Simulation data driven rolling bearing fault diagnosis method

Abstract

The invention provides a rolling bearing fault diagnosis method driven by simulation data, and the method comprises the steps: constructing a five-degree-of-freedom dynamical model of a rolling bearing according to the failure forms of different fault types of the bearing, so as to represent the dynamic response of a bearing system in a virtual space; the method comprises the following steps: collecting vibration signals under healthy and different fault conditions in a real environment through a bearing fault simulation experiment table, and establishing a data set; the feature distribution difference between the simulation data and the measurement data is considered, and four optimization targets are constructed and are respectively used for difference feature learning and domain invariant feature learning; a Transform model is used as a feature extractor, and fault diagnosis is carried out under a cross-domain condition in combination with a mixed loss function. Compared with the prior art, the model established by the invention can perform effective diagnosis under cross-domain and sparse training data conditions, solves the problems of data feature distribution difference and sparse target domain samples in engineering practice, and has certain robustness.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent fault diagnosis of rolling bearings, and in particular to a rolling bearing fault diagnosis method driven by simulation data. Background Art

[0002] As a key component in industrial machinery, rolling bearings play a vital role in the transmission systems of modern power machinery, such as ships, wind turbines, and aircraft. With the development of modern technology, the working scenarios faced by rolling bearings are becoming more and more complex, which accelerates the degradation process of rolling bearings. Once a rolling bearing fails, it will cause a lot of downtime and maintenance costs, and in severe cases even cause catastrophic damage and huge economic losses. Therefore, the development of advanced diagnostic methods for rolling bearings is crucial to enhance industrial maintenance strategies.

[0003] In recent years, deep learning techniques have made significant progress in the field of fault diagnosis, but these methods usually assume that the test data and training data come from the same data distribution. However, this assumption does not hold true in industrial applications because the operating environment, working conditions or noise levels of mechanical equipment under actual working conditions vary greatly. In order to address the problem of data distribution differences in bearing fault diagnosis, a variety of domain adaptation techniques based on transfer learning have been developed to narrow the gap between the source domain and the target domain with labeled data. However, existing methods often require that the source domain data contain information about all possible fault modes. Therefore, when the source domain data is insufficient, existing domain adaptation techniques may fail.

[0004] In recent years, digital twin technology has attracted widespread attention due to its huge transformative potential. By generating high-quality simulation data, digital twin technology can enhance fault diagnosis and help solve the problem of insufficient source domain data. Digital twins create virtual models of physical systems and simulate real working scenarios to generate diverse fault data. This capability can make up for the lack of source domain data and improve the robustness of fault diagnosis methods. By integrating these simulation data sets, it is expected to overcome the limitations of incomplete real data and achieve more accurate and reliable fault detection and analysis. Summary of the invention

[0005] According to the above technical problems of working condition changes and sparse training data in engineering practice, a simulation data-driven rolling bearing fault diagnosis method is proposed. The present invention first generates sufficient high-quality source domain data samples through a dynamic model, then collects vibration signals in a real environment through a test bench, and finally extracts high-level features from fault signals through a Transformer model, and further realizes global domain adaptation by combining four domain adaptation loss functions, thereby enhancing the diagnostic performance in the target domain.

[0006] The technical means adopted by the present invention are as follows:

[0007] A rolling bearing fault diagnosis method driven by simulation data, comprising:

[0008] S1. According to the failure modes of different bearing fault types, a five-degree-of-freedom dynamic model of the rolling bearing is constructed to characterize the dynamic response of the bearing system in virtual space;

[0009] S2. Collect vibration signals under healthy and different fault conditions in real environment through bearing fault simulation test bench to establish data set;

[0010] S3, considering the difference in feature distribution between simulated data and measured data, construct four optimization objectives for differential feature learning and domain invariant feature learning respectively;

[0011] S4. Use the Transformer model as a feature extractor and combine it with a hybrid loss function to perform fault diagnosis under cross-domain conditions.

[0012] Furthermore, step S1 specifically includes:

[0013] S11. When there is a defect in the outer ring of the rolling bearing, calculate the contact deformation between the rolling element and the raceway when passing through the defect as follows:

[0014] δ j =(x i -x o )cosθ j +(y i -y o )sinθ j -γ-δ o

[0015] Among them, x i , x o Respectively represent the horizontal displacement of the inner and outer rings; y i ,y o Respectively represent the vertical displacement of the inner and outer rings; θ j The position angle of the jth rolling element in the circumferential direction represents the radial clearance of the bearing; δ o Indicates the contact deformation when there is a defect in the outer ring;

[0016] S12. When there is a defect in the inner ring of the rolling bearing, calculate the contact deformation between the rolling element and the raceway when passing through the defect, as follows:

[0017] δ j =(x i -x o )cosθ j +(y i -y o)sinθ j -γ-δ i

[0018] Among them, δ i Indicates the contact deformation when there is a defect in the inner ring;

[0019] S13. When there is a defect on the rolling element, calculate the contact deformation between the rolling element and the raceway when the rolling element contacts the raceway, as follows:

[0020] δ j =(x i -x o )cosθ j +(y i -y o )sinθ j -γ-δ r

[0021] Among them, δ r Indicates contact deformation when there are defects in the rolling element;

[0022] S14. According to Newton's laws of motion, the dynamic differential equation of the rolling bearing is established as follows:

[0023]

[0024] Among them, m i , m o and m r Respectively represent the mass of the inner ring, outer ring and unit resonator; c i , c o and c r Represent the damping of the inner ring, outer ring and unit resonator respectively; k i , k o and k r Represent the stiffness of the inner ring, outer ring and unit resonator respectively; W represents the radial load applied to the inner ring; F x , F y Represents the contact force in the horizontal and vertical directions respectively; y r represents the y-direction displacement of the unit resonator;

[0025] S15, solve the dynamic differential equation of rolling bearing by the fourth-order fixed-step Runge-Kutta method, with a calculation step size of 5××10 -5 s, output simulation signals under different fault types.

[0026] Furthermore, the five-degree-of-freedom dynamic model of the rolling bearing established in step S1 is used to characterize the dynamic response of the rolling bearing system, wherein the inner ring and the outer ring have two degrees of freedom respectively, and the unit resonator has only one degree of freedom.

[0027] Further, step S2 specifically includes:

[0028] S21, setting two acceleration sensors in the horizontal direction and the vertical direction of the bearing seat to collect the vibration signal of the rolling bearing state monitoring;

[0029] S22, setting the sampling frequency and sampling time of each type;

[0030] S23. The simulation signal generated by dynamics and the measured signal collected by the experiment are divided into a source domain data set and a target domain data set through a sliding window algorithm.

[0031] Further, step S3 specifically includes:

[0032] S31. Introduce the source domain target classification loss function to ensure that the model accurately classifies samples from the labeled source domain data. The introduced function is as follows:

[0033]

[0034] Among them, x s and ys represent the samples and labels of the source domain respectively, n s Indicates the number of samples in the source domain;

[0035] S32, introduce the maximum mean difference loss function to align the domain distribution boundary. The introduced function is as follows:

[0036]

[0037] in, and Represent the sample features of the source domain and the target domain respectively, and f(·) represents the mapping to the reproducing kernel Hilbert space through the kernel function. The feature map, n s and n t Represents the number of samples in the source domain and the target domain respectively, P(F(x) s ) and Q(F(x) t ) represent the feature distribution of source domain and target domain respectively;

[0038] S33. Introduce supervised contrast loss function to promote cross-domain robust feature learning. The introduced function is as follows:

[0039]

[0040] Among them, i represents the anchor point, P represents the positive sample set, N represents the negative sample set, and F p and F k They represent the feature vectors of positive and negative samples respectively, and τ represents the temperature parameter to control the smoothness of the similarity distribution;

[0041] S34, use data enhancement techniques such as adding Gaussian white noise, scaling the overall amplitude, and cropping to perform data enhancement to improve the generalization ability of unlabeled target domain samples;

[0042] S35. Introduce an unsupervised contrast loss function to improve the clustering performance in the unlabeled target domain. The introduced function is as follows:

[0043]

[0044] Among them, F a Represents the features after data enhancement operation;

[0045] S36. Combine the introduced source domain target classification loss function, maximum mean difference loss function, supervised contrast loss function and unsupervised contrast loss function to achieve global and instance-level domain alignment. The combined function is as follows:

[0046]

[0047] Among them, λ, α, and β all represent weight parameters.

[0048] Further, step S4 specifically includes:

[0049] S41. Use the multi-head self-attention mechanism in the Transformer model to capture the long-range dependencies of the input data;

[0050] S42. Convert a set of input embeddings x into query items, keys and values ​​of d dimensions respectively. The expressions are as follows:

[0051]

[0052] in, Both represent trainable projection matrices;

[0053] S43. Calculate the single-head scaled dot product attention by the scaling factor. The calculation formula is as follows:

[0054]

[0055] in, represents the scaling factor;

[0056] S44. Perform H single-head attention on the multi-head self-attention mechanism. The expression is as follows:

[0057]

[0058] head j =Attention(Q,K,V) j ,j=1,2,...,H

[0059] in, represents the trainable projection matrix, H represents the number of attention heads;

[0060] S45. Connect a multi-layer perceptron after the multi-head attention mechanism to map the features to a high-dimensional space and output the maximum probability category through the Softmax function. The expression is as follows:

[0061] MLP(x)=GELU(0,xW1+b1)W2+b2

[0062] Among them, W1 and W2 both represent weights, b1 and b2 both represent biases, and GELU represents the activation function.

[0063] Compared with the prior art, the present invention has the following advantages:

[0064] 1. The present invention provides a rolling bearing fault diagnosis method driven by simulation data. It uses a virtual bearing dynamics model to generate simulation data as the source domain data for fault diagnosis, and uses a Transformer model to extract and learn feature distribution differences from the original data for fault diagnosis, thereby solving the problems of data feature distribution differences and sparse target domain samples in engineering practice.

[0065] 2. The present invention provides a simulation data-driven rolling bearing fault diagnosis method. In order to enhance domain alignment, the maximum mean difference loss is used to implement a domain adaptation strategy, and combined with a supervised contrastive learning method, the method integrates original and enhanced data for global domain alignment and instance-based alignment.

[0066] 3. The present invention provides a simulation data-driven rolling bearing fault diagnosis method, which applies unsupervised contrastive learning loss to the enhanced data of the target domain, further improving the diagnostic accuracy.

[0067] Based on the above reasons, the present invention can be widely promoted in the fields of rolling bearing intelligent fault diagnosis and the like. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0069] Figure 1 The figure is a flow chart of the method of the present invention.

[0070] Figure 2This is an outer race fault characterization provided by an embodiment of the present invention.

[0071] Figure 3 This is an inner race fault characterization provided by an embodiment of the present invention.

[0072] Figure 4 The present invention provides a rolling element fault characterization.

[0073] Figure 5 Comparison between the simulated signal and the real signal provided in the embodiment of the present invention.

[0074] Figure 6 A hybrid loss of the construction provided for an embodiment of the present invention.

[0075] Figure 7 The Transformer model provided by the embodiment of the present invention.

[0076] Figure 8 Diagnostic results of different models provided by the embodiments of the present invention.

[0077] Fig. 9 A confusion matrix provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0078] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0079] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0080] like Figure 1 As shown, the present invention provides a rolling bearing fault diagnosis method driven by simulation data, comprising:

[0081] S1. According to the failure modes of different bearing fault types, a five-degree-of-freedom dynamic model of the rolling bearing is constructed to characterize the dynamic response of the bearing system in virtual space;

[0082] S2. Collect vibration signals under healthy and different fault conditions in real environment through bearing fault simulation test bench to establish data set;

[0083] S3, considering the difference in feature distribution between simulated data and measured data, construct four optimization objectives for differential feature learning and domain invariant feature learning respectively;

[0084] S4. Use the Transformer model as a feature extractor and combine it with a hybrid loss function to perform fault diagnosis under cross-domain conditions.

[0085] In specific implementation, as a preferred embodiment of the present invention, step S1 specifically includes:

[0086] S11. When there is a defect in the outer ring of the rolling bearing, calculate the contact deformation between the rolling element and the raceway when passing through the defect as follows:

[0087] δ j =(x i -x o )cosθ j +(y i -y o )sinθ j -γ-δ o

[0088] Among them, x i , x o Respectively represent the horizontal displacement of the inner and outer rings; y i ,y o Respectively represent the vertical displacement of the inner and outer rings; θ j represents the position angle of the jth rolling element in the circumferential direction; γ represents the radial clearance of the bearing; δ o Indicates the contact deformation when there is a defect in the outer ring;

[0089] S12. When there is a defect in the inner ring of the rolling bearing, calculate the contact deformation between the rolling element and the raceway when passing through the defect, as follows:

[0090] δ j =(x i -x o )cosθ j +(y i -y o )sinθ j -γ-δ i

[0091] Among them, δ iIndicates the contact deformation when there is a defect in the inner ring;

[0092] S13. When there is a defect on the rolling element, calculate the contact deformation between the rolling element and the raceway when the rolling element contacts the raceway, as follows:

[0093] δ j =(x i -x o )cosθ j +(y i -y o )sinθ j -γ-δ r

[0094] Among them, δ r Indicates contact deformation when there are defects in the rolling element;

[0095] S14. According to Newton's laws of motion, the dynamic differential equation of the rolling bearing is established as follows:

[0096]

[0097] Among them, m i , m o and m r Respectively represent the mass of the inner ring, outer ring and unit resonator; c i , c o and c r Represent the damping of the inner ring, outer ring and unit resonator respectively; k i , k o and k r Represent the stiffness of the inner ring, outer ring and unit resonator respectively; W represents the radial load applied to the inner ring; F x , F y Represents the contact force in the horizontal and vertical directions respectively; y r represents the y-direction displacement of the unit resonator;

[0098] S15, solve the dynamic differential equation of rolling bearing by the fourth-order fixed-step Runge-Kutta method, with a calculation step size of 5××10 -5 s, output simulation signals under different fault types.

[0099] In this embodiment, Figure 5 The time domain waveforms of the simulated and measured signals under different fault conditions at a speed of 1800rpm are shown. It can be observed that the bearing fault generates periodic pulses in the simulated and measured signals, and the pulse trends of the simulated and measured signals for each fault are similar. This illustrates the rationality of the virtual dynamics model.

[0100] In specific implementation, as a preferred embodiment of the present invention, the five-degree-of-freedom dynamic model of the rolling bearing established in step S1 is used to characterize the dynamic response of the rolling bearing system, wherein the inner ring and the outer ring have two degrees of freedom respectively, and the unit resonator has only one degree of freedom. Figure 2-4 As shown in the figure, it is the failure characterization model under different bearing fault types.

[0101] In specific implementation, as a preferred embodiment of the present invention, step S2 specifically includes:

[0102] S21, setting two acceleration sensors in the horizontal direction and the vertical direction of the bearing seat to collect the vibration signal of the rolling bearing state monitoring;

[0103] S22, setting the sampling frequency and the sampling time of each type; in this embodiment, the sampling frequency is set to 20 kHz; the sampling time of each type is set to 10 s.

[0104] S23, the dynamics generated simulation signal and the experimental measured signal are divided into source domain data set and target domain data set through the sliding window algorithm. Table 1 shows the five cross-domain fault diagnosis tasks constructed.

[0105] Table 1. Division of different tasks

[0106] Task Source domain (fault category) Target domain (fault category) Sample length <![CDATA[T1]]> Outer ring, inner ring, rolling element, health Outer ring, inner ring, rolling element, health 2048 <![CDATA[T2]]> Outer ring, inner ring, rolling element, health Outer circle, inner circle, health 2048 <![CDATA[T3]]> Outer ring, inner ring, rolling element, health Inner ring, rolling element, health 2048 <![CDATA[T4]]> Outer ring, inner ring, rolling element, health Outer ring, health 2048 <![CDATA[T5]]> Outer ring, inner ring, rolling element, health Inner circle, health 2048

[0107] In specific implementation, as a preferred embodiment of the present invention, step S3 specifically includes:

[0108] S31. Introduce the source domain target classification loss function to ensure that the model accurately classifies samples from the labeled source domain data. The introduced function is as follows:

[0109]

[0110] Among them, x s and s Represent the samples and labels of the source domain respectively, n s Indicates the number of samples in the source domain;

[0111] S32, introduce the maximum mean difference loss function to align the domain distribution boundary. The introduced function is as follows:

[0112]

[0113] in, and Represent the sample features of the source domain and the target domain respectively, and f(·) represents the mapping to the reproducing kernel Hilbert space through the kernel function. The feature map, n s and n tRepresents the number of samples in the source domain and the target domain respectively, P(F(x) s ) and Q(F(x) t ) represent the feature distribution of source domain and target domain respectively;

[0114] S33. Introduce supervised contrast loss function to promote cross-domain robust feature learning. The introduced function is as follows:

[0115]

[0116] Among them, i represents the anchor point, P represents the positive sample set, N represents the negative sample set, and F p and F k Represent the feature vectors of positive and negative samples respectively, and τ represents the temperature parameter to control the smoothness of the similarity distribution; the loss function pulls samples with the same labels together in the shared feature space and separates samples with different labels.

[0117] S34, use data enhancement techniques such as adding Gaussian white noise, scaling the overall amplitude, and cropping to perform data enhancement to improve the generalization ability of unlabeled target domain samples;

[0118] S35. Introduce an unsupervised contrast loss function to improve the clustering performance in the unlabeled target domain. The introduced function is as follows:

[0119]

[0120] Among them, F a Represents the features after data enhancement operation;

[0121] S36. The introduced source domain target classification loss function, maximum mean difference loss function, supervised contrast loss function and unsupervised contrast loss function are combined to achieve global and instance-level domain alignment, further improving the diagnostic accuracy under cross-domain conditions. The combined function is as follows:

[0122]

[0123] Among them, λ, α, and β all represent weight parameters. Figure 6 As shown, the integrated loss function process is demonstrated.

[0124] In specific implementation, as a preferred embodiment of the present invention, step S4 specifically includes:

[0125] S41, using the multi-head self-attention mechanism in the Transformer model to capture the long-range dependency of the input data; in this embodiment, the constructed Transformer network structure is as follows Figure 7 shown.

[0126] S42. Convert a set of input embeddings x into query items (Q), keys (K) and values ​​(V) of d dimensions respectively, as shown below:

[0127]

[0128] in, Both represent trainable projection matrices;

[0129] S43. Calculate the single-head scaled dot product attention by the scaling factor. The calculation formula is as follows:

[0130]

[0131] in, represents the scaling factor;

[0132] S44. Perform H single-head attention on the multi-head self-attention mechanism. The expression is as follows:

[0133]

[0134] head j =Attention(Q,K,V) j ,j=1,2,...,H

[0135] in, represents the trainable projection matrix, H represents the number of attention heads;

[0136] S45. Connect a multi-layer perceptron after the multi-head attention mechanism to map the features to a high-dimensional space and output the maximum probability category through the Softmax function. The expression is as follows:

[0137] MLP(x)=GELU(0,xW1+b1)W2+b2

[0138] Among them, W1 and W2 both represent weights, b1 and b2 both represent biases, and GELU represents the activation function.

[0139] like Figure 8The experimental results for all tasks are shown in Figure 2. Non-DA methods have the lowest accuracy, ranging from 44.29% to 50.87%, which highlights its inadequacy in addressing the substantial domain conversion between simulated and measured data, ultimately leading to suboptimal diagnosis results. In contrast, traditional domain adaptation methods, such as MMD and DANN, have shown significant improvements over non-DA, with accuracies ranging from 69.76% to 79.93%. The MMDSC method consistently outperforms MMD and DANN in all tasks, with accuracies ranging from 81.84% to 86.64%. It is worth noting that the proposed method achieves excellent performance in all evaluation tasks, with the highest accuracy ranging between 82.51% and 87.44%. This shows that the proposed method effectively mitigates domain differences while providing stable and accurate diagnosis.

[0140] like Fig. 9 As shown, the confusion matrix visualization is used to show the diagnosis results under all conditions. Non-DA methods have low diagnostic accuracy for RF, IF, and OF fault conditions. In contrast, methods using domain adaptation techniques such as MMD, DANN, and MMDSC show higher accuracy, especially for OF and IF conditions. However, the proposed method achieves the highest average accuracy among all methods, reflecting its superior ability in accurately diagnosing these fault conditions.

[0141] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A rolling bearing fault diagnosis method driven by simulation data, characterized in that: include: S1. According to the failure modes of different bearing fault types, a five-degree-of-freedom dynamic model of the rolling bearing is constructed to characterize the dynamic response of the bearing system in virtual space; S2. Collect vibration signals under healthy and different fault conditions in real environment through bearing fault simulation test bench to establish data set; S3, considering the difference in feature distribution between simulated data and measured data, construct four optimization objectives for differential feature learning and domain invariant feature learning respectively; S4. Use the Transformer model as a feature extractor and combine it with a hybrid loss function to perform fault diagnosis under cross-domain conditions.

2. A rolling bearing fault diagnosis method driven by simulation data according to claim 1, characterized in that: Step S1 specifically includes: S11. When there is a defect in the outer ring of the rolling bearing, calculate the contact deformation between the rolling element and the raceway when passing through the defect as follows: d j =(x i -x o )cosθ j +(y i -y o )sinθ j -c-d o Among them, x i , x o Respectively represent the horizontal displacement of the inner and outer rings; y i ,y o Respectively represent the vertical displacement of the inner and outer rings; θ j represents the position angle of the jth rolling element in the circumferential direction; γ represents the radial clearance of the bearing; δ o Indicates the contact deformation when there is a defect in the outer ring; S12. When there is a defect in the inner ring of the rolling bearing, calculate the contact deformation between the rolling element and the raceway when passing through the defect, as follows: d j =(x i -x o )cosθ j +(y i -y o )sinθ j -c-d i Among them, δ i Indicates the contact deformation when there is a defect in the inner ring; S13. When there is a defect on the rolling element, calculate the contact deformation between the rolling element and the raceway when the rolling element contacts the raceway, as follows: d j =(x i -x o )cosθ j +(y i -y o )sinθ j -c-d r Among them, δ r Indicates contact deformation when there are defects in the rolling element; S14. According to Newton's laws of motion, the dynamic differential equation of the rolling bearing is established as follows: Among them, m i , m o and m r Respectively represent the mass of the inner ring, outer ring and unit resonator; c i , c o and c r Represent the damping of the inner ring, outer ring and unit resonator respectively; k i , k o and k r Represent the stiffness of the inner ring, outer ring and unit resonator respectively; W represents the radial load applied to the inner ring; F x , F y Represents the contact force in the horizontal and vertical directions respectively; y r represents the y-direction displacement of the unit resonator; S15, solve the dynamic differential equation of rolling bearing by the fourth-order fixed-step Runge-Kutta method, with a calculation step size of 5××10 -5 s, output simulation signals under different fault types.

3. The rolling bearing fault diagnosis method driven by simulation data according to claim 1 is characterized in that: The five-degree-of-freedom dynamic model of the rolling bearing established in step S1 is used to characterize the dynamic response of the rolling bearing system, wherein the inner ring and the outer ring have two degrees of freedom respectively, and the unit resonator has only one degree of freedom.

4. The rolling bearing fault diagnosis method driven by simulation data according to claim 1, characterized in that: Step S2 specifically includes: S21, setting two acceleration sensors in the horizontal direction and the vertical direction of the bearing seat to collect the vibration signal of the rolling bearing state monitoring; S22, setting the sampling frequency and sampling time of each type; S23. The simulation signal generated by dynamics and the measured signal collected by the experiment are divided into a source domain data set and a target domain data set through a sliding window algorithm.

5. The rolling bearing fault diagnosis method driven by simulation data according to claim 4 is characterized in that: In step S22, the sampling frequency is set to 20 kHz; the sampling time of each type is set to 10 s.

6. The rolling bearing fault diagnosis method driven by simulation data according to claim 1, characterized in that: Step S3 specifically includes: S31. Introduce the source domain target classification loss function to ensure that the model accurately classifies samples from the labeled source domain data. The introduced function is as follows: Among them, x s and s Represent the samples and labels of the source domain respectively, n s Indicates the number of samples in the source domain; S32, introduce the maximum mean difference loss function to align the domain distribution boundary. The introduced function is as follows: in, and Represent the sample features of the source domain and the target domain respectively, and f(·) represents the mapping to the reproducing kernel Hilbert space through the kernel function. The feature map, n s and n t Represents the number of samples in the source domain and the target domain respectively, P(F(x) s ) and Q(F(x) t ) represent the feature distribution of source domain and target domain respectively; S33. Introduce supervised contrast loss function to promote cross-domain robust feature learning. The introduced function is as follows: Among them, i represents the anchor point, P represents the positive sample set, N represents the negative sample set, and F p and F k They represent the feature vectors of positive and negative samples respectively, and τ represents the temperature parameter to control the smoothness of the similarity distribution; S34, use data enhancement techniques such as adding Gaussian white noise, scaling the overall amplitude, and cropping to perform data enhancement to improve the generalization ability of unlabeled target domain samples; S35. Introduce an unsupervised contrast loss function to improve the clustering performance in the unlabeled target domain. The introduced function is as follows: Among them, F a Represents the features after data enhancement operation; S36. Combine the introduced source domain target classification loss function, maximum mean difference loss function, supervised contrast loss function and unsupervised contrast loss function to achieve global and instance-level domain alignment. The combined function is as follows: Among them, λ, α, and β all represent weight parameters.

7. The rolling bearing fault diagnosis method driven by simulation data according to claim 1, characterized in that: Step S4 specifically includes: S41. Use the multi-head self-attention mechanism in the Transformer model to capture the long-range dependencies of the input data; S42. Convert a set of input embeddings x into query items, keys and values ​​of d dimensions respectively. The expressions are as follows: in, Both represent trainable projection matrices; S43. Calculate the single-head scaled dot product attention by the scaling factor. The calculation formula is as follows: in, represents the scaling factor; S44. Perform H single-head attention on the multi-head self-attention mechanism. The expression is as follows: head j =Attention(Q,K,V) j ,j=1,2,...,H in, represents the trainable projection matrix, H represents the number of attention heads; S45. Connect a multi-layer perceptron after the multi-head attention mechanism to map the features to a high-dimensional space and output the maximum probability category through the Softmax function. The expression is as follows: MLP(x)=GELU(0,xW1+b1)W2+b2 Among them, W1 and W2 both represent weights, b1 and b2 both represent biases, and GELU represents the activation function.

Images