 and 75.14 respectively, and the H-scores of the comparison method 2 in the embodiment are 69.24 and 78.56 respectively. Legal The H-score results are as high as 93.8 and 94.62, which reflects the effectiveness of the combination of multi-label learning and conditional diffusion model, especially the generalized zero-shot learning method that combines the conditional diffusion model of the present application with the multi-label learning method, which can use labeled training samples to help the model learn the fault characteristics of unseen compound faults, thereby improving the final fault diagnosis results.

[0096] Table 3

[0097]

[0098] The present invention proposes a learning model for composite fault diagnosis without composite fault training samples, which solves the problem that composite fault samples are difficult to obtain and cannot be used in industrial equipment operation scenarios. The experimental results show that the model can use the designed semantic label definition module and the existing single fault data to construct composite fault data with multi-feature information through the fault generation module, and classify faults according to the classification model based on multi-label learning, solving the composite fault diagnosis problem in the actual situation without composite fault samples, with significant results.

[0099] Example 3

[0100] like Figure 4 As shown, a generalized zero-sample bearing composite fault diagnosis device based on multi-label learning includes at least one processor, a memory connected to the at least one processor, and at least one input-output interface connected to the at least one processor; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the generalized zero-sample bearing composite fault diagnosis method based on multi-label learning described in the above embodiment. The input-output interface may include a display, a keyboard, a mouse, and a USB interface for inputting and outputting data.

[0101] Those skilled in the art can understand that: all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiments; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM), disks or optical disks, etc. Various media that can store program codes.

[0102] When the above-mentioned integrated unit of the present invention is implemented in the form of a software functional unit and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0103] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A generalized zero-shot bearing composite fault diagnosis method based on multi-label learning, characterized in that: The following steps are involved: Obtaining the fault signal to be analyzed; Inputting the fault signal to be analyzed into a pre-trained composite fault diagnosis model; The composite fault diagnosis model outputs a fault diagnosis result corresponding to the fault signal to be analyzed; Wherein, the composite fault diagnosis model includes a semantic label definition module, a fault generation module and a multi-label classification module; The semantic label definition module is used to store a single fault semantic label definition vector set and a compound fault semantic label definition vector set; The fault generation module generates a single fault data set and a compound fault data set according to the single fault semantic label definition vector set and the compound fault semantic label definition vector set in combination with Gaussian noise, and uses them together as training sets for training a multi-label classification module; The multi-label classification module is used to test single fault signal and compound fault signal samples that have no intersection with the training set according to the model parameters obtained through training, and outputs the single fault type / compound fault type with the highest similarity as the fault type corresponding to the fault signal to be analyzed; The pre-training of the composite fault diagnosis model includes the following steps: S1: Construct sample dataset; The sample data set includes a training set and a test set; the training set includes labeled single fault signal samples and compound fault signal samples of the fault generation module, and the test set includes untrained single fault signal samples and unlabeled compound fault signal samples; S2: construct a single fault semantic label definition vector set and a compound fault semantic label definition vector set, and store them in the semantic label definition module; S3: training the semantic label generation module through the training set and the single fault semantic label definition vector set, and after the training is completed, generating a single fault semantic data set and a composite fault semantic data set according to the single fault semantic label definition vector set and the composite fault semantic label definition vector set in combination with Gaussian noise; S4: The multi-label classification module is trained according to the training set to obtain a trained model, which has the ability to identify compound faults. The classifier is then tested and verified according to the test set, and the classification accuracy of the single fault signal set and the compound fault signal set is output; The S2 comprises the following steps: S21: Constructing a single fault semantic label definition vector set through single fault signal samples , and then according to the single fault type contained in the composite fault signal sample, the semantic label definition vector set corresponding to the composite fault signal sample is obtained ; S22: The single fault data corresponding to the kth single fault category composed of N consecutive data points in the single fault signal sample The signal input semantic label definition module obtains matrices of different levels from N data points through discrete wavelet transform , and the semantic label definition vector of each single fault category ; S23: storing the semantic label definition vectors of each single fault category of different types as a single fault semantic label definition vector set ; S24: Combine the correlation between single faults and compound faults to generate a semantic label definition for compound faults, which is expressed as: , Among them, i, j are the single fault category numbers, and are the dimension values ​​corresponding to the nth data point in the i-th and j-th single faults respectively; S25: Output composite fault semantic label definition vector set , The single fault category serial number that constitutes the composite fault semantic label definition vector; The S3 comprises the following steps: The fault generation module is composed of a conditional diffusion model, which includes forward diffusion and reverse diffusion. Gaussian noise is superimposed on the fault image through forward diffusion, and then the noise addition operation of forward diffusion is offset through reverse diffusion. After the model training is completed, a noise is randomly sampled from the standard normal distribution, and then the noise is restored to an image using the reverse process; The forward diffusion process of the fault generation module: , in, x t It is t The diffusion data of the step, x t-1 is the data from the previous step, ɛ t At every step t Noise sampled independently and identically from a standard normal distribution, α t is to control the amount of noise added in each step; The reverse diffusion process of the fault generation module predicts the noise loss function as follows: L 1: , in, x t yes ɛ The result after encoding and adding noise to the fault image is obtained. t It's time. ɛ is the actual noise, ɛ θ is the prediction noise, T θ ( y ) is to encode the fault label into the corresponding intermediate representation, It's expectation.

2. The generalized zero-sample bearing composite fault diagnosis method based on multi-label learning according to claim 1 is characterized in that: Training set D s and test set D t The following conditions are met: in, I (·) is used to calculate the two sample distributions p ( X s )and p ( X c )’s mutual information; X s and X c represent the single fault training set and the compound fault set respectively; represents a single-fault test set; L s and L c They represent single fault category label set and compound fault category label set respectively; Z s and Z c They represent single fault semantic label vector set and compound fault semantic label vector set respectively; Represents the mapping function between single fault semantics and compound fault semantics.

3. The generalized zero-sample bearing composite fault diagnosis method based on multi-label learning according to claim 1 is characterized in that: The loss function of the multi-label classification module L 2 is: , in, C is the total number of fault categories, y o,c is a binary indicator indicating the o Does the sample belong to c categories, p o,c The model predicts o The samples belong to c The probability of a class.

4. A generalized zero-shot bearing composite fault diagnosis method based on multi-label learning, characterized in that: It includes at least one processor and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method described in any one of claims 1 to 3.

Citation Information

Patent Citations

  • Discriminant generalized zero sample learning fault diagnosis method

    CN113609569A

  • Bearing composite fault diagnosis method based on embedded zero sample learning model

    CN114383845A

  • Bearing composite fault diagnosis method based on fault label information vector

    CN114383846A

  • Dynamic vision enabling non-contact equipment migration fault diagnosis method

    CN117649554A

  • Zero sample rolling bearing compound fault diagnosis system and method based on envelope spectrum

    CN117949209A