Time-varying working condition transfer learning bearing fault diagnosis method based on diffusion model

Abstract

The invention provides a time-varying working condition transfer learning bearing fault diagnosis method based on a diffusion model, and the method comprises the steps: obtaining and preprocessing an original fault vibration signal of a bearing, and obtaining a two-dimensional time-frequency spectrum image; constructing a de-noising diffusion model, and enabling the de-noising diffusion model to learn the image features of the two-dimensional time-frequency spectrum image under the stable working condition; through a Repaint method of a two-dimensional time-frequency spectrum image, a complete bearing fault data set under a stable working condition is constructed based on a vibration signal of a time-varying working condition; and carrying out bearing fault diagnosis by utilizing the expanded bearing fault data set and a field generalization method. According to the method, the problem of low bearing fault diagnosis accuracy caused by inconsistent distribution of field generalization bearing fault diagnosis data is solved, the bearing fault diagnosis accuracy is improved, and the method is more suitable for bearing fault diagnosis of different bearing application scenes.

Description

Technical Field

[0001] The present invention relates to the technical field of high-speed train bearing fault diagnosis, and in particular to a bearing fault diagnosis method based on time-varying working condition transfer learning of a diffusion model. Background Art

[0002] Intelligent fault diagnosis of rotating machinery plays an important role in industrial production. As an important component of the vehicle, the bogie of a high-speed train bears the weight of the train and ensures its smooth operation. Therefore, timely and effective condition monitoring and fault diagnosis are crucial to ensure the safe operation of the train. However, current condition monitoring and fault diagnosis methods face many challenges in processing massive monitoring data. For high-speed trains, a large amount of monitoring data will be generated during the operation of the train, and these data cover different working conditions. However, when processing these complex data, the existing diagnostic models usually have limited generalization capabilities, making it difficult to maintain high-precision diagnostic capabilities under various working conditions, which limits the practical application effect and reliability of the fault diagnosis system. At the same time, the label data contained in the massive monitoring data is often scarce and concentrated on specific working conditions, which further increases the difficulty of fault diagnosis. Therefore, how to complete the fault diagnosis task with limited available data is an urgent problem to be solved in intelligent fault diagnosis.

[0003] In order to meet the needs of intelligent fault diagnosis under time-varying working conditions in the current scenario of scarce sample data, a direct and effective method is to use domain migration technology. The current mainstream intelligent diagnosis solution under time-varying working conditions adopts the domain generalization method. However, the domain generalization method requires effective fault diagnosis data, and the limited amount of data cannot train an effective domain generalization model.

[0004] The denoising diffusion model has become one of the efficient and effective methods for generating data due to its excellent performance and training stability. The images generated by the denoising diffusion model are generated based on the trained images, while more data in bearing fault diagnosis are stable data, but more data in industrial applications are in a time-varying state, which leads to the low accuracy of bearing fault diagnosis in the domain generalization model due to inconsistent data distribution. Therefore, it is necessary to target the data characteristics of stable working conditions, let the data under time-varying working conditions be generated based on the data under stable working conditions, and improve the accuracy of domain generalization bearing fault diagnosis. Therefore, it is necessary to develop a model that allows a small amount of time-varying working condition data to learn stable working condition data based on its current state, so as to solve the problem of low fault diagnosis accuracy caused by inconsistent data distribution in domain generalization. Summary of the invention

[0005] The purpose of the present invention is to provide a bearing fault diagnosis method generalized in the field of time-varying working conditions in response to the deficiencies existing in the above-mentioned background technology. When the time-varying working condition data is limited, a bearing fault diagnosis signal under stable working conditions is generated according to some characteristics of the time-varying working conditions, so as to solve the problem of low fault diagnosis accuracy caused by inconsistent distribution of generalized bearing fault diagnosis data in the field.

[0006] In order to achieve the above object, the present invention provides a bearing fault diagnosis method based on time-varying working condition transfer learning of diffusion model, comprising the following steps:

[0007] S1, obtaining the original fault vibration signal of the bearing and preprocessing it to obtain a two-dimensional time-frequency spectrum image;

[0008] S2, constructing a denoising diffusion model, putting the acquired two-dimensional time-frequency spectrum image of the stable working condition into the denoising diffusion model for model training, so that the denoising diffusion model learns the image features of the two-dimensional time-frequency spectrum image of the stable working condition;

[0009] S3, through the Repaint method of two-dimensional time-frequency spectrum image, the vibration signal of the time-varying working condition is masked according to the established masking rules. The Repaint method generates a continuous stable vibration signal that conforms to the law of the vibration signal from the masked area based on the unmasked part of the vibration signal of the unstable working condition and the image features learned by the denoising diffusion model, and constructs a complete bearing fault data set under stable working conditions;

[0010] S4, bearing fault diagnosis is performed using the bearing fault dataset expanded by S3 and the domain generalization method.

[0011] Furthermore, the bearing fault vibration signal is collected in S1, and the one-dimensional vibration signal is converted into a two-dimensional time-frequency spectrum image using time-frequency analysis of complex morlet wavelet transform.

[0012] Furthermore, S2 includes the following sub-steps:

[0013] S21, the two-dimensional time-frequency spectrum image x0 of the stable working condition is input into the denoising diffusion model. At each step t, the denoising diffusion model adds Gaussian noise to the two-dimensional time-frequency spectrum image through the following formula, so that the two-dimensional time-frequency spectrum image becomes a Gaussian white noise image with independent distribution:

[0014]

[0015] Where t represents the step length, t=1,2,...,T; α t represents the adjustment parameter of the tth step, and its size decreases as the step size increases; ε t It follows a standard normal distribution The sampling noise of

[0016] S22, gradually denoise the Gaussian white noise image with independent distribution and restore it to a two-dimensional time-frequency spectrum image similar to the original image. T Initially, step-by-step noise removal is performed at each step t, as shown below:

[0017]

[0018] Among them, ε θ (x t ,t) is the prediction noise learned by S21; σ t is the adjustment parameter;

[0019] S23, repeating S21 to S22 for multiple times to obtain a denoising diffusion model capable of generating a two-dimensional time-frequency spectrum image of a stable working condition, and saving the denoising diffusion model.

[0020] Furthermore, S3 includes the following sub-steps:

[0021] S31, longitudinally masking the two-dimensional time-frequency spectrum image of the time-varying working condition;

[0022] S32, generating a stable two-dimensional time-frequency spectrum image for the masked area by using the Repaint method;

[0023] S33, repeat S32 multiple times until the generated image includes a complete two-dimensional time-frequency spectrum image of a stable working condition in which the image of the Repaint area is fused with the original image.

[0024] Furthermore, in S31, the left 1 / 8 position of the two-dimensional time-frequency spectrum image of the time-varying working condition is retained, and the right side is covered with a black image.

[0025] Furthermore, S4 includes the following sub-steps:

[0026] S41, taking the bearing fault data of the stable working condition at low speed and the generated bearing fault data of the stable working condition as the source domain, and taking the bearing fault data of the stable working condition at high speed as the target domain of domain generalization;

[0027] S42, training the model using the domain generalization method based on the source domain data, diagnosing the bearing fault on the target domain data, outputting the accuracy of the bearing fault diagnosis, and determining the bearing fault condition.

[0028] Furthermore, domain generalization methods include MMD, Mixup, ERM, DANN and ANDMask methods.

[0029] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned method for diagnosing bearing faults by time-varying working condition transfer learning based on a diffusion model.

[0030] The present invention also provides an electronic device, comprising at least one processor and a memory communicatively connected to the at least one processor, wherein 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 a bearing fault diagnosis method based on time-varying working condition transfer learning of a diffusion model as described above.

[0031] The above scheme of the present invention has the following beneficial effects:

[0032] The time-varying working condition transfer learning bearing fault diagnosis method based on the diffusion model provided by the present invention can generate bearing fault data under stable working conditions by using only a small amount of bearing fault data under unstable working conditions; the method only needs to train a denoising diffusion model that can generate stable working condition data to generate bearing fault data under specified working conditions, and the generated data of the specified working condition has the same data distribution characteristics as the existing stable working condition data, which solves the problem of low bearing fault diagnosis accuracy caused by inconsistent distribution of domain generalized bearing fault diagnosis data, improves the accuracy of bearing fault diagnosis, and is more suitable for bearing fault diagnosis in different bearing application scenarios;

[0033] This method is not limited to the bearing fault diagnosis of high-speed train bogies, but is also applicable to the intelligent fault diagnosis of rotating machinery in other fields. Of course, it is not limited to the fault diagnosis of rotating machinery bearings. Gearboxes and planetary gears in rotating machinery all involve intelligent fault diagnosis, and this method is also applicable to the intelligent fault diagnosis of such rotating machinery.

[0034] Other beneficial effects of the present invention will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 is a flow chart of the steps of the present invention;

[0036] Figure 2 Schematic diagram of the process of performing the Repaint method on a two-dimensional time-frequency spectrum image in an embodiment of the present invention, wherein (a) is a two-dimensional time-frequency spectrum image under time-varying conditions, (b) is a two-dimensional time-frequency spectrum image with a 7 / 8 mask, and (c) is a two-dimensional time-frequency spectrum image under stable conditions after the Repaint method;

[0037] Figure 3 A schematic diagram of a high-speed train bearing experimental device in an embodiment of the present invention;

[0038] Figure 4 Schematic diagram of the result of bearing fault diagnosis accuracy in an embodiment of the present invention; DETAILED DESCRIPTION

[0039] The following describes the embodiments of the present disclosure through specific examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all of the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present disclosure.

[0040] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein may be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present disclosure, it should be understood by those skilled in the art that an aspect described herein may be implemented independently of any other aspect, and two or more of these aspects may be combined in various ways. For example, any number of aspects described herein may be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein may be used to implement this device and / or practice this method.

[0041] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present disclosure, and the illustrations only show the components related to the present disclosure rather than the number, shape and size of the components in actual implementation. The type, quantity and proportion of each component in actual implementation may be changed at will, and the component layout type may also be more complicated. In addition, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the aspects described can be practiced without these specific details.

[0042] like Figure 1 As shown, an embodiment of the present invention provides a bearing fault diagnosis method based on time-varying working condition transfer learning of a diffusion model, which specifically includes the following steps:

[0043] S1, obtain the original fault vibration signal of the bearing and perform preprocessing.

[0044] In this step, for high-speed train bearing fault diagnosis, the bearing fault vibration signal of the high-speed train is collected, and the one-dimensional vibration signal is converted into a two-dimensional time-frequency spectrum image using time-frequency analysis of complex morlet wavelet transform.

[0045] It should be noted that the bandwidth of the complex morlet wavelet transform used in this embodiment is 3, the center frequency is 3, and the total scale is 256. The bearing fault vibration signal in this embodiment includes fault data of four types of high-speed train actual vehicle bearings.

[0046] S2, construct a denoising diffusion model that can generate two-dimensional time-frequency spectrum images of stable working conditions.

[0047] In this step, the acquired two-dimensional time-frequency spectrum image of the stable working condition is put into the denoising diffusion model for model training, so that the denoising diffusion model can learn the image features of the two-dimensional time-frequency spectrum image of the stable working condition.

[0048] Specifically in this embodiment, S2 can be decomposed into the following sub-steps:

[0049] S21, the two-dimensional time-frequency spectrum image x0 of the stable working condition is input into the denoising diffusion model. At each step t (t = 1, 2, ..., T), the denoising diffusion model adds Gaussian noise to it through the following formula to turn it into a Gaussian white noise image with independent distribution of each item:

[0050]

[0051] Among them, t represents the step size, that is, the current number of iterations; α t represents the adjustment parameter of the tth step, and its size decreases as the step size increases; ε t It follows a standard normal distribution The sampling noise.

[0052] S22, gradually denoise the Gaussian white noise image with independent distribution and restore it to a two-dimensional time-frequency spectrum image similar to the original image, that is, for the Gaussian pure noise data sample x T Initially, step-by-step noise removal is performed at each step t (t = T, T-1, ..., 1), as shown in the following formula:

[0053]

[0054] Among them, ε θ (x t ,t) is the prediction noise learned by S21; σ t It is to adjust the parameters to prevent the model from falling into the local optimum.

[0055] S23, repeating S21 to S22 for multiple times to obtain a denoising diffusion model that can generate a two-dimensional time-frequency spectrum image of a stable working condition, and saving the trained denoising diffusion model.

[0056] S3, through the Repaint method of two-dimensional time-frequency spectrum image, a complete bearing fault data set under stable working conditions is constructed based on the vibration signal of time-varying working conditions.

[0057] In this step, the vibration signal of the time-varying working condition (unstable working condition) is masked according to the established masking rules. The Repaint method generates a continuous stable vibration signal that conforms to the law of the vibration signal from the masked area based on the unmasked part of the vibration signal of the unstable working condition and the image features learned by the denoising diffusion model, thereby constructing a complete bearing fault data set under stable working conditions.

[0058] The purpose of this step is to make the two-dimensional time-frequency spectrum image maintain the characteristics of a small part of the original speed, and at the same time generate the data of the current stable working condition according to the existing denoising diffusion model, which finally solves the problem of data scarcity under time-varying working conditions in the field generalization process.

[0059] It should be noted that the Repaint method is a technology for data generation and repair, which is mainly used in the fields of signal processing and computer vision. The Repaint method is mainly used to restore a complete image from a partially missing or damaged image, or to generate a continuous signal that conforms to a specific rule in signal processing. In this embodiment, by inputting a partially masked time-varying signal into a denoising diffusion model, and using the image features of the stable signal learned by the model to generate a complete stable signal, the robustness of signal processing can be effectively improved, helping to generate a high-quality data set for further analysis and diagnosis.

[0060] At the same time Figure 2 As shown, specifically in this embodiment, S3 can be decomposed into the following sub-steps:

[0061] S31, masking the two-dimensional time-frequency spectrum image of the time-varying working condition vertically. As a preferred implementation, the left 1 / 8 position of the two-dimensional time-frequency spectrum image of the time-varying working condition is reserved, and the right side is masked with a black image.

[0062] S32, generating a stable two-dimensional time-frequency spectrum image for the masked area through a denoising diffusion model;

[0063] S33, repeating S32 for multiple times, and finally generating an image including a two-dimensional time-frequency spectrum image of a stable working condition in which the image of the Repaint area is perfectly integrated with the original image and is complete.

[0064] S4, generalizes bearing fault diagnosis in the field of time-varying working conditions.

[0065] In this step, bearing fault diagnosis is performed using the S3-expanded bearing fault dataset and domain generalization method, which can be broken down into the following sub-steps:

[0066] S41, taking the bearing fault data of the stable working condition at low speed and the generated bearing fault data of the stable working condition as the source domain, and taking the bearing fault data of the stable working condition at high speed as the target domain of domain generalization.

[0067] S42, training the model using the domain generalization method based on the source domain data, diagnosing the bearing fault on the target domain data, outputting the accuracy of the bearing fault diagnosis, and determining the bearing fault condition.

[0068] In this embodiment, the domain generalization methods include MMD, Mixup, ERM, DANN and ANDMask methods. Maximum Mean Discrepancy (MMD) is a distance metric used for domain alignment, which measures the distribution difference between the source domain and the target domain data, and the goal is to minimize the distribution difference between the source domain and the target domain. Mixup is a data enhancement method that enhances the robustness of the model by generating new training samples by linearly interpolating samples during training. Empirical Risk Minimization (ERM) is a classic training strategy that aims to train the model by minimizing the loss function on the training set. The Domain-Adversarial Neural Network (DANN) reduces the difference between the source domain and the target domain through adversarial training. The Adaptive Network Masking (ANDMask) aims to align the source domain and target domain data by adaptively adjusting the feature space.

[0069] Therefore, this method only uses a small amount of bearing fault data under unstable conditions to generate bearing fault data under stable conditions; this method only needs to train a denoising diffusion model that can generate stable condition data to generate bearing fault data under specified conditions, and the generated data for the specified conditions has the same data distribution characteristics as the existing data for stable conditions, which solves the problem of low bearing fault diagnosis accuracy caused by inconsistent data distribution of domain generalized bearing fault diagnosis.

[0070] It should be noted that this method is not limited to the bearing fault diagnosis of high-speed train bogies, but is also applicable to the intelligent fault diagnosis of rotating machinery in other fields. Of course, it is not limited to the fault diagnosis of rotating machinery bearings. Gearboxes, planetary gears, etc. in rotating machinery all involve intelligent fault diagnosis, and this method is also applicable to the intelligent fault diagnosis of this type of rotating machinery.

[0071] The following is a specific case to further illustrate the effect of this method. The data of the two-dimensional time-frequency spectrum image that has not been repainted is selected for domain generalization as a reference to verify the effectiveness of bearing fault diagnosis after repainting. Among them, in order to ensure the consistency of the data volume, the slicing method used for the variable speed data that has not been repainted is that the next two-dimensional time-frequency spectrum image retains part of the data of the previous two-dimensional time-frequency spectrum image, and the slicing method used for the variable speed data that has been repainted is that there is no repeated vibration information between adjacent two-dimensional time-frequency spectrum images.

[0072] This case uses real high-speed train bogie bearing failure data for verification. The NTNCRI-2692 double-row tapered roller bearing for high-speed trains is used as the research object. Normal, indented, spalled and corroded specimens are selected and installed on two sets of bogies in the entire rolling test bench. The position of the bearing is as follows Figure 3 As shown, the left and right sides of the bogie are divided into left and right bearings. The four axes of the two bogies are defined as the first axis, the second axis, the third axis and the fourth axis from the front end to the rear end. Different types of bearing faults are installed in the train for data collection. The test objects cover 8 bearings, and 2 measurement points are arranged on each bearing, with a total of 16 sensors. The test speed range is 100km / h, 100-150km / h and 150km / h, the sampling time is 20s, and the sampling frequency is 25.6KHz.

[0073] Based on the collected bearing fault data, this method is used to diagnose the bearing fault. First, the one-dimensional data of the high-speed train bogie bearing fault is converted into a two-dimensional time-frequency spectrum image, and then the two-dimensional time-frequency spectrum images of the bearing fault at speeds of 100km / h and 150km / h are used to train the denoising diffusion model; the two-dimensional time-frequency spectrum images of the bearing fault at 100-150km / h are masked by 7 / 8, and the image of the mask position is generated using the trained denoising diffusion model. Finally, the two-dimensional time-frequency spectrum images of 100km / h and the generated 100-150km / h are used as the source domain, and the two-dimensional time-frequency spectrum image of 150km / h is used as the target domain. The domain generalization method is used to diagnose the bearing fault, and the accuracy of the bearing fault diagnosis is output. The higher the accuracy of the bearing fault diagnosis, the more accurate the diagnosis, such as Figure 4 The comparison results are shown in Table 1.

[0074] Table 1 Test results of domain generalized bearing fault diagnosis before and after Repaint

[0075] Domain Generalization Method MMD Mixup ERM DANN ANDmask No Repaint 87.38 92.19 78.54 82.18 83.03 Repaint 97.23 97.4 98.13 96.73 92.55

[0076] Depend on Figure 4As can be seen from Table 1, the bearing fault diagnosis accuracy after Repaint is significantly higher than that without Repaint. This further proves that the method provided in this embodiment can solve the problem of low bearing fault diagnosis accuracy caused by inconsistent distribution of domain generalized bearing fault diagnosis data, and improve the accuracy of bearing fault diagnosis.

[0077] Based on the same inventive concept, this embodiment also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the aforementioned time-varying working condition transfer learning bearing fault diagnosis method based on a diffusion model is implemented.

[0078] The computer readable medium includes, but is not limited to, any type of disk (including floppy disk, hard disk, optical disk, CD-ROM, and magneto-optical disk), ROM, RAM, EPROM (Erasable Programmable Read-Only Memory), EEPROM, flash memory, magnetic card or optical card. In other words, the readable medium includes any medium that can store or transmit information in a readable form by a device (such as a computer).

[0079] Based on the same inventive concept, this embodiment also provides an electronic device, including at least one processor and a memory connected to the at least one processor in communication. The memory stores instructions that can be executed 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 aforementioned time-varying working condition transfer learning bearing fault diagnosis method based on a diffusion model.

[0080] The computer-readable storage medium, electronic device, etc. provided in this embodiment have the same inventive concept and the same beneficial effects as the aforementioned method, and will not be described in detail here.

[0081] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0082] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A bearing fault diagnosis method based on time-varying working condition transfer learning of diffusion model, characterized in that: The steps include: S1, obtaining the original fault vibration signal of the bearing and preprocessing it to obtain a two-dimensional time-frequency spectrum image; S2, constructing a denoising diffusion model, putting the acquired two-dimensional time-frequency spectrum image of the stable working condition into the denoising diffusion model for model training, so that the denoising diffusion model learns the image features of the two-dimensional time-frequency spectrum image of the stable working condition; S3, through the Repaint method of two-dimensional time-frequency spectrum image, the vibration signal of the time-varying working condition is masked according to the established masking rules. The Repaint method generates a continuous stable vibration signal that conforms to the law of the vibration signal from the masked area based on the unmasked part of the vibration signal of the unstable working condition and the image features learned by the denoising diffusion model, and constructs a complete bearing fault data set under stable working conditions; S4, bearing fault diagnosis is performed using the bearing fault dataset expanded by S3 and the domain generalization method.

2. According to claim 1, a time-varying working condition transfer learning bearing fault diagnosis method based on diffusion model is characterized in that: The bearing fault vibration signal is collected in S1, and the one-dimensional vibration signal is converted into a two-dimensional time-frequency spectrum image using time-frequency analysis of complex morlet wavelet transform.

3. The method for bearing fault diagnosis based on time-varying working condition transfer learning of diffusion model according to claim 1 is characterized in that: S2 includes the following sub-steps: S21, the two-dimensional time-frequency spectrum image x0 of the stable working condition is input into the denoising diffusion model. At each step t, the denoising diffusion model adds Gaussian noise to the two-dimensional time-frequency spectrum image through the following formula, so that the two-dimensional time-frequency spectrum image becomes a Gaussian white noise image with independent distribution: Where t represents the step length, t=1,2,...,T; α t represents the adjustment parameter of the tth step, and its size decreases as the step length increases; ε t It follows a standard normal distribution The sampling noise of S22, gradually denoise the Gaussian white noise image with independent distribution and restore it to a two-dimensional time-frequency spectrum image similar to the original image. T Initially, step-by-step noise removal is performed at each step t, as shown below: Among them, ε θ (x t ,t) is the prediction noise learned by S21; σ t is the adjustment parameter; S23, repeating S21 to S22 for multiple times to obtain a denoising diffusion model capable of generating a two-dimensional time-frequency spectrum image of a stable working condition, and saving the denoising diffusion model.

4. The method for bearing fault diagnosis based on time-varying working condition transfer learning of diffusion model according to claim 1 is characterized in that: S3 includes the following sub-steps: S31, longitudinally masking the two-dimensional time-frequency spectrum image of the time-varying working condition; S32, generating a stable two-dimensional time-frequency spectrum image for the masked area by using the Repaint method; S33, repeat S32 multiple times until the generated image includes a complete two-dimensional time-frequency spectrum image of a stable working condition in which the image of the Repaint area is fused with the original image.

5. The method for bearing fault diagnosis based on time-varying working condition transfer learning of diffusion model according to claim 4 is characterized in that: In S31, the left 1 / 8 position of the two-dimensional time-frequency spectrum image of the time-varying working condition is retained, and the right side is covered with a black image.

6. The method for bearing fault diagnosis based on time-varying working condition transfer learning of diffusion model according to claim 1 is characterized in that: S4 includes the following sub-steps: S41, taking the bearing fault data of the stable working condition at low speed and the generated bearing fault data of the stable working condition as the source domain, and taking the bearing fault data of the stable working condition at high speed as the target domain of domain generalization; S42, training the model using the domain generalization method based on the source domain data, diagnosing the bearing fault on the target domain data, outputting the accuracy of the bearing fault diagnosis, and determining the bearing fault condition.

7. The method for bearing fault diagnosis based on time-varying working condition transfer learning of diffusion model according to claim 6 is characterized in that: Domain generalization methods include MMD, Mixup, ERM, DANN and ANDMask methods.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, a bearing fault diagnosis method based on time-varying working condition transfer learning of a diffusion model is implemented as described in any one of claims 1 to 7.

9. An electronic device, characterized in that: It includes at least one processor and a memory that is communicatively connected to the at least one processor, wherein the memory stores instructions that can be executed 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 a bearing fault diagnosis method based on time-varying working condition transfer learning based on a diffusion model as described in any one of claims 1 to 7.

Images