A rolling bearing fault diagnosis method for discriminating information reverse guide input optimization

Abstract

This invention relates to a rolling bearing fault diagnosis method guided by discriminant information in input optimization. It includes: firstly, introducing a non-negative matrix factorization model with discriminant constraints at the input layer to perform structured modeling of the original vibration signal, achieving signal reconstruction and discriminant information fusion, thereby enhancing the expression of fault-related features while suppressing noise interference; secondly, constructing an input optimization mechanism guided by discriminant information in reverse during model training, using a loss function to controllably adjust the gradient information of the input signal, enabling the model to adaptively strengthen key regions that contribute significantly to the classification results; finally, combining a multi-scale convolutional structure with a frequency-aware feature fusion strategy to effectively represent the optimized signal, achieving decoupled expression and stable modeling of fault features across different frequency bands. This invention can effectively suppress noise interference while retaining key fault information; and it can maintain stable feature expression and high classification accuracy under multiple operating conditions.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology, and more specifically to a rolling bearing fault diagnosis method that optimizes input by back-guided discrimination information. Background Technology

[0002] With the ongoing transformation of the global energy structure, the importance of clean energy is becoming increasingly prominent. Wind energy, as a mature and stable green energy source, occupies an important position in the global energy system, and its safe and efficient operation is crucial for achieving low-carbon economic goals. Wind turbines operate in complex outdoor environments for extended periods, and the reliability of their transmission systems directly affects the stable output of the entire wind farm. As core components of the transmission system, rolling bearings and gearboxes not only bear the responsibility of energy transmission but are also most susceptible to factors such as variable loads, strong noise, and fluctuations in operating conditions, thus becoming the primary locations where wind turbine failures are most frequent.

[0003] However, wind turbines operate in a complex environment, with their operating conditions exhibiting significant dynamic variations depending on wind speed, load, and season. Bearing vibration signals often display strong nonlinear and non-stationary characteristics. In situations where strong noise backgrounds coexist with weak impact characteristics, effective fault features are often difficult to extract accurately, posing significant challenges to fault diagnosis.

[0004] Currently, in fault diagnosis research, time-frequency analysis methods are often used to process non-stationary vibration signals, such as short-time Fourier transform, empirical mode decomposition, variational mode decomposition, and continuous wavelet transform. Although these methods provide effective means for extracting bearing fault features, there is still room for improvement in their ability to capture transient impacts under conditions of strong noise, variable operating conditions, and weak features.

[0005] As data volumes continue to expand, deep learning methods are gradually becoming the mainstream technology for fault diagnosis. They can automatically learn high-level features from raw signals, reducing reliance on manual feature engineering and providing more reliable technical support for the safe and stable operation of wind farms. However, existing deep learning methods often ignore the differences in physical mechanisms corresponding to different frequency bands, and generally adopt a globally uniform feature aggregation strategy. This leads to coupling and information dilution of high and low frequency features in deep space, thus limiting the model's generalization ability and physical interpretability under complex operating conditions. Meanwhile, in feature extraction and structure enhancement, data-driven methods are currently widely used. While these methods can achieve good results under low speed and stable operating conditions, they are prone to weakening or even filtering out weak key fault features while suppressing background noise.

[0006] Therefore, how to overcome the above-mentioned shortcomings and achieve coordinated optimization of signal preprocessing and feature learning is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] In view of the above problems, the present invention is proposed to provide a rolling bearing fault diagnosis method that optimizes the input of discriminant information back-guided input to overcome or at least partially solve the above problems.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] In a first aspect, embodiments of the present invention provide a rolling bearing fault diagnosis method with reverse-guided input optimization based on discrimination information, the steps of which include: Based on the vibration signal under normal working conditions, a dictionary matrix is ​​constructed by introducing discriminant constraints; the input signal to be diagnosed is reconstructed using the dictionary matrix to obtain the reconstructed signal; The reconstructed signal is input into the rolling bearing fault diagnosis model to obtain the fault type; The rolling bearing fault diagnosis model is used for feature extraction, and adaptive aggregation of the extracted features based on channel dimension through frequency perception, and identification of fault type based on the aggregated features. During the training of the rolling bearing fault diagnosis model, the reconstructed signal is adaptively adjusted based on the gradient information of the fault classification loss function, and the model parameters are optimized based on the adjusted signal.

[0010] Preferably, based on the vibration signal under normal working conditions, a dictionary matrix is ​​constructed by introducing discriminant constraints, including: constructing the following objective function based on the discriminant constraints, and obtaining the dictionary matrix by iterative solution;

[0011] In the formula, , These represent vibration signals under normal operating conditions. The initial dictionary matrix and initial sparse coefficient matrix are obtained through nonnegative matrix decomposition. Describe the objective function. This refers to signal reconstruction error, used to ensure the physical reconstruction accuracy of the signal. for Norm sparse constraints are used to suppress noise and highlight key fault features in the signal. The sparse regularization coefficient; For classification loss function, For real labels, Represents a classification network mapping. The constraint weights are used to balance reconstruction accuracy and discrimination performance.

[0012] Preferably, during iterative solving, an alternating optimization strategy is adopted, including: in each iteration, first fixing the initial dictionary matrix and updating the initial sparse coefficient matrix, then fixing the updated sparse coefficient matrix to update the initial dictionary matrix, wherein the update method is as follows:

[0013] .

[0014] Preferably, the input signal to be diagnosed is reconstructed using the dictionary matrix to obtain a reconstructed signal, including:

[0015] In the formula, Let be the sparse coefficient vector corresponding to the input signal to be diagnosed. This represents the updated dictionary matrix.

[0016] Preferably, feature extraction includes extracting multi-scale features using convolutional kernels of different scales and concatenating the multi-scale features.

[0017] Preferably, adaptive aggregation of the extracted features based on channel dimension through frequency awareness includes: The amplitude spectrum is obtained by performing a Fast Fourier Transform on the channel features of the extracted features. ; Calculate the dominant frequency of the amplitude spectrum. ; With rotating fundamental frequency As a benchmark, a dimensionless frequency boundary parameter is introduced. and Define frequency gating coefficients ,in, This represents a gating function based on the dominant frequency position; The frequency gating coefficients are used to weight and aggregate the features of each channel to obtain aggregated features.

[0018] Preferably, the reconstructed signal is adaptively adjusted based on the gradient information of the fault classification loss function, including: The gradient of the reconstructed signal with respect to the fault classification loss function is calculated in reverse. The reconstructed signal is adaptively enhanced based on the gradient, as expressed by the following formula:

[0019]

[0020] In the formula, Indicates the reconstructed signal. Indicates an optimized signal. Represents the gradient. Indicates the adaptive dynamic enhancement step size. Indicates the maximum adjustment range. Indicates a length of The local standard deviation calculated within the sliding window. This represents the maximum local standard deviation within the global scope.

[0021] Secondly, embodiments of the present invention provide a rolling bearing fault diagnosis system with discriminant information reverse-guided input optimization, used to implement a rolling bearing fault diagnosis method with discriminant information reverse-guided input optimization as described in any of the preceding claims, comprising: The input signal reconstruction module is used to construct a dictionary matrix based on the vibration signal under normal operating conditions by introducing discrimination constraints; and to reconstruct the input signal to be diagnosed using the dictionary matrix to obtain the reconstructed signal. The fault type identification module has a built-in rolling bearing fault diagnosis model, which is used to identify the fault type based on the reconstructed signal; The model optimization training module is used to adaptively adjust the reconstructed signal based on the gradient information of the fault classification loss function, and optimize the model parameters according to the adjusted signal.

[0022] Thirdly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, it implements a rolling bearing fault diagnosis method with reverse guidance input optimization based on discrimination information as described in any of the preceding claims.

[0023] Fourthly, embodiments of the present invention provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements a rolling bearing fault diagnosis method with reverse guidance input optimization based on discrimination information as described in any of the preceding claims.

[0024] This invention provides a rolling bearing fault diagnosis method that uses discriminative information to guide input optimization. The method aims to apply the discriminative information of a classification model back to the input space to achieve synergistic optimization of signal preprocessing and feature learning, thereby improving the separability and noise resistance of fault features at the input layer. Compared with existing technologies, the beneficial effects of the above-mentioned technical solution provided by this invention include at least the following: 1. Construct a non-negative matrix factorization denoising model with discriminative constraints. Introduce classification loss constraints into the non-negative matrix factorization framework to achieve joint optimization of signal reconstruction and discriminative information. Enhance fault-sensitive features and suppress noise interference without the need for fault prior knowledge.

[0025] 2. A discriminant information-guided input optimization strategy is proposed. Based on the gradient response of the loss function to the input signal, the input is adjusted in a controlled manner, enabling the model to adaptively strengthen the local areas that play a key role in fault identification.

[0026] 3. Design a multi-scale and frequency-aware feature aggregation method. Extract features at different time scales through multi-scale convolution and adaptively adjust the feature aggregation method based on frequency information to achieve decoupling and effective fusion of features of different frequency bands in broadband signals.

[0027] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0028] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0030] Figure 1 This is a schematic diagram of the overall structural framework of the rolling bearing fault diagnosis method optimized by reverse guidance of discrimination information according to the present invention. Figure 2 This is a diagram illustrating the overall framework for feature extraction and fusion provided in this embodiment of the invention. Figure 3 This is a structural diagram of a motor-driven mechanical system using SEU provided in an embodiment of the present invention. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Most existing rolling bearing fault diagnosis methods rely on feature learning from raw or simply preprocessed signals, lacking explicit modeling of fault discrimination information in the input signals. This leads to key features being easily submerged by noise under complex operating conditions, thus affecting diagnostic performance.

[0033] Accordingly, this invention discloses a rolling bearing fault diagnosis method that uses discriminant information to guide input optimization, addressing the problems of strong noise interference and insufficient feature representation in the fault diagnosis of wind turbine gearboxes and rolling bearings. The aim is to achieve coordinated optimization of signal preprocessing and feature learning by feeding back the discriminant information of the classification model to the input space in the form of gradients from the perspective of optimizing the feature quality of the input layer.

[0034] In one embodiment, the rolling bearing fault diagnosis method that uses discriminant information to guide input optimization includes the following steps: Based on the vibration signal under normal working conditions, a dictionary matrix is ​​constructed by introducing discriminant constraints; the input signal to be diagnosed is reconstructed using the dictionary matrix to obtain the reconstructed signal; The reconstructed signal is input into the rolling bearing fault diagnosis model to obtain the fault type; The rolling bearing fault diagnosis model is used for feature extraction, and adaptive aggregation of the extracted features based on channel dimension through frequency perception, and identification of fault type based on the aggregated features. During the training of the rolling bearing fault diagnosis model, the reconstructed signal is adaptively adjusted based on the gradient information of the fault classification loss function, and the model parameters are optimized based on the adjusted signal.

[0035] In this embodiment, the above method comprises three stages: First, in the sparse denoising stage, a dictionary model is constructed using vibration signals under normal operating conditions, and discrimination constraints are designed to enhance fault-related components during signal reconstruction, thereby achieving noise suppression and retention of fault-sensitive features; Second, by combining multi-scale convolutional modeling and frequency prior constraints, the reconstructed signal is characterized, and the effective fusion of features in different frequency bands is achieved through frequency-aware adaptive feature aggregation, ultimately completing fault type identification.

[0036] Simultaneously, during the input optimization phase, based on the gradient response of the loss function to the reconstructed signal, the input is adaptively adjusted under constraints, explicitly applying discriminative information to the input space, thereby guiding the model to focus on local regions that play a crucial role in fault identification; the overall structural framework diagram is as follows. Figure 1 As shown.

[0037] In one alternative implementation, the vibration signals of wind turbine generators are typically affected by strong background noise and operating disturbances, resulting in fault characteristics exhibiting weak impact and local non-stationarity. Traditional denoising methods often aim to minimize signal reconstruction errors without explicitly considering fault discrimination information, which can easily weaken key fault features during denoising and thus affect subsequent diagnostic performance.

[0038] To this end, this embodiment designs discriminative constraints within the Non-negative Matrix Factorization (NMF) framework, integrating classification and discriminative information into the sparse decomposition process to achieve joint optimization of signal reconstruction and discriminative feature enhancement, thereby suppressing noise while retaining effective fault-related components.

[0039] For example, suppose the vibration signal under normal operating conditions is a matrix. It is represented as a dictionary matrix through nonnegative matrix decomposition. and sparse coefficient matrix :

[0040] In the formula, This is the initial dictionary matrix, used to characterize the fundamental vibration modes under the healthy state of the bearing; The initial sparse coefficient matrix satisfies the nonnegativity constraint. This decomposition ensures the physical non-negativity of the vibration signal amplitude and allows the signal to be represented by a linear combination of a small number of key dictionary atoms.

[0041] To improve the fault sensitivity of the decomposition results, a discriminative constraint is introduced based on the traditional reconstruction objective, and the following joint optimization objective function is constructed:

[0042] In the formula, , These represent vibration signals under normal operating conditions. The initial dictionary matrix and initial sparse coefficient matrix are obtained through nonnegative matrix decomposition. Describe the objective function. This refers to signal reconstruction error, used to ensure the physical reconstruction accuracy of the signal. for Norm sparse constraints are used to suppress noise and highlight key fault features in the signal. The sparse regularization coefficient; For classification loss function, For real labels, Represents a classification network mapping. The constraint weights are used to balance reconstruction accuracy and discrimination performance.

[0043] Furthermore, to address the aforementioned optimization problem, this embodiment employs an alternating optimization strategy. In each iteration, the initial dictionary matrix is ​​first fixed. Update the initial sparse coefficient matrix :

[0044] Then, fix the updated sparse coefficient matrix and update the initial dictionary matrix until convergence:

[0045] After training, the updated dictionary matrix is ​​used to reconstruct the input signal to be diagnosed, resulting in the reconstructed signal.

[0046] In the formula, Let be the sparse coefficient vector corresponding to the input signal to be diagnosed. This representation preserves the fault characteristics corresponding to high response coefficients while suppressing low response noise components, thereby obtaining a signal with discriminative enhancement characteristics. This represents the updated dictionary matrix.

[0047] As a preferred implementation, the reconstructed signal is finally normalized and segmented into segments with a fixed window length of 1024 to construct standardized input samples, providing a high-quality data foundation for the subsequent input optimization stage.

[0048] In an alternative implementation, to further highlight the discriminative region that plays a crucial role in fault identification, this embodiment introduces model discriminative information to controllably adjust the input signal, thereby improving the separability of the input layer features. Specific steps include: First, based on the reconstructed signal The classification loss function is calculated through backpropagation. gradient:

[0049] The gradient reflects the sensitivity of each position in the input signal to the classification result. A larger gradient magnitude indicates a more significant contribution of that position to the classification decision, and therefore should be strengthened during input optimization. To avoid excessive perturbation or under-optimization caused by fixed-step updates, this application introduces local statistical properties to construct an adaptive adjustment mechanism, dynamically constraining the input adjustment magnitude. The optimized input signal is expressed as:

[0050]

[0051] In the formula, Indicates the reconstructed signal. Indicates an optimized signal. Represents the gradient. Indicates the adaptive dynamic enhancement step size. Indicates the maximum adjustment range. Indicates a length of The local standard deviation calculated within the sliding window is used to characterize the local fluctuation intensity of the signal. This represents the maximum local standard deviation over a global range, when the signal is relatively stable in a local area. When the amplitude is small, the adjustment amplitude is suppressed, thus avoiding excessive amplification of the noise region; while in the region with significant impact characteristics and high gradient response, the adjustment step size will be adaptively increased, thereby enhancing the key characteristic response related to the fault.

[0052] During model training, this input optimization operation is performed only once before each forward propagation round and does not participate in the long-term cumulative update of samples, thus ensuring the stability of the training process. During the testing phase, no input optimization is performed; the reconstructed signal is directly used as the model input to improve the generalization ability of the method.

[0053] Based on the input optimization guided by discriminative information, the input signal has been enhanced in the key discrimination region, but the fault characteristics still exhibit significant non-uniformity in terms of time scale and frequency distribution. Different frequency bands correspond to different physical mechanisms: low-frequency components mainly reflect the periodic motion characteristics of the rotor system, while mid-to-high-frequency components include the impact response and harmonic structure excited by local defects. If features are aggregated indiscriminately, it can easily lead to the coupling of information from different frequency bands, thereby affecting the discriminative power of the feature representation.

[0054] In one alternative implementation, this application designs a multi-scale feature extraction and frequency-aware fusion method to achieve the synergistic effect of temporal multi-scale modeling and frequency-domain structural constraints.

[0055] For example, suppose the feature extraction network is Then the reconstructed signal The features are represented as follows:

[0056] In the formula, , and These represent the number of feature channels and the feature length, respectively.

[0057] In this embodiment, to fully characterize fault features at different time scales, a parallel one-dimensional convolutional branch structure is constructed. The input signal is mapped using convolutional kernels of different scales, and the data is then concatenated to obtain a multi-scale feature representation. Small-scale convolutional kernels are used to capture high-frequency transient impact features, medium-scale kernels are used to model periodic vibration information, and large-scale kernels are used to characterize low-frequency variation trends. Through parallel modeling of multi-scale branches, multi-level representation of broadband signals is achieved.

[0058] Figure 2 This is a diagram illustrating the overall framework for feature extraction and fusion. This embodiment obtains the feature tensor... Then, frequency prior information is introduced to adaptively modulate the features of different channels. This includes modulating the features of each channel. Perform a fast Fourier transform to obtain its amplitude spectrum. And calculate its dominant frequency. .

[0059] Then using the rotating fundamental frequency As a benchmark, a dimensionless frequency boundary parameter is introduced. and Define frequency gating coefficients to divide the frequency domain into different functional regions:

[0060] In the formula, This represents a gating function based on the dominant frequency location, used to characterize the importance weight of the channel's features in different frequency bands. The frequency gating function... The following parameterized double sigmoid gating function is used to achieve soft selection for a specific frequency band:

[0061] In the formula, This is a smoothing factor used to control the steepness of the gating boundary. and The frequency domain interval boundaries of interest for each feature channel are defined, and the dominant frequency of a certain channel feature is... fall into When the interval is defined, the function outputs weights close to 1, preserving this feature. When When the frequency deviates from this range, the weight rapidly decays to 0, thereby suppressing interference from irrelevant frequency bands.

[0062] Finally, the frequency gating coefficients are used to weight and aggregate the features of each channel to obtain the aggregated features, which are expressed as follows:

[0063] This mechanism enables the feature fusion process to adaptively adjust based on the frequency domain properties of the channels, thereby suppressing coupling interference between features in different frequency bands and enhancing key frequency components related to the fault mechanism. Ultimately, the features of each channel are... splicing to form a global representation And input it into the classifier to complete the fault type identification.

[0064] To verify the effectiveness of the proposed method, a case-based study was conducted in one embodiment. The main parameters of the fault diagnosis model are shown in Table 1. Table 1

[0065] To verify the diagnostic performance of the proposed method, a series of experiments were conducted, comparing it with five other methods. The comparison methods included Swin-T, LMSCNN, CWT-CNN-SSA-ELM, 1DCNN-PSO-SVM, and DMAAN.

[0066] The experiment used the gearbox dataset from Southeast University for verification. The test bench was as follows: Figure 3 As shown, the system consists of a motor, motor controller, reduction gearbox, planetary gearbox, and load controller. The dataset includes gear and rolling bearing datasets, obtained by adjusting operating conditions to 20 Hz -0 V and 30 Hz -2 V, respectively, and each dataset contains five fault states. This experiment uses datasets under 20 Hz -0 V and 30 Hz -2 V conditions to verify the diagnostic performance of the proposed method. Specific dataset descriptions are shown in Table 2. Each fault class has 100 samples, which are divided into training and test sets in a 7:3 ratio. The experimental batch size is 100.

[0067] Table 2

[0068] To comprehensively evaluate the classification performance of the proposed method, this application selects accuracy, precision, recall, and F1-score as evaluation metrics to compare the overall performance of different methods on the Southeast University Gearbox dataset. The experimental results are shown in Table 3.

[0069] Table 3

[0070] As shown in Table 3, the proposed method outperforms the comparative methods in all evaluation metrics, demonstrating more stable and balanced classification capabilities. Specifically, the proposed method achieves 97.84% accuracy, a 3.61% improvement over the second-best method, DMAAN (94.23%), demonstrating a significant advantage in overall classification performance. In terms of precision and recall, the proposed method achieves 97.52% and 97.63% respectively, indicating that the model not only effectively reduces false positives but also has high recognition capabilities for various fault samples. The corresponding F1-score reaches 97.57%, further demonstrating that the model achieves a good balance between precision and recall. In contrast, traditional deep learning methods such as Swin-T and LMSCNN exhibit some performance degradation under complex conditions, mainly because they rely on the original input signal for feature learning and lack explicit modeling of input layer discriminative information. While methods such as 1DCNN-PSO-SVM and DMAAN improve performance to some extent, they still have shortcomings under conditions of strong noise interference and weak fault features. The proposed method constructs an input optimization mechanism guided by discriminative information to achieve noise suppression and key feature enhancement at the input layer. It also combines multi-scale and frequency-aware feature fusion strategies to achieve optimal results in all metrics.

[0071] Based on the same inventive concept, embodiments of the present invention also provide a rolling bearing fault diagnosis system with reverse guidance of discrimination information for input optimization, comprising: The input signal reconstruction module is used to construct a dictionary matrix based on the vibration signal under normal operating conditions by introducing discrimination constraints; and to reconstruct the input signal to be diagnosed using the dictionary matrix to obtain the reconstructed signal. The input signal optimization module is used to adaptively enhance the reconstructed signal using a fault classification loss function to obtain an optimized signal. The signal feature aggregation module is used to extract features from the optimized signal and adaptively aggregate the features of each channel through frequency sensing to obtain aggregated features. The fault type identification module is used to identify the fault type based on the aggregated features.

[0072] Since the principles by which the above modules solve problems are consistent with the aforementioned rolling bearing fault diagnosis method that optimizes input by reverse guidance of discrimination information, the implementation of this system can refer to the implementation of the aforementioned method, and the repetitive parts will not be repeated.

[0073] Furthermore, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements a rolling bearing fault diagnosis method for reverse-guided input optimization based on discrimination information as described in any of the preceding claims.

[0074] Alternatively, a computer device may be provided, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements a rolling bearing fault diagnosis method optimized by back-guided input based on discriminant information as described in any of the preceding claims.

[0075] In some embodiments, the processor may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits packaged with the same or different functions. This includes combinations of one or more central processing units, microprocessors, digital processing chips, graphics processors, and various control chips. The processor is the control core of the electronic device, connecting various components of the device through various interfaces and lines. It executes programs or modules stored in memory and calls data stored in the memory to perform various functions and process data within the electronic device.

[0076] Any content or technical means not mentioned in the embodiments of this invention can be obtained by referring to the prior art. This disclosure does not limit the scope of the invention and therefore will not be elaborated further.

[0077] The embodiments of the present invention have been described in detail above, and the principles and implementation methods of the present invention have been explained. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of the present invention. Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, computer software program products, or electronic devices, etc. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of one or more storage media (including but not limited to disk storage, CDs) containing computer-usable program code. The form of a computer program product implemented on ROM, optical memory, etc.

[0078] It should be noted that the word "comprising" does not exclude the presence of components or steps not listed in the claims. The words "a" or "an" preceding a component do not exclude the presence of a plurality of such components. This invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer.

[0079] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0080] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for diagnosing rolling bearing faults by optimizing input based on discriminant information back-guided input, characterized in that, include: Based on vibration signals under normal operating conditions, a dictionary matrix is ​​constructed by introducing discrimination constraints; The input signal to be diagnosed is reconstructed using the dictionary matrix to obtain the reconstructed signal; The reconstructed signal is input into the rolling bearing fault diagnosis model to obtain the fault type; The rolling bearing fault diagnosis model is used for feature extraction, and adaptive aggregation of the extracted features based on channel dimension through frequency perception, and identification of fault type based on the aggregated features. During the training of the rolling bearing fault diagnosis model, the reconstructed signal is adaptively adjusted based on the gradient information of the fault classification loss function, and the model parameters are optimized based on the adjusted signal.

2. The method as described in claim 1, characterized in that, Based on vibration signals under normal operating conditions, a dictionary matrix is ​​constructed by introducing discriminant constraints, including: constructing the following objective function based on the discriminant constraints, and obtaining the dictionary matrix through iterative solution; In the formula, , These represent vibration signals under normal operating conditions. The initial dictionary matrix and initial sparse coefficient matrix are obtained through nonnegative matrix decomposition. Describe the objective function. This refers to signal reconstruction error, used to ensure the physical reconstruction accuracy of the signal. for Norm sparse constraints are used to suppress noise and highlight key fault features in the signal. The sparse regularization coefficient; For classification loss function, For real labels, Represents a classification network mapping. The constraint weights are used to balance reconstruction accuracy and discrimination performance.

3. The method of claim 2, wherein, During iterative solution, an alternating optimization strategy is adopted, including: in each iteration, first fix the initial dictionary matrix and update the initial sparse coefficient matrix, then fix the updated sparse coefficient matrix to update the initial dictionary matrix, wherein the update method is as follows: 。 4. The method of claim 1, wherein, The input signal to be diagnosed is reconstructed using the dictionary matrix to obtain the reconstructed signal. include: In the formula, is a sparse coefficient vector corresponding to the input signal to be diagnosed, denotes the updated dictionary matrix.

5. The method of claim 1, wherein, Feature extraction includes extracting multi-scale features using convolutional kernels of different scales and then concatenating the multi-scale features.

6. The method of claim 1, wherein, Based on the channel dimension, adaptive aggregation of extracted features is performed using frequency awareness, including: The amplitude spectrum is obtained by performing a Fast Fourier Transform on the channel features of the extracted features. ; calculating a dominant frequency of the amplitude spectrum ; with the rotational base frequency as a reference, introducing a dimensionless frequency boundary parameter and , defining a frequency gating coefficient wherein denotes a gating function based on the dominant frequency position; The frequency gating coefficients are used to weight and aggregate the features of each channel to obtain aggregated features.

7. The method of claim 1, wherein, The reconstructed signal is adaptively adjusted based on the gradient information of the fault classification loss function, including: The gradient of the reconstructed signal with respect to the fault classification loss function is calculated in reverse. The reconstructed signal is adaptively enhanced based on the gradient, as expressed by the following formula: In the formula, Indicates the reconstructed signal. Indicates an optimized signal. Represents the gradient. Indicates the adaptive dynamic enhancement step size. Indicates the maximum adjustment range. Indicates a length of The local standard deviation calculated within the sliding window. This represents the maximum local standard deviation within the global scope.

8. A rolling bearing fault diagnosis system that discriminates information reverse direction guide input optimization, characterized by, A rolling bearing fault diagnosis method for implementing the discriminant information reverse-guided input optimization method as described in any one of claims 1 to 7 includes: The input signal reconstruction module is used to construct a dictionary matrix based on the vibration signal under normal operating conditions by introducing discrimination constraints; and to reconstruct the input signal to be diagnosed using the dictionary matrix to obtain the reconstructed signal. The fault type identification module has a built-in rolling bearing fault diagnosis model, which is used to identify the fault type based on the reconstructed signal; The model optimization training module is used to adaptively adjust the reconstructed signal based on the gradient information of the fault classification loss function, and optimize the model parameters according to the adjusted signal.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by the processor, it implements a rolling bearing fault diagnosis method optimized by reverse guidance of discrimination information as described in any one of claims 1 to 7.

10. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements a rolling bearing fault diagnosis method optimized by reverse guidance of discrimination information as described in any one of claims 1 to 7.

Images