Methods, systems, and media for open-set domain generalized bearing fault diagnosis under unknown operating conditions

Abstract

This invention discloses a method, system, and medium for open-set domain generalized bearing fault diagnosis under unknown operating conditions, belonging to the field of bearing fault diagnosis technology. The method includes: utilizing source domain bearing vibration data; pre-training a fault diagnosis model based on subdomain constraints and domain adversarial mechanisms; constructing class centers based on the average activation vectors of known fault categories in the source domain; acquiring target domain bearing vibration signals in real time and inputting them into the pre-trained fault diagnosis model to obtain test set sample scores; classifying the test set samples into known faults and unknown faults based on the distance and distance threshold of each sample score to the class centers; differentially adjusting the test set sample scores corresponding to known and unknown faults; and determining the probabilities of known and unknown faults based on the adjusted scores. This invention can effectively improve the accuracy of open-set fault diagnosis for rotating machinery under unknown operating conditions.

Description

Technical Field

[0001] This invention relates to the field of bearing fault diagnosis technology, specifically to a bearing fault diagnosis method, system, and medium based on open set domain generalization under unknown operating conditions, and particularly to an openmax open set fault diagnosis method based on domain adversarial network generalization of class conditional subdomain and adaptive distance discrimination under unknown operating conditions. Background Technology

[0002] Currently, fault diagnosis has become an indispensable part of complex mechanical equipment. Fault diagnosis methods can effectively prevent accidents and avoid economic losses.

[0003] However, in actual mechanical equipment fault diagnosis, target domain data cannot be obtained in advance, and due to factors such as cost and time, it is impossible to collect a complete fault dataset. At the same time, the working conditions are usually unknown, making it impossible to train an accurate fault prediction model.

[0004] In addition, problems such as undetected faults and distribution differences exist in practical applications. These factors together pose a huge challenge to fault diagnosis and have a significant impact on the stable operation and maintenance of mechanical equipment.

[0005] Therefore, how to solve the above problems is currently the main research direction for those skilled in the art. Summary of the Invention

[0006] In view of the above problems, this application proposes an open-set domain generalization bearing fault diagnosis method, system, and medium based on domain generalization under unknown operating conditions. Open-set domain generalization does not require target domain data during the training phase; it only uses source domain data and known fault data to train the model. During the testing phase, it can effectively diagnose known faults and identify unknown faults under unknown operating conditions.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] In a first aspect, embodiments of the present invention provide an open-set domain generalized bearing fault diagnosis method under unknown operating conditions, comprising:

[0009] Using source domain bearing vibration data, a fault diagnosis model is pre-trained based on subdomain constraints and domain adversarial mechanisms; and class centers are constructed based on the average activation vectors of known fault categories in the source domain.

[0010] Real-time acquisition of bearing vibration signals in the target domain is input into a pre-trained fault diagnosis model to obtain test set sample scores.

[0011] Based on the distance from the score of each sample in the test set to the class center and the distance threshold, the test set samples are divided into known faults and unknown faults;

[0012] The test set sample scores corresponding to known faults and unknown faults are differentially adjusted, and the probabilities of known faults and unknown faults are determined based on the adjusted scores.

[0013] Preferably, the fault diagnosis model is pre-trained using source domain bearing vibration data and based on subdomain constraints for domain adversarial analysis. This aims to make the relationships between the same faults under different operating conditions more compact, including:

[0014] The fault diagnosis model includes a feature extractor and a classifier, and is pre-trained by minimizing the classification loss, combined with subdomain constraints and domain adversarial mechanisms. The loss function expression is as follows:

[0015]

[0016] In the formula, Indicates feature extractor, Represents a classifier. Represents classification loss, and This represents the hyperparameters of a balanced classification task. This represents the source domain discrimination loss after forward pass through GRL, where GRL stands for gradient reversal layer. Discriminator for the domain, Indicates the subdomain discrimination loss. This represents a subdomain discriminator.

[0017] Preferably, the domain discriminator and the subdomain discriminator Optimization is achieved by maximizing the discriminative loss, as shown in the following expression:

[0018]

[0019] In the formula, Indicates the discriminator, This indicates the source domain discrimination loss.

[0020] Preferably, the distance from each sample score to the class center is calculated according to the following formula:

[0021]

[0022] In the formula, Represents the weight parameters. Indicates sample The score, Indicates the fault category The class center, This represents the fault category index.

[0023] Preferably, the distance threshold is obtained through the following steps:

[0024] Determine the mean and standard deviation based on the distance distribution from known fault samples within the source domain to their respective class centers;

[0025] Determine the distance threshold based on the mean and standard deviation:

[0026]

[0027] In the formula, and Let represent the mean and standard deviation of the distance distribution, respectively. Indicates scaling factor

[0028] The threshold of this application can be adaptively adjusted according to different datasets. That is, when the method needs to be applied to different devices or scenarios, there is no need for tedious manual adjustment of the threshold.

[0029] Preferably, the test set samples are divided into known faults and unknown faults, including:

[0030] When the minimum distance is greater than the distance threshold, the corresponding test set sample is determined to be without fault; otherwise, it is determined to be a known fault.

[0031] Preferably, the test set sample scores corresponding to known faults and unknown faults are differentially adjusted, and the probabilities of known faults and unknown faults are determined based on the adjusted scores, including:

[0032] When a fault is identified as an unknown fault, the scores of the remaining samples identified as known faults are scaled.

[0033]

[0034] in, Indicates the scaling factor. Indicates the known fault category The sample score, This represents the total number of known fault categories. Indicates the index of known fault categories. Indicates the adjusted fault category Sample score;

[0035] The known failure probabilities are:

[0036]

[0037] The probability of an unknown fault is:

[0038]

[0039] In the formula, Indicates the sample failure category number. This represents the scaled score of the k-th sample. This represents a parameter that controls the probability of the unknown class.

[0040] When a known fault is identified, the one with the highest score is selected. Each category is assigned a decreasing weight factor in order of sorting:

[0041]

[0042] The scores of the known fault samples are adjusted according to the weighting factors as follows:

[0043]

[0044] In the formula, Indicates faulty sample Linear decay weights, Indicates the known fault category The sample score, Indicates the index of known fault categories. Indicates the adjusted fault category Sample score;

[0045] The known failure probabilities are:

[0046]

[0047] The probability of an unknown fault is:

[0048]

[0049] In the formula, Indicates the known probability of failure. This represents the total number of known fault categories. Indicates the probability of an unknown failure. This represents the adjusted score for the k-th known category, where k represents the sample fault category number.

[0050] Secondly, embodiments of the present invention provide an open-set domain generalized bearing fault diagnosis system under unknown operating conditions. This system is used to implement the open-set domain generalized bearing fault diagnosis method under unknown operating conditions as described in any of the preceding claims, including:

[0051] The pre-training module is used to pre-train the fault diagnosis model based on subdomain constraints and domain adversarial analysis using source domain bearing vibration data.

[0052] The class center construction module is used to construct class centers based on the average activation vector of known fault categories in the source domain.

[0053] The fault prediction module is used to acquire bearing vibration signals in the target domain in real time, input them into the pre-trained fault diagnosis model, and obtain test set sample scores.

[0054] The fault classification module is used to classify the test set samples into known faults and unknown faults based on the distance from the score of each sample in the test set to the class center and the distance threshold.

[0055] The post-fault processing module is used to differentiate the test set sample scores corresponding to known faults and unknown faults, and to determine the probability of known faults and the probability of unknown faults based on the adjusted scores.

[0056] Thirdly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the open-set domain generalized bearing fault diagnosis method under unknown operating conditions as described in any of the preceding claims.

[0057] This invention provides a method, system, and medium for open-set domain generalized bearing fault diagnosis under unknown operating conditions. Compared with the prior art, this application fully considers the problem of the simultaneous existence of unknown operating conditions and unseen faults in reality, and can effectively improve the accuracy of open-set fault diagnosis of rotating machinery under unknown operating conditions. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0059] Figure 1 This is a schematic diagram of the open set domain generalized bearing fault diagnosis method under unknown working conditions provided in this embodiment of the invention.

[0060] Figure 2 This is a time-frequency result diagram of wavelet transform provided in an embodiment of the present invention;

[0061] Figure 3 This is a diagram showing the fault diagnosis results based on the confusion matrix provided in this embodiment of the invention.

[0062] Figure 4 The images provided in this embodiment of the invention are a domain adversarial network based on class-conditional subdomain constraints and a feature visualization diagram after processing with only a domain adversarial network. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] This invention discloses an open-set domain generalization bearing fault diagnosis method, system, and medium under unknown operating conditions. Open-set domain generalization does not require target domain data during the training phase, but only uses source domain data and known fault data to train the model. During the testing phase, it can effectively diagnose known faults and identify unknown faults under unknown operating conditions.

[0065] The following is a description through specific embodiments.

[0066] Figure 1 This is a schematic flowchart of a generalized bearing fault diagnosis method for open-set domains under unknown operating conditions. The method includes the following steps:

[0067] Using source domain bearing vibration data, a fault diagnosis model is pre-trained based on subdomain constraints and domain adversarial mechanisms; and class centers are constructed based on the average activation vectors of known fault categories in the source domain.

[0068] Real-time acquisition of bearing vibration signals in the target domain is input into a pre-trained fault diagnosis model to obtain test set sample scores.

[0069] Based on the distance from the score of each sample in the test set to the class center and the distance threshold, the test set samples are divided into known faults and unknown faults;

[0070] The test set sample scores corresponding to known faults and unknown faults are differentially adjusted, and the probabilities of known faults and unknown faults are determined based on the adjusted scores.

[0071] This embodiment first utilizes source domain data to perform adversarial training on the fault diagnosis model based on subdomain constraints and domain adversarial mechanisms, in order to achieve generalized fault diagnosis in the open set target domain. The fault diagnosis model includes a feature extractor and a classifier.

[0072] In one optional implementation, after acquiring multiple source domain bearing vibration signals, the time-frequency diagram of the source domain bearing vibration signals is preferentially obtained through wavelet transform. This includes first constructing a sample of length 2048 using a sliding window method, with 120 samples for each fault, and then using wavelet transform on these 120 samples to convert them into a time-frequency diagram, the result of which is as follows. Figure 2 Then it is fed into a feature extractor for feature extraction, and a classifier is used to obtain the source domain sample scores.

[0073] For example, source domain data is represented as Where K represents the number of source domains, This represents the Kth source domain data, which contains a total of Marked samples ,and Here, It is the input sample data. These are the input sample labels.

[0074] In this embodiment, adversarial training is achieved through minimax game. First, the discriminator maximizes the differentiation of the source domain. Then, during adversarial training in that domain, a conditional subdomain constraint is introduced to reduce the confusion between known and unknown faults caused by generalization to the target domain, as shown in the following equation:

[0075]

[0076] In the formula, Indicates the discriminator, Indicates source domain discrimination loss, Indicates feature extractor, The domain discriminant is used to determine whether a sample belongs to the source domain or the target domain. Indicates the subdomain discrimination loss. This represents a subdomain discriminator, used to determine the subdomain to which a sample belongs.

[0077] It should be noted that in this embodiment, the subdomain discriminator does not participate in adversarial training.

[0078] Then, the classification loss of the feature extractor and classifier is minimized, and adversarial training is implemented through a gradient inversion layer (GRL) and a domain discriminator, as shown in the following equation:

[0079]

[0080] In the formula, Indicates feature extractor, Represents a classifier. Represents classification loss, and This represents the hyperparameters of a balanced classification task. Discriminator for the domain, Indicates the subdomain discrimination loss. Subdomain discriminator, This represents the source domain discrimination loss after forward pass through GRL, where GRL stands for gradient inversion layer, and its gradient with respect to the features satisfies:

[0081]

[0082] In the formula, h represents the feature representation output by the feature extractor, and a represents the adversarial weight coefficient.

[0083] In summary, we can obtain an overall optimization objective:

[0084]

[0085] In an optional embodiment, a class center is further constructed based on the average activation vector of known fault categories in the source domain; so that the distance to the class center can be calculated using the pre-trained fault diagnosis model based on the test set sample scores obtained from the real-time acquired bearing vibration signal in the target domain, and the test set samples can be divided into unknown faults and known faults based on this distance and a distance threshold.

[0086] In some specific implementation schemes, the class centers of fault categories are constructed based on the average activation vector of known fault categories in the source domain, where the average activation vector of fault categories is obtained by the following formula:

[0087]

[0088] In the formula, The score for the i-th sample in the source domain. The fault category is The number of samples, Fault Category The average activation vector.

[0089] In this implementation scheme, the adaptive distance threshold is further determined based on this, and the steps include:

[0090] Calculate the path from all known fault samples in the source domain to their respective known classes. The Euclidean distance is given by the following formula:

[0091]

[0092] in These are weight parameters. These are known fault samples within the source domain. The sample represents the output of the classifier. The score is expressed as , The number of known classes.

[0093] Based on the aforementioned distance distribution, the mean and standard deviation are determined; and an adaptive distance threshold is determined based on the mean and standard deviation.

[0094]

[0095] In the formula, and Let represent the mean and standard deviation of the distance distribution, respectively. This represents the scaling factor, preferably 3.3.

[0096] In some specific implementation schemes, after obtaining the bearing vibration signal in the target domain, it is similarly converted into a time-frequency diagram by wavelet transform and then input into the fault diagnosis model to obtain the test set sample score.

[0097] In this stage, the distance from the score of each test set sample to the class center of each known fault is further calculated, thus obtaining the set of distances from all samples in the test set to other faults. ,

[0098]

[0099] For each sample, the minimum distance is selected as the final distance to the class center.

[0100] Furthermore, known and unknown faults are determined based on an adaptive distance threshold. In this embodiment, when the following conditions are met... You can If the initial assessment is that no fault is found, otherwise it will be classified as a known fault.

[0101] In an optional embodiment, classifying known faults and unseen faults allows for differentiated adjustment, or recalibration, of sample scores. In this embodiment, different strategies are employed for adjusting known faults and unseen faults. Specifically,

[0102] When a fault is identified as an unknown fault, the scores of the remaining samples identified as known faults are scaled.

[0103]

[0104] in, This represents the adjustment coefficient, with a value of 1 - unpen, where unpen is an unknown penalty coefficient. Indicates the known fault category The sample score, This represents the total number of known fault categories. Indicates the index of known fault categories. Indicates the adjusted fault category Sample score;

[0105] Then, a probability space is reserved for unseen faults by reducing the score of known faults. The adjusted total probability is less than 1, and the remaining portion is allocated to unseen faults.

[0106] The known probabilities are:

[0107]

[0108] The probability of the unknown class is:

[0109]

[0110] in It is a parameter that controls the probability of the unknown class. This represents the scaled score of the k-th sample, where k represents the sample fault category number.

[0111] Furthermore, when a known fault is identified, the one with the highest score is selected. Each category, for The scores of one category are reweighted based on their scores to reduce their confidence level, while the scores of the remaining categories remain unchanged. The adjusted scores are as follows:

[0112]

[0113] The scores of the known fault samples are adjusted according to the weighting factors as follows:

[0114]

[0115] in, Indicates faulty sample Linear decay weights, Indicates the known fault category The sample score, Indicates the index of known fault categories. Indicates the adjusted fault category Sample score.

[0116] Furthermore, after adjusting the scores, the probabilities need to be normalized again to ensure that their sum is 1, as shown in the following formula:

[0117]

[0118] The probability of an unknown fault is:

[0119]

[0120] In the formula, Indicates the known probability of failure. This represents the total number of known fault categories. Indicates the index of known fault categories. Indicates the fault category The sample score, Indicates the adjusted fault category Sample score, Indicates the probability of an unknown failure. This represents the adjusted score for the k-th known category, where k represents the sample fault category number.

[0121] To verify the effectiveness of the method of this invention, the bearing datasets from Huazhong University of Science and Technology and Case Western Reserve University were used to validate the algorithm. To more intuitively illustrate the superiority of the algorithm, a confusion matrix was used to visualize the diagnostic results. Figure 3 As shown, Figure 3 In the middle, A is the classification confusion matrix based on the Case dataset. Figure 3 B is a classification confusion matrix based on the bearing dataset from Huazhong University of Science and Technology; furthermore, Figure 4 The left image shows a domain adversarial network based on class-conditional subdomain constraints, while the right image shows a feature visualization after processing with only a domain adversarial network. The image demonstrates the effectiveness of the proposed domain adversarial diagnostic model based on class-conditional subdomain constraints. Adding subdomain constraints makes the relationships between the same faults under different operating conditions more compact. Verification confirms the effectiveness of the proposed method, demonstrating that it can not only diagnose known faults but also effectively identify unknown faults.

[0122] Based on the same inventive concept, embodiments of the present invention also provide an open-set domain generalized bearing fault diagnosis system under unknown operating conditions, the system comprising:

[0123] The pre-training module is used to pre-train the fault diagnosis model based on subdomain constraints and domain adversarial analysis using source domain bearing vibration data.

[0124] The fault prediction module is used to acquire bearing vibration signals in the target domain in real time, input them into the pre-trained fault diagnosis model, and obtain test set sample scores.

[0125] The class center construction module is used to construct class centers based on the average activation vector of known fault categories in the source domain.

[0126] The fault classification module is used to classify the test set samples into known faults and unknown faults based on the distance from the score of each sample in the test set to the class center and the distance threshold.

[0127] The post-fault processing module is used to differentiate the test set sample scores corresponding to known faults and unknown faults, and to determine the probability of known faults and the probability of unknown faults based on the adjusted scores.

[0128] Since the principle of the problem solved by the above module is consistent with the steps in the aforementioned open set domain generalized bearing fault diagnosis method under unknown working conditions, the implementation of this system can refer to the implementation of the aforementioned method, and the repeated parts will not be repeated.

[0129] Furthermore, the open-set domain generalized bearing fault diagnosis method under unknown operating conditions provided in this embodiment of the invention can be run in the form of a computer-readable storage medium. Specifically, the storage medium stores a computer program, which, when executed by a processor, implements the open-set domain generalized bearing fault diagnosis method under unknown operating conditions as described in any of the preceding claims.

[0130] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0131] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for generalizing bearing fault diagnosis in an open set domain under unknown operating conditions, characterized in that, include: Using source domain bearing vibration data, a fault diagnosis model is pre-trained based on subdomain constraints and domain adversarial mechanisms; And class centers are constructed based on the average activation vector of known fault categories in the source domain; Real-time acquisition of bearing vibration signals in the target domain is input into a pre-trained fault diagnosis model to obtain test set sample scores. Based on the distance from the score of each sample in the test set to the class center and the distance threshold, the test set samples are divided into known faults and unknown faults; The test set sample scores corresponding to known faults and unknown faults are differentially adjusted, and the probabilities of known faults and unknown faults are determined based on the adjusted scores.

2. The open-set domain generalized bearing fault diagnosis method under unknown operating conditions as described in claim 1, characterized in that, Using source domain bearing vibration data, a fault diagnosis model is pre-trained based on subdomain constraints and domain adversarial mechanisms, including: The fault diagnosis model includes a feature extractor and a classifier, and is pre-trained by minimizing the classification loss, combined with subdomain constraints and domain adversarial mechanisms. The loss function expression is as follows: ； In the formula, Indicates feature extractor, Represents a classifier. Represents classification loss, and This represents the hyperparameters of a balanced classification task. This represents the source domain discrimination loss after forward pass through GRL, where GRL stands for gradient reversal layer. Discriminator for the domain, Indicates the subdomain discrimination loss. This represents a subdomain discriminator.

3. The open-set domain generalized bearing fault diagnosis method under unknown operating conditions as described in claim 2, characterized in that, The domain discriminator and the subdomain discriminator Optimization is achieved by maximizing the discriminative loss, as shown in the following expression: ； In the formula, Indicates the discriminator, This indicates the source domain discrimination loss.

4. The open-set domain generalized bearing fault diagnosis method under unknown operating conditions as described in claim 1, characterized in that, The distance from each sample score to the class center is calculated using the following formula: ； In the formula, Represents the weight parameters. Indicates sample The score, Indicates the fault category The class center, This represents the fault category index.

5. The open-set domain generalized bearing fault diagnosis method under unknown operating conditions as described in claim 1, characterized in that, The distance threshold is obtained through the following steps: Determine the mean and standard deviation based on the distance distribution from known fault samples within the source domain to their respective class centers; Determine the distance threshold based on the mean and standard deviation: ； In the formula, and Let represent the mean and standard deviation of the distance distribution, respectively. This represents the scaling factor.

6. The open-set domain generalized bearing fault diagnosis method under unknown operating conditions as described in claim 1, characterized in that, The test set samples are divided into known faults and unknown faults, including: When the minimum distance is greater than the distance threshold, the corresponding test set sample is determined to be without fault; otherwise, it is determined to be a known fault.

7. The method for generalizing bearing fault diagnosis under unknown operating conditions in an open set domain as described in claim 1, characterized in that, The test set sample scores corresponding to known and unknown faults are differentially adjusted, and the probabilities of known and unknown faults are determined based on the adjusted scores, including: When a fault is identified as an unknown fault, the scores of the remaining samples identified as known faults are scaled. ； in, Indicates the scaling factor. Indicates the known fault category The sample score, This represents the total number of known fault categories. Indicates the index of known fault categories. Indicates the adjusted fault category Sample score; The known failure probabilities are: ； The probability of an unknown fault is: ； In the formula, Indicates the sample failure category number. This represents the scaled score of the k-th sample. This represents a parameter that controls the probability of the unknown class.

8. The open-set domain generalized bearing fault diagnosis method under unknown operating conditions as described in claim 1, characterized in that, The test set sample scores corresponding to known and unknown faults are differentially adjusted, and the probabilities of known and unknown faults are determined based on the adjusted scores, including: When a known fault is identified, the one with the highest score is selected. Each category is assigned a decreasing weight factor in order of sorting: ； The scores of the known fault samples are adjusted according to the weighting factors as follows: ； In the formula, Indicates faulty sample Linear decay weights, Indicates the known fault category The sample score, Indicates the index of known fault categories. Indicates the adjusted fault category Sample score; The known failure probabilities are: ； The probability of an unknown fault is: ； In the formula, Indicates the known probability of failure. This represents the total number of known fault categories. Indicates the probability of an unknown failure. This represents the adjusted score for the k-th known category, where k represents the sample fault category number.

9. A generalized bearing fault diagnosis system for open-set domains under unknown operating conditions, characterized in that, The method for generalizing bearing fault diagnosis under unknown operating conditions according to any one of claims 1 to 8 includes: The pre-training module is used to pre-train the fault diagnosis model based on subdomain constraints and domain adversarial analysis using source domain bearing vibration data. The class center construction module is used to construct class centers based on the average activation vector of known fault categories in the source domain. The fault prediction module is used to acquire bearing vibration signals in the target domain in real time, input them into the pre-trained fault diagnosis model, and obtain test set sample scores. The fault classification module is used to classify the test set samples into known faults and unknown faults based on the distance from the score of each sample in the test set to the class center and the distance threshold. The post-fault processing module is used to differentiate the test set sample scores corresponding to known faults and unknown faults, and to determine the probability of known faults and the probability of unknown faults based on the adjusted scores.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the open set domain generalized bearing fault diagnosis method under unknown operating conditions as described in any one of claims 1 to 8.

Images