A bearing cross-domain diagnosis method and device and a storage medium

Abstract

The application discloses a bearing cross-domain diagnosis method and device and a storage medium, and belongs to the field of intelligent diagnosis of mechanical equipment. In view of the problem that the existing cross-domain diagnosis model lacks the representation ability of the internal causal relationship between fault features and categories, leading to the fault discrimination deviation under cross conditions, the application starts from the causal perspective, and through a double-branch entanglement feature learning mechanism, the vibration signal is decomposed into causal features strongly related to faults and redundant features related to domain distribution; a residual discrete wavelet enhancement module is integrated to perform frequency domain structural purification and deep decoupling on the two types of features; a causal invariance constraint is designed, the false association between the redundant features and the causal features is broken through random replacement of the redundant features, and the cross-domain invariance of the causal features is stabilized. The application realizes effective separation of the causal features and the redundant features, weakens the interference of domain distribution deviation on fault discrimination, and can meet the actual needs of reliable bearing fault diagnosis under complex and variable conditions.

Description

Technical Field

[0001] This invention relates to a cross-domain diagnostic method, device, and storage medium for bearings, belonging to the field of intelligent diagnostics for key components of mechanical equipment. Background Technology

[0002] Bearings, as critical components of rotating machinery, directly affect the stable operation and service life of equipment. In actual industrial scenarios, changes in rotational speed and load cause shifts in fault characteristic frequencies, resulting in distributional offsets in data for the same fault type. Cross-domain transfer learning, which can apply learned knowledge to new domains, is widely used in intelligent fault diagnosis. However, most existing models focus on the alignment of data distribution, lacking the ability to represent the intrinsic causal relationship between fault features and categories. When significant distributional offsets occur between training and test data due to changes in operating conditions or environmental disturbances, the model is prone to fault discrimination bias due to feature confusion, reducing diagnostic robustness. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide a bearing cross-domain diagnostic method, device and storage medium, which realizes the effective separation of causal features and redundant features, reduces the interference of domain distribution offset on fault identification, and can meet the actual needs of bearing reliable fault diagnosis under complex and variable working conditions.

[0004] To achieve the above objectives, the present invention is implemented using the following technical solution:

[0005] In a first aspect, the present invention provides a bearing cross-domain diagnostic method, comprising:

[0006] The acquired bearing vibration signal to be diagnosed is input into the trained causal constraint unentangled network model, and the bearing fault diagnosis result is output.

[0007] The causal constraint unentanglement network model is trained in the following way:

[0008] Acquire the original vibration signals of the bearing under different fault categories and operating conditions;

[0009] The original vibration signal is input into a dual-path feature extractor and separated into causal features strongly correlated with the fault category and redundant features related to the domain distribution.

[0010] The causal features and the redundancy features are respectively input into the residual discrete wavelet enhancement module, and frequency domain structured decomposition and residual enhancement are performed to output frequency domain enhanced causal features and frequency domain enhanced redundancy features.

[0011] The frequency domain enhanced redundancy features are randomly permuted to obtain the permuted redundancy feature matrix;

[0012] Based on the permuted redundancy features and the frequency domain enhanced causal features, a causal invariance constraint is applied through the decoder, and the constrained reconstructed signal is output.

[0013] Based on the reconstructed signal after constraints and the frequency domain enhanced redundancy features, a joint loss function is constructed. The model is trained and optimized through backpropagation to obtain the trained causal constraint unentangled network model.

[0014] Furthermore, the original vibration signal is input into a dual-path feature extractor, which separates it into causal features strongly correlated with the fault category and redundant features correlated with the neighborhood distribution, including:

[0015] This will include source domain signals. and target domain signal The original vibration signal is input to the causal feature extractor. Extracting causal features from the source domain causal features of the target domain The formula is as follows:

[0016] ;

[0017] in, These are the learnable parameters for the causal feature extractor;

[0018] Source domain signal and target domain signal Input noncausal feature extractor Extracting redundant features from the source domain and target domain redundancy features The formula is as follows:

[0019] ;

[0020] in, These are the learnable parameters for the non-causal feature extractor.

[0021] Furthermore, the causal features and the redundancy features are respectively input into the residual discrete wavelet enhancement module to perform frequency domain structured decomposition and residual enhancement, outputting frequency domain enhanced causal features and frequency domain enhanced redundancy features, including:

[0022] Causal characteristics Applying discrete wavelet transform, the components are decomposed into low-frequency and high-frequency components, which are then concatenated along the channel dimension to output frequency domain causal features. :

[0023] ;

[0024] in, For discrete wavelet operators; This represents the concatenation operation of tensors along the channel dimension; and These are the low-frequency and high-frequency components of the causal characteristics, respectively. Indicates batch size; Number of channels; It is the feature length;

[0025] For redundant features Applying discrete wavelet transform, the components are decomposed into low-frequency and high-frequency components, which are then concatenated along the channel dimension to output frequency domain redundancy features. :

[0026] ;

[0027] in, and These are the low-frequency and high-frequency components of the redundant feature, respectively.

[0028] The frequency domain causal features are enhanced by residual enhancement operators. Enhancement is performed to obtain frequency domain enhanced causal features. :

[0029] ;

[0030] in, This represents a one-dimensional convolution operation; It is a dimension alignment operator; For activation functions; For batch normalization; The convolution kernel weights represent the causal feature paths; This refers to the bias term for the causal feature path;

[0031] The frequency domain redundancy features are enhanced by the residual enhancement operator. Enhancement is performed to obtain frequency domain enhanced redundancy features. :

[0032] ;

[0033] in, The convolution kernel weights are for redundant feature paths; This is the bias term for redundant feature paths.

[0034] Furthermore, the frequency domain enhanced redundancy features are randomly permuted to obtain a permuted redundancy feature matrix, including:

[0035] For batch size Frequency domain enhanced redundancy features Perform random permutations to obtain the redundant feature matrix after permutation. :

[0036] ;

[0037] in, A randomly generated index sequence; It is a random permutation function; Indicates the first One redundant feature; T is the transpose.

[0038] Furthermore, based on the permuted redundancy features and the frequency domain enhanced causal features, a causal invariance constraint is applied through the decoder, and the constrained reconstructed signal is output, including:

[0039] The permuted redundant feature matrix and the frequency domain enhanced causal features Input Decoder Imposing causal invariance constraints:

[0040] For any sample that follows the same distribution and When the frequency domain enhanced causal features remain unchanged while the permutation-based redundant feature matrix changes, the following invariance relationship is satisfied:

[0041] ;

[0042] in, Indicates the given input Time Category The conditional probability of ; Representative sample Frequency domain enhancement causal features; For the sample Redundancy features after permutation;

[0043] Based on the decoder, the constrained reconstructed signal is output. , To reconstruct the signal from the source domain, This represents the target domain reconstruction signal.

[0044] Furthermore, a joint loss function is constructed based on the constrained reconstructed signal and the frequency domain enhanced redundancy features, including:

[0045] Reconstructed signal based on the constraints and the original input signal Construct reconstruction loss :

[0046] ;

[0047] in, Indicates the reconstruction loss; and They represent the first Each source domain signal and the reconstructed signal; and They represent the first Each target domain signal and reconstructed signal; Represents the L2 norm;

[0048] Based on the frequency domain enhanced causal features of the source and target domains, a maximum mean difference loss is constructed. :

[0049] ;

[0050] in, This represents the loss due to the maximum mean difference. and These represent the number of samples in the source domain and the number of samples in the target domain, respectively. Represents the kernel function; and The source domain and the target domain are respectively the first Frequency domain enhancement causal features for each sample;

[0051] Based on the frequency domain enhanced redundancy features and the frequency domain enhanced causality features, a domain discrimination loss is constructed. and classification loss ;

[0052] Construct the total loss function Taking the individual loss components and their weighting coefficients as input, the output is the total loss value used for backpropagation:

[0053] ;

[0054] in, Indicates the total loss; Indicates learnable parameters; , , , These are the weighting coefficients for each loss. Classification loss; Domain discrimination loss;

[0055] The model parameters are updated through backpropagation using the total loss value until convergence, and the trained causal constraint unentangled network model is output.

[0056] Secondly, the present invention provides a bearing cross-domain diagnostic device for implementing the bearing cross-domain diagnostic method described in any one of the foregoing claims, comprising:

[0057] The diagnostic module is used to input the acquired bearing vibration signal to be diagnosed into the trained causal constraint unentangled network model and output the bearing fault diagnosis result.

[0058] The causal constraint unentanglement network model is trained in the following way:

[0059] Acquire the original vibration signals of the bearing under different fault categories and operating conditions;

[0060] The original vibration signal is input into a dual-path feature extractor and separated into causal features strongly correlated with the fault category and redundant features related to the domain distribution.

[0061] The causal features and the redundancy features are respectively input into the residual discrete wavelet enhancement module, and frequency domain structured decomposition and residual enhancement are performed to output frequency domain enhanced causal features and frequency domain enhanced redundancy features.

[0062] The frequency domain enhanced redundancy features are randomly permuted to obtain the permuted redundancy feature matrix;

[0063] Based on the permuted redundancy features and the frequency domain enhanced causal features, a causal invariance constraint is applied through the decoder, and the constrained reconstructed signal is output.

[0064] Based on the reconstructed signal after constraints and the frequency domain enhanced redundancy features, a joint loss function is constructed. The model is trained and optimized through backpropagation to obtain the trained causal constraint unentangled network model.

[0065] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above.

[0066] Fourthly, the present invention provides an electronic device, comprising:

[0067] Memory, used to store computer programs / instructions;

[0068] A processor for executing the computer program / instructions to implement the steps of any of the methods described above.

[0069] Fifthly, the present invention provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of any of the methods described above.

[0070] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0071] This invention provides a bearing cross-domain diagnostic method, device, and storage medium. By inputting the original vibration signal into a dual-path feature extractor, it separates the original signal into causal features and redundant features, achieving directional decoupling between the essential fault features and domain-related interference. Through a residual discrete wavelet enhancement module, it performs frequency-domain structured decomposition and residual enhancement on the two types of features, constructing a frequency-domain invariant subspace for causal features and a frequency-domain difference subspace for redundant features, thus enhancing feature purification and decoupling. By randomly permuting redundant features and applying causal invariance constraints, it further solidifies the cross-domain stability of causal features, ultimately constructing a cross-domain stable causal invariant diagnostic mechanism. This effectively reduces the interference of domain distribution shift on fault identification, meeting the practical needs of reliable bearing fault diagnosis under complex and variable operating conditions. Attached Figure Description

[0072] Figure 1 This is a flowchart of a bearing cross-domain diagnostic method provided in an embodiment of the present invention;

[0073] Figure 2 These are causal feature correlation analysis diagrams provided in embodiments of the present invention; wherein, (a) is a causal feature correlation analysis diagram of 10 different samples from the same label; and (b) is a causal feature correlation analysis diagram of 10 different samples from different labels.

[0074] Figure 3 These are non-causal feature representation maps provided in this embodiment of the invention; wherein, (a) is a non-causal feature representation map of the source domain and the target domain; and (b) is a non-causal feature representation map of samples with different labels in the target domain.

[0075] Figure 4 This is a clustering effect diagram of causal and non-causal features provided in the embodiments of the present invention. Detailed Implementation

[0076] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0077] Example 1: This example introduces a bearing cross-domain diagnostic method, including:

[0078] The acquired bearing vibration signal to be diagnosed is input into the trained causal constraint unentangled network model, and the bearing fault diagnosis result is output.

[0079] The causal constraint unentanglement network model is trained in the following way:

[0080] Acquire the original vibration signals of the bearing under different fault categories and operating conditions;

[0081] The original vibration signal is input into a dual-path feature extractor and separated into causal features strongly correlated with the fault category and redundant features related to the domain distribution.

[0082] The causal features and the redundancy features are respectively input into the residual discrete wavelet enhancement module, and frequency domain structured decomposition and residual enhancement are performed to output frequency domain enhanced causal features and frequency domain enhanced redundancy features.

[0083] The frequency domain enhanced redundancy features are randomly permuted to obtain the permuted redundancy feature matrix;

[0084] Based on the permuted redundancy features and the frequency domain enhanced causal features, a causal invariance constraint is applied through the decoder, and the constrained reconstructed signal is output.

[0085] Based on the reconstructed signal after constraints and the frequency domain enhanced redundancy features, a joint loss function is constructed. The model is trained and optimized through backpropagation to obtain the trained causal constraint unentangled network model.

[0086] This embodiment uses bearing vibration signals collected from the wheelset bearing test bench at Soochow University to describe the proposed method in detail. The test bench includes an AC motor, transmission mechanism, accelerometer, and data acquisition system. The wheel-rail simulation consists of a 280mm diameter wheelset as the track support and a 200mm diameter smaller wheelset simulating a train wheel. The test bearing, an NJ208E, is mounted on one side of the smaller wheelset. An accelerometer is mounted on the bearing housing to collect vibration signals at a sampling frequency of 32768Hz. During the experiment, the rotational speeds were set to 400rpm, 600rpm, and 800rpm, and various bearing fault data were set using wire cutting, such as... Figure 1 As shown, the application process of the bearing cross-domain diagnostic method involves the following steps:

[0087] Step 1: Obtain vibration signals under different fault categories and operating conditions using an accelerometer, divide the data using a sliding window, and assign corresponding label information to the segmented signals according to different categories.

[0088] Step 2: Separate fault-related causal features and domain-related redundant interference features using a dual-path feature extractor.

[0089] Includes source domain signals and target domain signal The original vibration signal is input to the causal feature extractor. Extracting causal features from the source domain causal features of the target domain The formula is as follows:

[0090] ;

[0091] in, These are the learnable parameters for the causal feature extractor.

[0092] At the same time, the source domain signal and target domain signal Input noncausal feature extractor Extracting redundant features from the source domain and target domain redundancy features The formula is as follows:

[0093] ;

[0094] in, These are the learnable parameters for the non-causal feature extractor.

[0095] Typically, vibrational signals are observable, while shared factors are unobservable. Therefore, causal characteristics are not learned directly by reconstructing causal factors, but rather by forcing causal manifestations to possess the same properties as the causal factors. For example... Figure 2 As shown, the correlation analysis of causal features indicates that samples from different labels have high correlation coefficients, while samples from different labels have low correlation coefficients, suggesting that the causal representations have high consistency and independence. Figure 3 As shown, the source and target domains in the non-causal representations exhibit significant differences, while the non-causal representations of different fault categories within the same domain maintain a high degree of consistency, verifying the effective separation of causal features and redundant features.

[0096] Step 3: To further enhance the discriminative properties and decoupling of the features, residual discrete wavelet transform is performed on the shared features and private features output by the feature extractor, respectively, to achieve frequency domain structured decomposition and construct a decoupled feature subspace.

[0097] Causal characteristics Applying discrete wavelet transform, the components are decomposed into low-frequency and high-frequency components, which are then concatenated along the channel dimension to output frequency domain causal features. :

[0098] ;

[0099] in, For discrete wavelet operators; This represents the concatenation operation of tensors along the channel dimension; and These are the low-frequency and high-frequency components of the causal characteristics, respectively. Indicates batch size; Number of channels; It is the feature length.

[0100] For redundant features Applying discrete wavelet transform, the components are decomposed into low-frequency and high-frequency components, which are then concatenated along the channel dimension to output frequency domain redundancy features. :

[0101] ;

[0102] in, and These are the low-frequency and high-frequency components of the redundant features, respectively.

[0103] To avoid attenuation and distortion of feature information during frequency domain separation, residual enhancement operators are applied to the two types of separated frequency domain features. These residual enhancement operators are then used to enhance the frequency domain causal features. Enhancement is performed to obtain frequency domain enhanced causal features. :

[0104] ;

[0105] in, This represents a one-dimensional convolution operation; It is a dimension alignment operator; For activation functions; For batch normalization; The convolution kernel weights represent the causal feature paths; This is the bias term for the causal feature path.

[0106] The frequency domain redundancy features are enhanced by the residual enhancement operator. Enhancement is performed to obtain frequency domain enhanced redundancy features. :

[0107] ;

[0108] in, The convolution kernel weights are for redundant feature paths; This is the bias term for redundant feature paths.

[0109] Step 4: Design causal invariance constraints and break the false associations between features through a redundant feature random replacement strategy.

[0110] For batch size Frequency domain enhanced redundancy features Perform random permutations to obtain the redundant feature matrix after permutation. :

[0111] ;

[0112] in, A randomly generated index sequence; It is a random permutation function; Indicates the first One redundant feature; T is the transpose.

[0113] Step 5: Based on the permuted redundant feature matrix and the frequency domain enhanced causal features, apply causal invariance constraints through the decoder.

[0114] The permuted redundant feature matrix and the frequency domain enhanced causal features Input Decoder Imposing causal invariance constraints:

[0115] For any sample that follows the same distribution and When the frequency domain enhanced causal features remain unchanged while the permutation-based redundant feature matrix changes, the following invariance relationship is satisfied:

[0116] ;

[0117] in, Indicates the given input Time Category The conditional probability of ; Representative sample Frequency domain enhancement causal features; For the sample Redundancy features after permutation;

[0118] Based on the decoder, the constrained reconstructed signal is output. , To reconstruct the signal from the source domain, This represents the reconstructed signal in the target domain. This constraint forces the model to rely solely on causal features to restore the essence of the signal, breaking the false correlation between redundant features and causal features.

[0119] Step 6: Construct a deep learning loss function and train and optimize the model using backpropagation.

[0120] Reconstructed signal based on the constraints and the original input signal Construct reconstruction loss :

[0121] ;

[0122] in, Indicates the reconstruction loss; and They represent the first Each source domain signal and the reconstructed signal; and They represent the first Each target domain signal and reconstructed signal; Represents the L2 norm;

[0123] Based on the frequency domain enhanced causal features of the source and target domains, a maximum mean difference loss is constructed. :

[0124] ;

[0125] in, This represents the loss due to the maximum mean difference. and These represent the number of samples in the source domain and the number of samples in the target domain, respectively. Represents the kernel function; and The source domain and the target domain are respectively the first Frequency domain enhancement causal features for each sample;

[0126] Based on the frequency domain enhanced redundancy features and the frequency domain enhanced causality features, a domain discrimination loss is constructed. and classification loss ;

[0127] Construct the total loss function Taking the individual loss components and their weighting coefficients as input, the output is the total loss value used for backpropagation:

[0128] ;

[0129] in, Indicates the total loss; Indicates learnable parameters; , , , These are the weighting coefficients for each loss. Classification loss; Domain discrimination loss;

[0130] The model parameters are updated through backpropagation using the total loss value until convergence, and the trained causal constraint unentangled network model is output.

[0131] Step 7: Verify the diagnostic performance of the proposed method. Six cross-domain diagnostic tasks were designed, and domain adaptation methods were selected as comparison models, including Domain Adversarial Neural Network (DANN), Wide Convolutional Network (WDCNN), and Deep Correlation Alignment Network (DCORAL). The accuracy comparison results are shown in Table 1. The proposed method outperforms the comparison methods in all different tasks, maintaining a high fault diagnosis accuracy.

[0132] Table 1. Accuracy Comparison of Different Methods

[0133]

[0134] Ablation experiments were conducted, with four ablation variants: A1 containing only fault category classification loss, A2 containing both fault category classification loss and domain classification loss, A3 adding maximum mean difference loss to A2, and A4 introducing reconstruction loss to A2. The ablation experiment results are shown in Table 2. The proposed causal constraint unentanglement network achieved optimal diagnostic performance.

[0135] Table 2. Accuracy Comparison of Different Ablation Variants

[0136]

[0137] like Figure 4 As shown in the clustering results of causal and non-causal features, causal features of the same fault category are highly aggregated in the low-dimensional space, the class boundaries between different categories are clear, and domain-related information forms independently distributed clusters in the space, further verifying the effective decoupling capability of the model.

[0138] Step 8: Input the acquired bearing vibration signal to be diagnosed into the causal constraint unentangled network model trained in steps 1 to 6, and output the bearing fault diagnosis result. In this embodiment, the bearing vibration signal to be diagnosed is input into the model, and the model outputs the corresponding fault category label, realizing accurate diagnosis of bearing faults across operating conditions.

[0139] This embodiment separates the original vibration signal into causal features and redundant features by inputting it into a dual-path feature extractor, thereby achieving directional decoupling between the essential fault features and domain-related interference. A residual discrete wavelet enhancement module performs frequency-domain structured decomposition and residual enhancement on the two types of features, constructing a frequency-domain invariant subspace for causal features and a frequency-domain difference subspace for redundant features, enhancing feature purification and decoupling. By randomly permuting redundant features and applying causal invariance constraints, the cross-domain stability of causal features is further solidified, ultimately constructing a cross-domain stable causal invariant diagnostic mechanism. This effectively weakens the interference of domain distribution shift on fault identification, meeting the practical needs of reliable bearing fault diagnosis under complex and variable operating conditions.

[0140] Example 2: This example provides a bearing cross-domain diagnostic device, including:

[0141] The diagnostic module is used to input the acquired bearing vibration signal to be diagnosed into the trained causal constraint unentangled network model and output the bearing fault diagnosis result.

[0142] The causal constraint unentanglement network model is trained in the following way:

[0143] Acquire the original vibration signals of the bearing under different fault categories and operating conditions;

[0144] The original vibration signal is input into a dual-path feature extractor and separated into causal features strongly correlated with the fault category and redundant features related to the domain distribution.

[0145] The causal features and the redundancy features are respectively input into the residual discrete wavelet enhancement module, and frequency domain structured decomposition and residual enhancement are performed to output frequency domain enhanced causal features and frequency domain enhanced redundancy features.

[0146] The frequency domain enhanced redundancy features are randomly permuted to obtain the permuted redundancy feature matrix;

[0147] Based on the permuted redundancy features and the frequency domain enhanced causal features, a causal invariance constraint is applied through the decoder, and the constrained reconstructed signal is output.

[0148] Based on the reconstructed signal after constraints and the frequency domain enhanced redundancy features, a joint loss function is constructed. The model is trained and optimized through backpropagation to obtain the trained causal constraint unentangled network model.

[0149] The specific functions of each module described above are explained in the relevant content of the method in Embodiment 1, and will not be repeated here.

[0150] Example 3: This example provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described in Example 1.

[0151] Example 4: This example provides an electronic device, including:

[0152] Memory, used to store computer programs / instructions;

[0153] A processor for executing the computer program / instructions to implement the steps of any of the methods described in Embodiment 1.

[0154] Example 5: This example provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the method described in any one of Examples 1.

[0155] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

[0156] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0157] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0158] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0159] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0160] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this disclosure and not to limit its protection scope. Although this disclosure has been described in detail with reference to the above embodiments, those skilled in the art should understand that after reading this disclosure, they can still make various changes, modifications or equivalent substitutions to the specific implementation of the invention, but these changes, modifications or equivalent substitutions are all within the protection scope of the pending claims.

Claims

1. A method for cross-domain diagnosis of bearings, characterized in that, include: The acquired bearing vibration signal to be diagnosed is input into the trained causal constraint unentangled network model, and the bearing fault diagnosis result is output. The causal constraint unentanglement network model is trained in the following way: Acquire the original vibration signals of the bearing under different fault categories and operating conditions; The original vibration signal is input into a dual-path feature extractor and separated into causal features strongly correlated with the fault category and redundant features related to the domain distribution. The causal features and the redundancy features are respectively input into the residual discrete wavelet enhancement module, and frequency domain structured decomposition and residual enhancement are performed to output frequency domain enhanced causal features and frequency domain enhanced redundancy features. The frequency domain enhanced redundancy features are randomly permuted to obtain the permuted redundancy feature matrix; Based on the permuted redundancy features and the frequency domain enhanced causal features, a causal invariance constraint is applied through the decoder, and the constrained reconstructed signal is output. Based on the reconstructed signal after constraints and the frequency domain enhanced redundancy features, a joint loss function is constructed. The model is trained and optimized through backpropagation to obtain the trained causal constraint unentangled network model.

2. The bearing cross-domain diagnostic method according to claim 1, characterized in that, The original vibration signal is input into a dual-path feature extractor, which separates it into causal features strongly correlated with the fault category and redundant features correlated with the neighborhood distribution, including: Includes source domain signals and target domain signal The original vibration signal is input to the causal feature extractor. Extracting causal features from the source domain causal features of the target domain The formula is as follows: ； in, These are the learnable parameters for the causal feature extractor; Source domain signal and target domain signal Input noncausal feature extractor Extracting redundant features from the source domain and target domain redundancy features The formula is as follows: ； in, These are the learnable parameters for the non-causal feature extractor.

3. The bearing cross-domain diagnostic method according to claim 2, characterized in that, The causal features and the redundancy features are respectively input into the residual discrete wavelet enhancement module, and frequency domain structured decomposition and residual enhancement are performed to output frequency domain enhanced causal features and frequency domain enhanced redundancy features, including: Causal characteristics Applying discrete wavelet transform, the components are decomposed into low-frequency and high-frequency components, which are then concatenated along the channel dimension to output frequency domain causal features. : ； in, For discrete wavelet operators; This represents the concatenation operation of tensors along the channel dimension; and These are the low-frequency and high-frequency components of the causal characteristics, respectively. Indicates batch size; Number of channels; It is the feature length; For redundant features Applying discrete wavelet transform, the components are decomposed into low-frequency and high-frequency components, which are then concatenated along the channel dimension to output frequency domain redundancy features. : ； in, and These are the low-frequency and high-frequency components of the redundant feature, respectively. The frequency domain causal features are enhanced by residual enhancement operators. Enhancement is performed to obtain frequency domain enhanced causal features. : ； in, This represents a one-dimensional convolution operation; It is a dimension alignment operator; For activation functions; For batch normalization; The convolution kernel weights represent the causal feature paths; This refers to the bias term for the causal feature path; The frequency domain redundancy features are enhanced by the residual enhancement operator. Enhancement is performed to obtain frequency domain enhanced redundancy features. : ； in, The convolution kernel weights are for redundant feature paths; This is the bias term for redundant feature paths.

4. The bearing cross-domain diagnostic method according to claim 3, characterized in that, The frequency domain enhanced redundancy features are randomly permuted to obtain a permuted redundancy feature matrix, including: For batch size Frequency domain enhanced redundancy features Perform random permutations to obtain the redundant feature matrix after permutation. : ； in, A randomly generated index sequence; It is a random permutation function; Indicates the first One redundant feature; T is the transpose.

5. The bearing cross-domain diagnostic method according to claim 4, characterized in that, Based on the permuted redundancy features and the frequency domain enhanced causal features, a causal invariance constraint is applied through the decoder, and the constrained reconstructed signal is output, including: The permuted redundant feature matrix and the frequency domain enhanced causal features Input Decoder Imposing causal invariance constraints: For any sample that follows the same distribution and When the frequency domain enhanced causal features remain unchanged while the permutation-based redundant feature matrix changes, the following invariance relationship is satisfied: ； in, Indicates the given input Time Category The conditional probability of ; Representative sample Frequency domain enhancement causal features; For the sample Redundancy features after permutation; Based on the decoder, the constrained reconstructed signal is output. , To reconstruct the signal from the source domain, This represents the target domain reconstruction signal.

6. The bearing cross-domain diagnostic method according to claim 5, characterized in that, A joint loss function is constructed based on the constrained reconstructed signal and the frequency domain enhanced redundancy features, including: Reconstructed signal based on the constraints and the original input signal Construct reconstruction loss : ； in, Indicates the reconstruction loss; and They represent the first Each source domain signal and the reconstructed signal; and They represent the first Each target domain signal and reconstructed signal; Represents the L2 norm; Based on the frequency domain enhanced causal features of the source and target domains, a maximum mean difference loss is constructed. : ； in, This represents the loss due to the maximum mean difference. and These represent the number of samples in the source domain and the number of samples in the target domain, respectively. Represents the kernel function; and The source domain and the target domain are respectively the first Frequency domain enhancement causal features for each sample; Based on the frequency domain enhanced redundancy features and the frequency domain enhanced causality features, a domain discrimination loss is constructed. and classification loss ; Construct the total loss function Taking the individual loss components and their weighting coefficients as input, the output is the total loss value used for backpropagation: ； in, Indicates the total loss; Indicates learnable parameters; , , , These are the weighting coefficients for each loss. Classification loss; Domain discrimination loss; The model parameters are updated through backpropagation using the total loss value until convergence, and the trained causal constraint unentangled network model is output.

7. A bearing cross-domain diagnostic device, used to implement the bearing cross-domain diagnostic method according to any one of claims 1-6, characterized in that, include: The diagnostic module is used to input the acquired bearing vibration signal to be diagnosed into the trained causal constraint unentangled network model and output the bearing fault diagnosis result. The causal constraint unentanglement network model is trained in the following way: Acquire the original vibration signals of the bearing under different fault categories and operating conditions; The original vibration signal is input into a dual-path feature extractor and separated into causal features strongly correlated with the fault category and redundant features related to the domain distribution. The causal features and the redundancy features are respectively input into the residual discrete wavelet enhancement module, and frequency domain structured decomposition and residual enhancement are performed to output frequency domain enhanced causal features and frequency domain enhanced redundancy features. The frequency domain enhanced redundancy features are randomly permuted to obtain the permuted redundancy feature matrix; Based on the permuted redundancy features and the frequency domain enhanced causal features, a causal invariance constraint is applied through the decoder, and the constrained reconstructed signal is output. Based on the reconstructed signal after constraints and the frequency domain enhanced redundancy features, a joint loss function is constructed. The model is trained and optimized through backpropagation to obtain the trained causal constraint unentangled network model.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When executed by a processor, the computer program implements the steps of the method described in any one of claims 1-6.

9. An electronic device, characterized in that, include: Memory, used to store computer programs / instructions; A processor for executing the computer program / instructions to implement the steps of the method according to any one of claims 1-6.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method described in any one of claims 1-6.

Images