Bearing fault diagnosis method, system, equipment and medium

Abstract

The invention relates to a bearing fault diagnosis method, system, equipment and medium, and belongs to the technical field of fault diagnos.The bearing fault diagnosis method comprises the following steps that firstly, according to the characteristics of bearing vibration signals, a DCT-Laplace dictionary is designed by using a discrete cosine transform primary function and Laplace wavelets as atoms so as to represent harmonic waves and impact components of the signals; then, estimating a standard deviation of noise in the signal by using a wavelet transform noise estimation method of the de-impact region; and finally, according to the standard deviation of the noise, comparing the power of the noise and the power of the reconstructed residual signal so as to adaptively determine the expansion number of the network, thereby obtaining an unsupervised noise adaptive matching pursuit algorithm expanded de-noising network. According to the invention, the problem of insufficient optimization of the network expansion number is solved, the network interpretability is improved, and the noise reduction capability is good.

Description

Technical Field

[0001] The present disclosure belongs to the technical field of fault diagnosis, and in particular relates to a bearing fault diagnosis method, system, equipment and medium. Background Art

[0002] Rolling bearings are essential supporting components widely used in rotating machinery. However, in practical applications such as high-speed trains, helicopter main reducers, and wind turbines, bearings often experience rapid wear and damage due to the complex and harsh operating conditions. Consequently, bearing failures can cause significant economic losses and even safety issues. Therefore, health monitoring of rolling bearings is crucial. Mechanical vibration signals can more accurately reflect fault characteristics than temperature, sound, current, and voltage. Vibration signal analysis is the most commonly used method in existing bearing health monitoring research.

[0003] In real-world engineering, mechanical equipment often operates in harsh and noisy environments. Collected vibration signals inevitably contain significant noise interference, which can overwhelm the characteristics of the vibration signal that reflect bearing health. This is especially true in the presence of strong background noise and subtle impacts that indicate early bearing failures. This limits diagnostic accuracy in noisy environments, making denoising the collected vibration signals essential.

[0004] Existing signal processing-based denoising methods include fast Fourier transform, wavelet transform, empirical mode decomposition and its improved variants, as well as sparse representation-based methods such as orthogonal matching pursuit and basis pursuit. However, these methods all have the disadvantages of relying on expert knowledge, lacking adaptability, and difficulty in selecting hyperparameters.

[0005] With the rapid development and widespread application of deep learning, denoising methods based on deep learning have been studied. A non-local fully convolutional neural network has been proposed, in which non-local methods are used to enhance the learning ability of long-distance dependencies. AI et al. designed an encoder-decoder network with cross-layer residual connections for signal denoising. A bearing fault diagnosis method based on multi-layer noise reduction technology and improved convolutional neural networks has also been proposed. However, deep neural networks are regarded as "black box" function approximators when used, lacking interpretability. Their structure is usually designed based on empirical trial and error, and there is a lack of systematic network design methods.

[0006] To address the "black box" problem of deep neural networks, sparse algorithm expansion provides a method for systematically designing neural networks by expanding iterative algorithms, thereby improving the interpretability of deep neural networks. This paper first proposed a learnable iterative shrinkage threshold algorithm designed by expanding iterative shrinkage threshold algorithms. By applying the convolutional sparse coding model to image representation, a convolutional sparse coding forward propagation network that follows the MMSE (Minimum MSE, MMSE) approximation process was proposed for supervised image denoising. A deep expanded convolutional dictionary learning network for image denoising has also been proposed. Using a multi-layer sparse coding model to improve the CDL network, a multi-layer convolutional dictionary learning network was proposed for signal denoising.

[0007] A drawback of existing signal processing methods is that they often rely on expert knowledge to select appropriate filtering parameters, such as the mother wavelet function or threshold settings for wavelet transforms, or the construction of dictionaries for sparse representations. This subjective selection results in a lack of adaptability and suboptimal performance.

[0008] The disadvantages of data-driven methods are: currently, most methods are supervised and require a large number of clean signal samples, which are difficult to obtain in actual engineering; at the same time, the supervised denoising network uses artificially added noise to obtain training sample label pairs during training, which has certain approximations; in addition, the number of layers of the sparse algorithm in the existing methods needs to be pre-set. Too many layers will increase the difficulty of training, while insufficient layers will reduce the denoising ability of the network.

[0009] Therefore, it is necessary to provide a new bearing fault diagnosis method, system, equipment and medium to solve the above technical problems. Summary of the Invention

[0010] The purpose of the present disclosure is to provide a bearing fault diagnosis method, system, device and medium in order to solve the above problems.

[0011] The present disclosure achieves the above objectives through the following technical solutions: A bearing fault diagnosis method comprises the following steps: Acquire a vibration signal of the bearing, determine and remove an impact region in the vibration signal, and obtain a noisy signal in the impact-free region; perform a wavelet transform on the noisy signal in the impact-free region, and estimate a noise standard deviation based on the wavelet coefficients; Expanding the iterative steps of the matching pursuit algorithm into a denoising network based on the DCT-Laplace dictionary, adaptively determining the network expansion number based on a comparison between the noise standard deviation and the power of the residual signal reconstructed in the matching pursuit algorithm, and constructing a denoising network based on the network expansion number to achieve vibration signal denoising; Fault features are extracted based on the denoised vibration signal to complete bearing fault diagnosis.

[0012] As a further optimization solution of the present disclosure, the impact area in the vibration signal is determined and removed to obtain a noise-containing signal without the impact area, including: Performing Hilbert transform on the vibration signal to obtain an analytical signal, and calculating the absolute value of the analytical signal to obtain an envelope signal; The area in the envelope signal where the amplitude is greater than a preset threshold is defined as the impact area, and the signal value in the area is set to zero to obtain a noise-containing signal without the impact area.

[0013] As a further optimization solution of the present disclosure, the noise standard deviation is estimated based on the wavelet coefficients, including: The noisy signal in the de-shocked area is decomposed into one layer using Db3 wavelet to obtain the detail wavelet coefficients; A noise standard deviation is estimated based on the detail wavelet coefficients.

[0014] As a further optimization solution of the present disclosure, the DCT-Laplace dictionary is composed of discrete cosine transform basis functions and Laplace wavelets, which are used to represent the harmonic components and impulse components of the signal respectively.

[0015] As a further optimization solution of the present disclosure, the iterative steps of the matching pursuit algorithm include: Initialize the residual signal and sparse representation vector; Calculate the inner product of the residual signal and each atom in the DCT-Laplace dictionary, select the atom with the largest absolute value of the inner product, update the sparse representation vector and the residual signal, and repeat until the power of the residual signal meets the stopping condition.

[0016] As a further optimization solution of the present disclosure, the network expansion number is adaptively determined based on the comparison between the noise standard deviation and the power of the reconstructed residual signal, including: Calculating the power and noise power of the current residual signal, where the noise power is the square of the noise standard deviation; When the power of the residual signal is less than the noise power, the iteration is stopped and the current number of iterations is the optimal expansion number; otherwise, the number of expansion layers is increased and the comparison is repeated.

[0017] As a further optimization solution of the present disclosure, fault features are extracted based on the denoised vibration signal to complete bearing fault diagnosis, including: Extract the fault characteristic frequency of the denoised vibration signal; Using the pre-established mapping relationship between fault characteristic frequency and bearing fault type, find the bearing fault type corresponding to the extracted fault characteristic frequency; The found bearing fault type is output as the diagnosis result.

[0018] A bearing fault diagnosis system, comprising: A wavelet transform noise estimation module for removing the impact area is used to obtain the vibration signal of the bearing, determine the impact area in the vibration signal and remove it to obtain a noisy signal in the impact area; perform a wavelet transform on the noisy signal in the impact area and estimate the noise standard deviation based on the wavelet coefficients; A noise-adaptive algorithm expansion denoising network module is used to expand the iterative steps of the matching pursuit algorithm into a denoising network based on the DCT-Laplace dictionary, and adaptively determine the number of network expansions based on a comparison between the noise standard deviation and the power of the residual signal reconstructed in the matching pursuit algorithm, and construct a denoising network based on the number of expansion layers to achieve vibration signal denoising; The fault diagnosis module is used to extract fault features based on the denoised vibration signal and complete bearing fault diagnosis.

[0019] An electronic device comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory for storing computer programs; The processor is used to execute the program stored in the memory to implement the bearing fault diagnosis method.

[0020] A computer-readable storage medium stores a computer program, which implements the bearing fault diagnosis method when executed by a processor.

[0021] The beneficial effects of the present disclosure are: The present invention proposes a wavelet transform noise estimation method for the de-impacted area, and based on this method designs an unsupervised noise adaptive matching pursuit algorithm to expand the denoising network based on the power of noise and reconstruction residual to adaptively determine the expansion number of the sparse algorithm expansion network, which solves the problem of suboptimal network expansion number and has good noise reduction ability while improving the interpretability of the network.

[0022] The present invention proposes a wavelet transform noise estimation method for the de-shocked area. Taking into account the influence of the fault impulse response on the accuracy of the noise intensity estimation, the Hilbert transform is used to process the signal to be analyzed, and a threshold method is constructed to obtain the position of the fault impulse. On this basis, the signal of the de-shocked area is processed by wavelet decomposition and the noise intensity is estimated using the wavelet coefficients, thereby improving the accuracy of the noise intensity estimation.

[0023] This paper proposes an unsupervised noise-adaptive matching pursuit algorithm expansion denoising network that adaptively determines the number of sparse algorithm expansions based on the power of noise and reconstruction residuals. Considering the characteristics of bearing vibration signals, discrete cosine transform basis functions and Laplace wavelets are used as atoms to design a DCT-Laplace dictionary to improve the accuracy of impact region estimation. A method for adaptively determining the number of expansion layers based on the power of the reconstructed residual signal and the noise power of the de-impulsed signal is constructed, forming an unsupervised noise-adaptive matching pursuit algorithm expansion denoising network, which improves the network's noise reduction capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 is a flow chart of a method in an embodiment of the present disclosure; Figure 2 (a) is a comparison chart of the accuracy of the noise intensity estimation method of the present disclosure in an embodiment of the present disclosure; Figure 2 (b) is a comparison chart of the accuracy of the noise intensity estimation method based on wavelet transform in an embodiment of the present disclosure; Figure 3 (a) is a comparison diagram of the diagnostic accuracy using the denoising method of the present disclosure and the other 10 different denoising methods in an embodiment of the present disclosure; Figure 3 (b) is a histogram showing the comparison range of diagnostic accuracy using the denoising method of the present disclosure and the other 10 different denoising methods in an embodiment of the present disclosure; Figure 4 is a system structure block diagram in an embodiment of the present disclosure; Figure 5 It is a block diagram of the device structure in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0025] The present application will be described in further detail below in conjunction with the accompanying drawings. It is necessary to point out that the following specific implementation methods are only used to further illustrate the present application and cannot be understood as limiting the scope of protection of the present application. Technicians in this field can make some non-essential improvements and adjustments to the present application based on the above application content.

[0026] like Figure 1 As shown, a bearing fault diagnosis method includes the following steps: Acquire a vibration signal of the bearing, determine and remove an impact region in the vibration signal, and obtain a noisy signal in the impact-free region; perform a wavelet transform on the noisy signal in the impact-free region, and estimate a noise standard deviation based on the wavelet coefficients; Expanding the iterative steps of the matching pursuit algorithm into a denoising network based on the DCT-Laplace dictionary, adaptively determining the network expansion number based on a comparison between the noise standard deviation and the power of the residual signal reconstructed in the matching pursuit algorithm, and constructing a denoising network based on the network expansion number to achieve vibration signal denoising; Fault features are extracted based on the denoised vibration signal to complete bearing fault diagnosis.

[0027] In this embodiment, a bearing fault diagnosis method mainly includes two steps: wavelet transform noise estimation in the impact removal area and noise adaptive algorithm expansion denoising network. The specific contents are as follows: In the wavelet transform noise estimation to remove the impact area, the impact area in the vibration signal is first determined. , its analytical signal can be obtained through Hilbert transform: ;(1) ;(2) in, represents the Hilbert transform operator, represents the Cauchy principal value, Indicates time, Indicates time delay, represents pi, represents an imaginary number, is the real part of the signal, is the imaginary part of the signal, represents the instantaneous phase of the analytical signal, represents the instantaneous amplitude of the analytical signal, is the Hilbert transform operator. The envelope signal can be calculated by analyzing the absolute value of the signal.

[0028] By setting a threshold parameter (taken as 5), with 1 / 5 of the maximum value of the envelope signal as the threshold, any area in the envelope signal with an amplitude greater than this set threshold is considered to be the impact area that needs to be removed. Assume that the calculation time value is defined as , then the length is The noise sequence The impact area is: ;(3) Where, Represents the maximum operator.

[0029] Remove the signal values ​​in the time region of the signal that falls within the impact region defined by the above formula to obtain the noisy signal without the impact region: ;(4) in, is the complete time region of the signal; is the potential of the set of impact areas, which indicates the number of data points contained in the impact area.

[0030] Then perform wavelet transform on the result of removing the shock area. Assign it as the initial value of the algorithm, that is, , decomposed according to Mallat's fishbone algorithm: ;(5) ;(6) Where, and are respectively called the decomposition level The approximate wavelet coefficients and detail wavelet coefficients under Z represent integers, m belongs to Z and is also an integer. Represents the filter coefficient. This disclosure uses the detail wavelet coefficient of Db3 wavelet when the decomposition level is 1 to estimate the noise. The estimated value of the noise standard deviation is calculated as follows: ;(7) Where, Represents the median operator. In the noise adaptive algorithm, the matching pursuit algorithm is first used to find the optimal atom in the denoising network. , the matching pursuit algorithm is to express the signal as ,in and Respectively represent atoms and their corresponding coefficients, is sparse and satisfies , only a very small number of atoms are used to represent the signal, and most elements in the sparse representation vector are zero. The specific steps are as follows: 1) The initial residual of the signal is recorded as , assign the signal to the initial value of the residual signal, that is The initial value of the sparse representation vector is set to zero, that is .

[0031] 2) Calculate the inner product between the residual and each atom, and find the largest absolute value of the inner product to obtain the atom that best matches the residual, that is: ;(8) Where, represents the supremum operator.

[0032] 3) Update the residual by subtracting the best matching atom from the residual: ;(9) In the formula, it involves the update of the sparse representation vector, .

[0033] 4) Iteration stops. The signal is decomposed into a linear combination of all the selected best matching atoms: ;(10) The iterative steps of the matching pursuit algorithm can then be expanded into a denoising network in time. Note that the computational process of finding the atom corresponding to the maximum inner product and calculating the coefficient in step 2 is non-differentiable, so the original operation needs to be replaced by a maximum projection threshold unit: ;(11) Where, Indicates the maximum projection and its operator. In addition, it is not assumed that the atoms in the dictionary are normalized, so when updating the sparse representation vector in step 3, it is necessary to multiply it by the weight matrix: ;(12) In the formula Denotes the diagonalization operator. Using the estimated value of the wavelet transform noise estimation method in the de-shocked region, a noise-adaptive iterative stopping condition is proposed. For larger noise levels, a smaller expansion number is used to avoid noise in the reconstructed signal, while for smaller noise levels, a larger expansion number is used to accurately reconstruct the signal. This results in the noise-adaptive expansion number of the expansion network. The specific concept is: at each iteration, the reconstructed residual signal is considered to be the removed noise, and the power of the removed noise is calculated from the residual signal. As the number of iterations increases, more atoms are selected, the amplitude of the reconstructed signal gradually increases, and the power of the residual signal decreases accordingly. The residual signal is initialized to the entire noisy signal, and its power must be greater than the noise power estimated by the improved wavelet noise estimation method in the de-shocked region. As the power of the residual signal decreases until it is less than or equal to the estimated noise power (the noise power is calculated as the square of the estimated noise standard deviation, since for zero-mean Gaussian white noise, the power is equal to the standard deviation), the noise removal is considered to be of appropriate magnitude. The number of iterations of the expansion algorithm at this point is the noise-adaptive expansion number obtained: ; (13) In the formula According to the above analysis on the insufficiency of MP algorithm with fixed expansion number, for larger noise, a smaller expansion number should be used to avoid noise in the reconstructed signal, while for smaller noise, a larger expansion number should be used to achieve accurate reconstruction of the signal.

[0034] Based on the aforementioned noise-adaptive expansion number, an unsupervised noise-adaptive matching pursuit algorithm is proposed. The algorithm's iterative steps are expanded over time into the network's forward structure, resulting in an unsupervised noise-adaptive matching pursuit algorithm expansion denoising network. Compared to supervised denoising methods, the proposed network is more suitable for practical situations where clean signal labels are lacking, and has a wider range of applicability.

[0035] In order to verify the effectiveness and advancement of the method proposed in the present invention, the performance of the noise power estimation method and the unsupervised noise adaptive matching pursuit algorithm denoising network proposed in the present invention were verified on the rolling bearing dataset.

[0036] Table 1 shows the accuracy comparison results of the noise intensity estimation method of the disclosed method and the noise intensity estimation method based on wavelet transform. As shown in Table 1, compared with the noise intensity estimation method based on wavelet transform, the estimation error of the method proposed by the present invention is smaller. At the same time, Figure 2 The figure shows the estimation results of the disclosed method and the wavelet transform method under different noise intensities. Figure 2 It can be seen from the shown content that the estimation results of the method proposed in the present disclosure within the noise intensity range of -15dB to 10dB are all derived from the method based on wavelet transform.

[0037] Table 1

[0038] Figure 3 The figure shows the model diagnosis accuracy of the method proposed in this disclosure and nearly 10 noise reduction methods. Figure 3It can be seen that several classic denoising methods, including the fast Fourier transform (FFT), wavelet transform (Wavelet), orthogonal matching pursuit (OMP), and basis pursuit denoising (BPDN), can achieve an intermediate level of denoising performance. Partial ensemble empirical mode decomposition (PEEMD) performs slightly better than these methods. Three networks, the multi-layer convolutional dictionary learning network (ML-CDL), the convolutional autoencoder (ConvAE), and the non-local fully convolutional neural network (NL-FCNN), trained using unsupervised loss, use noisy samples as labels due to the lack of clean signal labels. The improvement in the signal-to-noise ratio (SNR) after denoising is relatively small compared to the input noise, and the improvement in diagnostic accuracy compared to the undenoised accuracy is also relatively small. The NL-FCNN performs slightly better than the other two networks. The unsupervised noise adaptive matching pursuit algorithm denoising network (NAUMP) proposed in this paper outperforms all compared unsupervised denoising methods on both metrics. Compared with the supervised ML-CDL network, its results are slightly worse. Although its denoised SNR is lower than that of the ML-CDL network, it achieves similar or even better results on ACC. This shows that NAUMP achieves denoising results on unsupervised tasks that are somewhat close to those of supervised networks, validating the superiority of the proposed method.

[0039] This paper proposes a wavelet transform noise estimation method for shock-removed regions. First, the Hilbert transform is used to process the signal to be analyzed and the absolute value of the transform result is calculated to obtain the signal envelope. Then, using 1 / 5 of the maximum signal value as a threshold, the shock-existing region of the envelope result is determined and the signal amplitude in this region is set to zero, resulting in a shock-removed noisy signal. Finally, the shock-removed envelope result is subjected to a wavelet transform, and the noise intensity is estimated based on the wavelet coefficients in the first-layer wavelet decomposition result.

[0040] This disclosure proposes an unsupervised noise-adaptive matching pursuit algorithm expansion denoising network that adaptively determines the number of sparse algorithm expansions based on the power of the noise and reconstructed residual signal. First, the signal is processed using a single-layer expansion model to obtain a reconstructed signal and a residual signal. Then, the residual signal power is calculated and compared with the estimated noise power of the de-shocked signal. If the residual signal power meets the set conditions, the current number of expansion layers is optimal. Otherwise, the number of expansion layers is increased and the process is repeated until the set conditions are met. Finally, the optimal unsupervised noise-adaptive matching pursuit algorithm expansion denoising network is obtained based on the optimal number of expansion layers.

[0041] like Figure 4 As shown, an embodiment of the present disclosure provides a bearing fault diagnosis system, comprising: A wavelet transform noise estimation module for removing the impact area is used to obtain the vibration signal of the bearing, determine and remove the impact area in the vibration signal, perform a wavelet transform on the signal after the impact is removed, and estimate the noise standard deviation based on the wavelet coefficients; A noise-adaptive algorithm expansion denoising network module is used to expand the iterative steps of the matching pursuit algorithm into a denoising network based on the DCT-Laplace dictionary, and adaptively determine the number of network expansions based on the comparison between the noise standard deviation and the reconstructed residual signal power, and construct a denoising network based on the number of expansion layers to achieve vibration signal denoising; The fault diagnosis module is used to extract fault features based on the denoised vibration signal and complete bearing fault diagnosis.

[0042] The implementation process of the functions and effects of each module in the above system is specifically described in the implementation process of the corresponding steps in the above method, which will not be repeated here.

[0043] For the system embodiment, since it basically corresponds to the method embodiment, the relevant parts can be referred to the partial description of the method embodiment. The system embodiment described above is only schematic, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, that is, they may be located in one place, or they may be distributed on multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. Those of ordinary skill in the art can understand and implement it without paying any creative work.

[0044] In the above embodiments, any number of all modules can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality of other modules and implemented in one module. At least one of all modules can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented in hardware or firmware by any other reasonable means of integrating or packaging the circuit, or can be implemented in any one of the three implementation methods of software, hardware, and firmware, or in an appropriate combination of any of them. Alternatively, at least one of all modules can be at least partially implemented as a computer program module, which can perform the corresponding function when the computer program module is executed.

[0045] See also Figure 5The electronic device provided by an embodiment of the present disclosure includes a processor 1110, a communication interface 1120, a memory 1130 and a communication bus 1140, wherein the processor 1110, the communication interface 1120, and the memory 1130 communicate with each other through the communication bus 1140; Memory 1130, for storing computer programs; The processor 1110 is configured to implement the above-mentioned bearing fault diagnosis method when executing the program stored in the memory 1130 .

[0046] The communication bus 1140 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The communication bus 1140 can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, the figure shows only one thick line, but this does not mean that there is only one bus or only one type of bus.

[0047] The communication interface 1120 is used for communication between the electronic device and other devices.

[0048] The memory 1130 may include a random access memory (RAM) or a non-volatile memory, such as at least one disk storage. Alternatively, the memory 1130 may be at least one storage device located away from the processor 1110.

[0049] The above-mentioned processor 1110 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0050] The embodiments of the present disclosure further provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the bearing fault diagnosis method described above.

[0051] The computer-readable storage medium may be included in the device / apparatus described in the above embodiments, or may exist independently without being incorporated into the device / apparatus. The computer-readable storage medium carries one or more programs that, when executed, implement the bearing fault diagnosis method according to the embodiments of the present disclosure.

[0052] According to embodiments of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium, such as, but not limited to, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0053] The above embodiments merely illustrate several implementation methods of the present disclosure, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present disclosure. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the scope of the present disclosure, all of which fall within the scope of protection of the present disclosure.

Claims

1. A bearing fault diagnosis method, characterized in that: The following steps are involved: Acquire a vibration signal of the bearing, determine and remove an impact region in the vibration signal, and obtain a noisy signal in the impact-free region; perform a wavelet transform on the noisy signal in the impact-free region, and estimate a noise standard deviation based on the wavelet coefficients; Expanding the iterative steps of the matching pursuit algorithm into a denoising network based on the DCT-Laplace dictionary, adaptively determining the network expansion number based on a comparison between the noise standard deviation and the power of the residual signal reconstructed in the matching pursuit algorithm, and constructing a denoising network based on the network expansion number to achieve vibration signal denoising; Fault features are extracted based on the denoised vibration signal to complete bearing fault diagnosis.

2. A bearing fault diagnosis method according to claim 1, characterized in that: Determining and removing the impact area in the vibration signal to obtain a noise-containing signal without the impact area includes: Performing Hilbert transform on the vibration signal to obtain an analytical signal, and calculating the absolute value of the analytical signal to obtain an envelope signal; The area in the envelope signal where the amplitude is greater than a preset threshold is defined as the impact area, and the signal value in the area is set to zero to obtain a noise-containing signal without the impact area.

3. A bearing fault diagnosis method according to any one of claims 1 or 2, characterized in that: Estimate the noise standard deviation based on the wavelet coefficients, including: The noisy signal in the de-shocked area is decomposed into one layer using Db3 wavelet to obtain the detail wavelet coefficients; A noise standard deviation is estimated based on the detail wavelet coefficients.

4. A bearing fault diagnosis method according to claim 1, characterized in that: The DCT-Laplace dictionary is composed of discrete cosine transform basis functions and Laplace wavelets, which are used to represent the harmonic components and impulse components of the signal respectively.

5. A bearing fault diagnosis method according to claim 1, characterized in that: The iterative steps of the matching pursuit algorithm include: Initialize the residual signal and sparse representation vector; Calculate the inner product of the residual signal and each atom in the DCT-Laplace dictionary, select the atom with the largest absolute value of the inner product, update the sparse representation vector and the residual signal, and repeat until the power of the residual signal meets the stopping condition.

6. A bearing fault diagnosis method according to claim 5, characterized in that: The network expansion number is adaptively determined based on the comparison between the noise standard deviation and the power of the reconstructed residual signal, including: Calculating the power and noise power of the current residual signal, where the noise power is the square of the noise standard deviation; When the power of the residual signal is less than the noise power, the iteration is stopped and the current number of iterations is the optimal expansion number; otherwise, the number of expansion layers is increased and the comparison is repeated.

7. A bearing fault diagnosis method according to claim 1, characterized in that: Extract fault features based on the denoised vibration signal to complete bearing fault diagnosis, including: Extract the fault characteristic frequency of the denoised vibration signal; Using the pre-established mapping relationship between fault characteristic frequency and bearing fault type, find the bearing fault type corresponding to the extracted fault characteristic frequency; The found bearing fault type is output as the diagnosis result.

8. A bearing fault diagnosis system, characterized in that: include: A wavelet transform noise estimation module for removing the impact area is used to obtain the vibration signal of the bearing, determine the impact area in the vibration signal and remove it to obtain a noisy signal in the impact area; perform a wavelet transform on the noisy signal in the impact area and estimate the noise standard deviation based on the wavelet coefficients; A noise-adaptive algorithm expansion denoising network module is used to expand the iterative steps of the matching pursuit algorithm into a denoising network based on the DCT-Laplace dictionary, and adaptively determine the number of network expansions based on a comparison between the noise standard deviation and the power of the residual signal reconstructed in the matching pursuit algorithm, and construct a denoising network based on the number of expansion layers to achieve vibration signal denoising; The fault diagnosis module is used to extract fault features based on the denoised vibration signal and complete bearing fault diagnosis.

9. An electronic device, characterized in that: The processor, the communication interface, the memory and the communication bus are connected to each other via the communication bus. Memory for storing computer programs; The processor is used to execute the program stored in the memory to implement the bearing fault diagnosis method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the bearing fault diagnosis method according to any one of claims 1 to 7 is implemented.

Images