Rolling bearing fault signal noise reduction and feature extraction method based on ASTENED and STFT

Abstract

The invention provides a rolling bearing fault signal noise reduction and feature extraction method based on adaptive semi-soft threshold set noise reconstruction empirical mode decomposition (ASTENEEMD) and (Sort-time Fourier Transform, STFT), and belongs to the field of intelligent diagnosis of rolling bearing faults. The rolling bearing fault signal noise reduction and feature extraction method based on the adaptive semi-soft threshold set noise reconstruction empirical mode decomposition (ASTENEEMD) and the Sort-time Fourier Transform (STFT) is characterized in that the rolling bearing fault signal noise reduction and feature extraction method based on the adaptive semi-soft threshold set noise reconstruction empirical mode decomposition (Sort-time Fourier Transform (STFT) is provided. According to the method, firstly, noise reduction of an original vibration signal of the rolling bearing is achieved through self-adaptive semi-soft threshold set noise reconstruction empirical mode decomposition, then, the noise reduction signal is converted into a time-frequency image through short-time Fourier transform, and time-frequency feature extraction of a non-linear and non-stable vibration signal is achieved. According to the method, the rolling bearing fault signal quality is enhanced, the noise reduction effect can be remarkably improved, the fault features in the signal are clearer, rich time-frequency features can be extracted, and reliable input is provided for subsequent fault diagnosis by using a deep learning model.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent diagnosis of rolling bearing faults, and particularly relates to a method for noise reduction and feature extraction of rolling bearing fault signals based on adaptive semi-soft threshold ensemble noise reconstructed empirical mode decomposition and short-time Fourier transform. Background Art

[0002] Under the background of intelligent manufacturing, mechanical equipment is continuously developing towards integration, automation and intelligence. Under this background, rolling bearings are increasingly widely used in the ship field. Especially in the propeller shaft power system, as a core component, it plays a key role in ensuring the safety and reliability of ships. The propeller shaft power system is one of the most core systems of a ship. Efficient diagnosis of the faults of this system is crucial for ensuring the safe navigation of the ship and is also the core link to realize an intelligent ship. As a key component in the ship power propeller shaft system, during the actual variable working condition operation, due to the complex working environment of the ship and various external interference factors, the fault signals are often masked by high-intensity background noises such as engine vibration, wave impact and hydrodynamic force, showing non-linear and non-stationary characteristics. These noises will not only weaken the significant features of the fault signals, but may also introduce misleading information, resulting in inaccurate extraction of fault features, thereby reducing the sensitivity and accuracy of the diagnosis system. In addition, noise reduction processing is particularly crucial for capturing early weak fault signals of bearings, which can effectively enhance the intensity of key feature signals and provide a clearer data basis for subsequent feature extraction and classification. Therefore, it is extremely important to develop an efficient signal noise reduction technology to perform data noise reduction on the original signal, so as to eliminate multi-source noise interference under complex working conditions, improve the diagnosis effect, and better detect and prevent equipment faults.

[0003] After retrieval, the application publication number is CN112504679A, a method for noise reduction of bearing fault signals based on least mean square empirical mode decomposition, including the following steps: 1) Decompose the collected bearing signals by the CEEMDAN method and decompose the signals into different modal spaces; 2) Use the IMF components of different modes as the source signals of normalized adaptive filtering, add Gaussian white noise as the desired signal, and use the normalized adaptive filtering method to filter the IMF components of different modes. The difference between the source signal and the output signal is the IMF component after adaptive filtering; 3) Reconstruct the filtered IMF components of each mode. The present invention applies normalized least mean square complete ensemble empirical mode decomposition to the noise reduction of bearing fault signals. The steps are simple and easy to implement, can effectively eliminate the noise of bearing fault signals, enhance the fault feature information, and improve the signal-to-noise ratio of the signal.

[0004] CN112504679A, a bearing fault signal denoising method based on complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), indeed achieves fault signal denoising by using CEEMDAN and enhances the fault feature information. This method needs to introduce Gaussian white noise during CEEMDAN decomposition to improve the decomposition stability. However, the added noise may cause certain interference to the decomposition result. Especially when the noise amplitude is not properly selected, it will affect the quality of the IMF components. Although it is processed by normalized adaptive filtering, for high-noise scenarios, it may not be able to completely eliminate the masking effect of noise on fault features. Determining the amplitude of white noise depends on expert experience. If the noise amplitude is too large, it will introduce too many random components, resulting in the decomposition result deviating from the original signal characteristics. If the noise amplitude is too small, it cannot effectively smooth the mode mixing problem. Improper noise amplitude may lead to unstable or distorted decomposition results, affecting the diagnostic accuracy.

[0005] Aiming at the problem that noise and interference often mask the true fault features and reduce the robustness and accuracy of the diagnostic model, this patent proposes an adaptive noise estimation method based on the peak factor and a semi-soft threshold technique. It directly uses the inherent noise existing in the original signal itself instead of adding external Gaussian white noise. Through the noise estimation technique, repeated iterative decomposition is carried out. This processing avoids the mode mixing problem and reduces the inherent noise itself. Compared with the original fixed threshold, the adaptive noise estimation method based on the peak factor can reduce the influence of noise while retaining the key fault information by introducing the reciprocal of the peak factor as the threshold adjustment factor, thus achieving more accurate signal denoising. The adaptive threshold reduces the interference of subjective factors and eliminates the dependence on artificial parameters. The semi-soft threshold technique smoothly adjusts the signal components close to the threshold instead of directly setting them to zero or significantly reducing them, thus retaining more weak signal feature information, reducing the risk of signal distortion, and avoiding the boundary effect and artifact problems. Through the strategy of semi-soft threshold reconstruction and ensemble averaging, the present invention significantly improves the signal-to-noise ratio of the signal and enhances the visibility of fault features. This patent directly generates a time-frequency image, providing an optimized input format for the deep learning model. Compared with the simple IMF component denoising in CN112504679A, this patent can extract richer and more intuitive feature information, facilitating subsequent high-precision rolling bearing fault diagnosis through the deep learning model. Summary of the Invention

[0006] The present invention aims to solve the above problems of the prior art. A rolling bearing fault signal denoising and feature extraction method based on ASTENEMD and STFT is proposed. The technical solution of the present invention is as follows:

[0007] A rolling bearing fault signal denoising and feature extraction method based on ASTENEMD and STFT, comprising the following steps:

[0008] - Collect the original vibration signal of the rolling bearing;

[0009] - Use the adaptive semi-soft threshold ensemble noise reconstructed empirical mode decomposition (ASTENEMD) for noise reduction to obtain pure intrinsic mode functions (IMFs) and the residual component;

[0010] - Convert the noise-reduced signal into a time-frequency image through the short-time Fourier transform (STFT) to extract time-frequency features.

[0011] Furthermore, the ASTENEMD method includes:

[0012] - Use EMD to decompose the original vibration signal to obtain a series of intrinsic mode components (IMFs) and the residual component; denoted as {ω j (t), 1 ≤ j ≤ n, 1 ≤ t ≤ N} and r n respectively; where n represents the number of IMFs, and N represents the length of the original time-domain signal;

[0013] - Calculate the noise energy of the IMFs and determine the confidence interval to identify the pure noise IMFs;

[0014] - Use the adaptive noise threshold estimation based on the kurtosis factor to perform noise reduction processing on the pure noise IMFs;

[0015] - Reconstruct the noise-reduced signal.

[0016] Furthermore, the calculation of the noise energy of the IMFs and determining the confidence interval to identify the pure noise IMFs specifically includes:

[0017] Calculate the noise energy of each IMF, denoted as {E j , 1 ≤ j ≤ n}, and the noise energy calculation formula is as follows:

[0018]

[0019] where median(|ω j |) represents the median of the |w j (i)| component;

[0020] Calculate the confidence intervals of the noise energy of each IMF at the 95% and 99% confidence levels and

[0021]

[0022] where The β and ρ corresponding to the 95% and 99% confidence levels are β = 0.719, ρ = 1.919 and β = 0.719, ρ = 2.449 respectively.

[0023] Furthermore, the denoising process for the pure noise IMFs using the adaptive noise threshold estimation based on the kurtosis factor specifically includes:

[0024] Select the pure noise IMFs. If then the j-th IMF can be used as the pure noise IMF;

[0025] Re-group the selected pure noise IMFs and denote them as {ω l (t), l = 1, 2, …}, and then calculate the thresholds of each pure noise IMF, denoted as {τ l , l = 1, 2, …}. The adaptive noise threshold estimation method based on the kurtosis factor is as follows;

[0026]

[0027] where σ j represents the mean absolute deviation calculated from the median of the w j (i) component, median(w j ) represents the median of the w j (i) component, N represents the length of the j-th IMF, CF represents the peak factor of the signal, Peak represents the maximum absolute value of the signal, and RMS represents the root mean square value of the signal;

[0028] Use the semi-soft threshold technique to separate the estimated noise from the above pure noise IMFs The mathematical expression of the semi-soft threshold technique is as follows;

[0029]

[0030] where x(i) represents the original signal, sgn(x) is the sign function, τ is the noise threshold, and α is the dynamic adjustment factor;

[0031] Randomly permute to obtain the m-th version of the estimated noise Then calculate the m-th version of the reconstructed signal

[0032] Perform EMD on x m (t) to obtain the m-th version of the IMFs and the residual component, denoted as {ω mj (t), 1 ≤ j ≤ n, 1 ≤ t ≤ N} and r mn ;

[0033] According to the stopping criterion, calculate the mean square error ε between ω m,j (i) and ω m+1,j (i) obtained from two iterations:

[0034]

[0035] Stop the iteration when the mean square error ε between two consecutive iterations is less than or equal to the set value ε threshold , and the stopping condition is:

[0036] ε ≤ ε threshold (6)

[0037] Calculate the mean values of all IMFs and the residual component after each iteration, and use them as the final IMFs and the residual component, denoted as and

[0038]

[0039] According to the IMF selection method based on the cross-correlation coefficient, select the IMFs rich in fault information from the obtained IMFs, denoted as {ω s (t), s = 1, 2, …}, and the IMF selection method based on the cross-correlation coefficient is expressed as follows;

[0040]

[0041] where CC j is a variable to measure the correlation degree between the original signal and the IMFs, are the mean values of the original signal x(i) and the j-th IMF respectively, and Kurtosis IMFs is the kurtosis value of each IMF, and Kurtosis Normal is the kurtosis value of the healthy signal, usually 3;

[0042] Recalculate the noise thresholds of the selected IMFs using the adaptive noise threshold estimation method based on the kurtosis factor, denoted as {τ s , s = 1, 2, …};

[0043] According to the obtained noise thresholds, use the semi-soft threshold technique to denoise the selected IMFs to obtain the denoised IMFs rich in fault information, denoted as

[0044] Reconstruct the signal

[0045] Furthermore, convert the denoised signal into a time-frequency image through the short-time Fourier transform STFT to extract time-frequency features, which specifically includes the following steps:

[0046] Perform sliding window overlapping sampling on the denoised signal of length N with m data points in each group, and the overlapping rate is r. Each sample contains m sampling points;

[0047] Therefore, the number of samples obtained is

[0048] The Hanning window is used as the window function to perform STFT on each sample respectively, extract the corresponding time-frequency diagram, and the window width is set to 256.

[0049] Furthermore, the steps of the short-time Fourier transform are as follows:

[0050] STFT divides each sample signal into many short time segments by using the window function, and performs Fourier transform on each short time segment respectively, realizing the effective extraction of time-domain and frequency-domain information of non-stationary vibration signals;

[0051] The window function is expressed as ω(t). To perform transformation on the window function, a time variable τ is introduced to represent the center position of the current window, and the transformed window function is expressed as ω(t - τ);

[0052] The Fourier transform formula is as follows:

[0053]

[0054] Among them, x(t) is a one-dimensional time-domain signal, ω is the frequency variable, and exp(-jwt) represents the complex conjugate operation;

[0055] Based on the above process, the STFT result of the denoised signal sample can be given by the following integral:

[0056]

[0057] Among them, X(t,f) represents the amplitude value at time t and frequency f.

[0058] An electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the rolling bearing fault signal denoising and feature extraction method based on ASTENEMD and STFT as described in any one of the above.

[0059] A non-transitory computer-readable storage medium stores a computer program. When the computer program is executed by a processor, it implements the rolling bearing fault signal denoising and feature extraction method based on ASTENEMD and STFT as described in any one of the above.

[0060] A computer program product includes a computer program. When the computer program is executed by a processor, it implements the rolling bearing fault signal denoising and feature extraction method based on ASTENEMD and STFT as described in any one of the above.

[0061] The advantages and beneficial effects of the present invention are as follows:

[0062] In view of the problem that the non-linear and non-stationary vibration signals in the fault diagnosis of rolling bearings contain a large degree of noise, resulting in inaccurate extraction of fault features and thus reducing the sensitivity and accuracy of the diagnosis system, a method for noise reduction and feature extraction of rolling bearing fault signals based on adaptive semi-soft threshold ensemble empirical mode decomposition and short-time Fourier transform is proposed.

[0063] 1) The adaptive noise threshold estimation method based on kurtosis factor proposed in the present invention uses the reciprocal of the peak factor to replace the original empirical coefficient C, and can adaptively denoise according to the information contained in the signal itself. Compared with the original fixed threshold, by introducing the reciprocal of the peak factor as the threshold adjustment factor, the influence of noise can be reduced while retaining the key fault information, so as to achieve more accurate signal denoising. The adaptive threshold reduces the interference of subjective factors and eliminates the dependence on artificial parameters.

[0064] 2) The semi-soft threshold technology proposed in the present invention smoothly adjusts the signal components close to the threshold instead of directly setting them to zero or significantly reducing them, thus retaining more weak signal feature information, reducing the risk of signal distortion, avoiding the problems of boundary effects and artifacts, and more effectively balancing the signal fidelity and noise reduction performance.

[0065] 3) The present invention uses STFT to convert the denoised signal into a time-frequency diagram, which can integrate the time-domain and frequency-domain information, effectively process non-linear and non-stationary vibration signals, and is helpful for robust feature extraction. Description of the Drawings

[0066] Figure 1 is a schematic diagram of the process for noise reduction and feature extraction of rolling bearing fault signals provided by the preferred embodiment of the present invention;

[0067] Figure 2 is a diagram of the experimental device of the embodiment of the present invention;

[0068] Figure 3 is the original signal of the rolling bearing in different health states in the embodiment of the present invention;

[0069] Figure 4 is the IMFs and residuals finally obtained after the original vibration signal is decomposed by ASTENEMD in the embodiment of the present invention;

[0070] Figure 5 is the relationship between the IMFs, cross-correlation coefficients and kurtosis values in different health states in the embodiment of the present invention;

[0071] Figure 6 is the IMFs of the signals in different health states before and after noise reduction in the embodiment of the present invention;

[0072] Figure 7It is the time-domain waveform before and after noise reduction of different health status signals in the embodiments of the present invention;

[0073] Figure 8 It is the frequency-domain waveform before and after noise reduction of different health status signals in the embodiments of the present invention. Specific embodiments

[0074] Next, the technical solutions in the embodiments of the present invention will be clearly and detailedly described in conjunction with the accompanying drawings in the embodiments of the present invention. The described embodiments are only a part of the embodiments of the present invention.

[0075] The technical solution for the present invention to solve the above technical problems is:

[0076] In view of the problem that the non-linear and non-stationary vibration signals in rolling bearing fault diagnosis contain a large degree of noise, resulting in inaccurate extraction of fault features, thereby reducing the sensitivity and accuracy of the diagnosis system, a method for noise reduction and feature extraction of rolling bearing fault signals based on adaptive semi-soft threshold ensemble noise-reconstructed empirical mode decomposition and short-time Fourier transform is proposed. The technical solution of the present invention is as follows:

[0077] A method for noise reduction and feature extraction of rolling bearing fault signals based on adaptive semi-soft threshold ensemble noise-reconstructed empirical mode decomposition and short-time Fourier transform, and the process framework of the method is as Figure 1 shown, and the specific steps are as follows:

[0078] A method for noise reduction and feature extraction of rolling bearing fault signals based on adaptive semi-soft threshold ensemble noise-reconstructed empirical mode decomposition and short-time Fourier transform, the method includes the following steps:

[0079] Step S1: Use a vibration sensor to collect the original vibration signal of the rolling bearing, obtain a number of signal samples, and use the Adaptive Semi-Soft Threshold EnsembleNoise-reconstructed Empirical Mode Decomposition (ASTENEMD) method to process the original vibration signal to achieve signal noise reduction;

[0080] Step S2: Use the Short-time Fourier Transform (STFT) to convert the one-dimensional time-domain signal after noise reduction into a two-dimensional time-frequency image to achieve time-frequency feature extraction.

[0081] The step S1 specifically includes the following steps:

[0082] Step S1.1: Use a vibration sensor to collect the original vibration signals of the rolling bearing, obtaining a number of signal samples. Process the original vibration signal data samples using Empirical Mode Decomposition (EMD) to obtain the Intrinsic Mode Functions (IMFs) and Residuals at different time scales, denoted as {ω j (t), 1 ≤ j ≤ n, 1 ≤ t ≤ N} and r n ;

[0083] where n represents the number of IMFs and N represents the length of the original time-domain signal.

[0084] Step S1.2: Calculate the noise energy of each IMF, denoted as {E j , 1 ≤ j ≤ n}. The formula for calculating the noise energy is as follows:

[0085]

[0086] where median(|ω j |) represents the median of the |w j (i)| component.

[0087] Step S1.3: Calculate the confidence intervals of the noise energy of each IMF at the 95% and 99% confidence levels and

[0088]

[0089] where the β and ρ corresponding to the 95% and 99% confidence levels are β = 0.719, ρ = 1.919 and β = 0.719, ρ = 2.449 respectively.

[0090] Step S1.4: Select the pure-noise IMFs. If then the j-th IMF can be used as the pure-noise IMF;

[0091] Step S1.5: Re-group the selected pure-noise IMFs and denote them as {ω l (t), l = 1, 2, …}, and then calculate the thresholds of each pure-noise IMF, denoted as {τ l , l = 1, 2, …}. The adaptive noise threshold estimation method based on the kurtosis factor is as follows;

[0092]

[0093] where σ j represents the w j(i) The mean absolute deviation of the median calculation, median(w j ) represents the median of the w j (i) component, N represents the length of the j-th IMF, Peak represents the maximum absolute value of the signal, and RMS represents the root mean square value of the signal.

[0094] Step S1.6: Use the semi-soft threshold technique to separate the estimated noise from the above pure noise IMFs The mathematical expression of the semi-soft threshold technique is as follows;

[0095]

[0096] where x(i) represents the original signal, sgn(x) is the sign function, τ is the noise threshold, and α is the dynamic adjustment factor.

[0097] Step S1.7: Random permutation to obtain the m-th version of the estimated noise Then calculate the m-th version of the reconstructed signal

[0098] Step S1.8: Perform EMD on x m (t) to obtain the m-th version of the IMFs and the residual component, denoted as {ω mj (t), 1 ≤ j ≤ n, 1 ≤ t ≤ N} and r mn ;

[0099] Step S1.9: According to the stopping criterion, calculate the mean square error ε between ω m,j (i) and ω m+1,j (i) obtained from two iterations:

[0100]

[0101] When the mean square error ε between two iterations is less than or equal to the set value ε threshold , stop the iteration, and the stopping condition is:

[0102] ε ≤ ε threshold (6)

[0103] Step S1.10: Calculate the mean of all IMFs and the residual component after each iteration, and use this as the final IMFs and the residual component, denoted as and

[0104]

[0105] Step S1.11: Select the IMFs rich in fault information from the IMFs obtained in step S1.10 according to the IMF selection method based on the cross-correlation coefficient, denoted as {ω s (t), s = 1, 2, …}, and the IMF selection method based on the cross-correlation coefficient is expressed as follows;

[0106]

[0107] where CC j is a variable to measure the correlation degree between the original signal and the IMFs, are the means of the original signal x(i) and the jth IMF respectively, and Kurtosis IMFs is the kurtosis value of each IMF, and Kurtosis Normal is the kurtosis value of the healthy signal, usually 3.

[0108] Step S1.12: Recalculate the noise thresholds of the IMFs selected in step (11) using the adaptive noise threshold estimation method based on the kurtosis factor, denoted as {τ s , s = 1, 2, …};

[0109] Step S1.13: According to the noise thresholds obtained in step S1.12, use the semi-soft threshold technology to denoise the IMFs selected in step S1.11 to obtain the denoised IMFs rich in fault information, denoted as

[0110] Step S1.14: Reconstruct the signal

[0111] The aforementioned step S2 specifically includes the following steps:

[0112] Step S2.1: Slide-window overlapping sampling is performed on the denoised signal with a length of N by taking every m data points as a group, and the overlapping rate is r, and each sample contains m sampling points.

[0113] Therefore, the number of samples obtained is

[0114] Step S2.2: Use the Hanning window as the window function to perform STFT on each sample respectively, extract the corresponding time-frequency diagram, and the window width is set to 256.

[0115] The steps of the short-time Fourier transform are as follows:

[0116] STFT divides each sample signal into many short time segments by using the window function, and performs Fourier transform on each short time segment respectively, realizing the effective extraction of the time-domain and frequency-domain information of the non-stationary vibration signal.

[0117] The window function is denoted as ω(t). To transform the window function, a time variable τ is introduced to represent the center position of the current window, and the transformed window function is denoted as ω(t - τ).

[0118] The Fourier transform formula is as follows:

[0119]

[0120] Among them, x(t) is a one-dimensional time-domain signal, ω is the frequency variable, and exp(-jwt) represents the complex conjugate operation.

[0121] Based on the above process, the STFT result of the noise-reduced signal sample can be given by the following integral:

[0122]

[0123] Among them, X(t, f) represents the amplitude value at time t and frequency f.

[0124] The experimental data used in this embodiment is the bearing dataset of Case Western Reserve University (CWRU), USA. The experimental setup is as Figure 2 shown, consisting of a 1.5KW motor (left), a torque sensor (middle), and a power meter (right). The torque sensor adjusts the motor speed by adjusting the load. Accelerometers are installed at the drive end (DE) and the fan end (FE) to collect the operating data of different positions of the bearing. The bearing model used is SKF6205. By performing electrical discharge machining (EDM) on the bearing, two fault types, inner race pitting and outer race pitting, are introduced. Each fault type is further divided into three severity levels (fault diameter: 0.007 inches, 0.014 inches, 0.021 inches). Bearings with different defects are respectively installed on the test bench, and vibration data is collected under constant loads of 0 horsepower (0hp), 1 horsepower (1hp), 2 horsepower (2hp), and 3 horsepower (3hp). The corresponding motor speeds are 1797, 1772, 1750, and 1730 RPM respectively. The data collected in each test is the acceleration in the vertical direction at both the drive and fan ends, and the sampling frequencies are 12KHz and 48KHz respectively. In this experiment, the inner and outer race faults with a fault diameter of 0.014 inches at the drive end and the original vibration signal data in the normal state at a speed of 1772 RPM are selected, and the first 10s are selected as the experimental data. The original signals of different health states of the bearing are as Figure 3 shown.

[0125] First, perform ASTENEMD on the signals of different health states respectively. The final obtained IMFs and residuals after decomposition are as Figure 4As shown, 15, 18, and 17 IMFs are obtained after decomposing the three signals respectively. The more IMFs obtained after decomposition indicates that more information is contained, and the degree of decomposition of each state signal is slightly different. From the decomposition results, the first few IMFs obtained after decomposition are consistent or similar to the original signal waveform, indicating that this part contains the main fault information. Usually, these IMFs containing the main fault information are low-frequency components, with no rapid fluctuations in their waveforms and the amplitudes remaining stable, while the high-frequency components are mainly composed of noise and some minor vibration components. Compared with the original signal, obvious periodic pulses appear in the low-frequency IMFs, and these periodic pulses correspond to different fault characteristics.

[0126] Then, the most effective IMFs are selected from all the decomposed IMFs using the IMFs selection method based on the cross-correlation coefficient. The relationships between each IMF, the cross-correlation coefficient, and the kurtosis value under different health states are as Figure 5 shown.

[0127] The selection results of IMFs rich in fault information for signals of different health states of the bearing are shown in Table 1.

[0128] Table 1 Selection results of IMFs rich in fault information

[0129]

[0130] To obtain the denoised signal, semi-soft threshold denoising needs to be performed on the selected IMFs. The IMFs before and after denoising are as Figure 6 shown. Table 2 gives the noise thresholds of IMFs for different fault types.

[0131] Table 2 Noise thresholds of IMFs for different fault types

[0132]

[0133] It can be seen from the results that the higher the cross-correlation coefficient of a component, the larger its threshold, and the noise components in the IMFs rich in fault information are removed more effectively. While removing noise, semi-soft threshold denoising can retain the key signal components, especially the periodic fault information in the low-frequency IMFs, and better retain the original characteristics of the signal, especially in the presence of fault information such as mutations, periodic vibrations, or impacts.

[0134] Finally, the denoised IMFs are added together to finally obtain the denoised signal, Figure 7 which shows the time-domain waveforms before and after denoising. It is difficult to clearly see the denoising effect only from the time-domain graph, Figure 8 which shows the frequency-domain waveforms before and after denoising. Next, the time-domain signal will be converted into a frequency-domain signal for analysis. From Figure 8It can be clearly found that after ASTENEMD processing, while retaining the main frequency components of the original signal, the noise interference is suppressed to a certain extent, and the frequency waveform diagram becomes smoother.

[0135] Through the above adaptive semi-soft threshold set noise reconstruction empirical mode decomposition, the noise reduction processing of the original vibration signal is carried out, and the noise reduction signals in different healthy states of the bearing are obtained.

[0136] Step S2, use STFT to convert the noise-reduced signal into a time-frequency image to achieve time-frequency feature extraction.

[0137] The noise-reduced signal is segmented using a sliding window with a size of m = 1024 and an overlap rate of r = 0.5. Each segmented sample contains 1024 data points.

[0138] The Hanning window is used as the window function, and the STFT is performed on the segmented samples to extract the corresponding time-frequency diagram. The window width is set to 256. The steps of the short-time Fourier transform are as follows:

[0139] STFT divides each sample signal into many short time periods by using the window function, and performs the Fourier transform on each short time period respectively, realizing the effective extraction of the time-domain and frequency-domain information of the non-stationary vibration signal.

[0140] The window function is expressed as ω(t). To transform the window function, a time variable τ is introduced to represent the center position of the current window. The transformed window function is expressed as ω(t - τ).

[0141] The Fourier transform formula is as follows:

[0142]

[0143] Among them, x(t) is a one-dimensional time-domain signal, w is the frequency variable, and exp(-jwt) represents the complex conjugate operation.

[0144] Based on the above process, the STFT result of the noise-reduced signal sample can be given by the following integral:

[0145]

[0146] Among them, X(t,f) represents the amplitude value at time t and frequency f.

[0147] The systems, devices, modules or units illustrated in the above embodiments may be specifically implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0148] Computer-readable media includes both permanent and non-permanent, removable and non-removable media and can be implemented by any method or technology for storing information. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media such as modulated data signals and carrier waves.

[0149] It should also be noted that the term "comprises," "comprising," or any other variation thereof is intended to cover a non-exclusive inclusion, such that a process, method, commodity, or device that comprises a series of elements includes not only those elements but also other elements not expressly listed, or elements that are inherent to such process, method, commodity, or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, commodity, or device that comprises the element.

[0150] The above embodiments should be understood as being illustrative of the present invention only and not limiting the scope of protection of the present invention. After reading the content described in the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent changes and modifications also fall within the scope defined by the claims of the present invention.

Claims

1. A rolling bearing fault signal denoising and feature extraction method based on ASTENEMD and STFT, characterized in that: The following steps are involved: - Collect the original vibration signal of rolling bearing; - Use adaptive semi-soft threshold ensemble noise reconstruction empirical mode decomposition (ASTENEMD) to reduce noise to obtain pure IMFs and residual components; -The denoised signal is converted into a time-frequency image through short-time Fourier transform (STFT) to extract time-frequency features.

2. The rolling bearing fault signal denoising and feature extraction method based on ASTENEMD and STFT according to claim 1 is characterized in that: The ASTENEMD method includes: - Use EMD to decompose the original vibration signal to obtain a series of intrinsic modal components IMFs and residual components; respectively denoted as {ω j (t),1≤j≤n,1≤t≤N} and r n ; Where n represents the number of IMFs, and N represents the length of the original time domain signal; - Calculate the noise energy of IMFs and determine confidence intervals to identify pure noise IMFs; - Adopting adaptive noise threshold estimation based on kurtosis factor to denoise pure noise IMFs; -Reconstruct the denoised signal.

3. The rolling bearing fault signal denoising and feature extraction method based on ASTENEMD and STFT according to claim 2 is characterized in that: The step of calculating the noise energy of IMFs and determining the confidence interval to identify pure noise IMFs specifically includes: Calculate the noise energy of each IMFs separately, denoted as {E j ,1≤j≤n}, the noise energy calculation formula is as follows: where median(|ω j |) means |w j (i) |Median of the components; Calculate the confidence interval of each IMFs noise energy at 95% and 99% confidence levels respectively and in The β and ρ corresponding to the confidence levels of 95% and 99% are β=0.719, ρ=1.919 and β=0.719, ρ=2.449 respectively.

4. The rolling bearing fault signal denoising and feature extraction method based on ASTENEMD and STFT according to claim 3 is characterized in that: The method of using an adaptive noise threshold estimation based on a kurtosis factor to perform denoising on pure noise IMFs specifically includes: Choose pure noise IMFs if Then the jth IMF can be used as a pure noise IMF; The selected pure noise IMFs are regrouped and recorded as {ω l (t), l=1,2,…}, and then calculate the threshold of each pure noise IMFs, denoted as {τ l ,l=1,2,…}, the adaptive noise threshold estimation method based on kurtosis factor is as follows; where σ j Indicated by w j (i) The mean absolute deviation of the median calculation of the components, median(w j ) indicates w j (i) The median of the component, N represents the length of the jth IMF, CF represents the crest factor of the signal, Peak represents the maximum absolute value of the signal, and RMS represents the root mean square value of the signal; The estimated noise is separated from the pure noise IMFs mentioned above using a semi-soft thresholding technique. The mathematical expression of semi-soft thresholding technique is as follows; Where x(i) represents the original signal, sgn(x) is the sign function, τ is the noise threshold, and α is the dynamic adjustment factor; Random Permutation Get the mth version of the estimated noise Then calculate the mth version of the reconstructed signal x m (t) Execute EMD to obtain the mth version of IMFs and residual components, which are denoted as {ω mj (t),1≤j≤n,1≤t≤N} and r mn ; According to the stopping criterion, calculate the ω obtained from two iterations m,j (i) and ω m+1,j (i) The mean square error ε between When the mean square error ε between two iterations is less than or equal to the set value ε threshold The iteration stops when , and the stopping condition is: ε≤ε threshold (6) Calculate the mean of all IMFs and residual components after each iteration and use it as the final IMFs and residual components, denoted as and According to the IMFs selection method based on the mutual correlation coefficient, the IMFs rich in fault information are selected from the obtained IMFs, denoted as {ω s (t), s=1,2,…}, the IMFs selection method based on the mutual correlation coefficient is expressed as follows; CC j To measure the correlation between the original signal and IMFs, are the original signal x(i) and the mean of the j-th IMFs, Kurtosis IMFs is the kurtosis value of each IMFs, Kurtosis Normal is the kurtosis value of the healthy signal, usually 3; The noise threshold of the selected IMFs is recalculated using the adaptive noise threshold estimation method based on the kurtosis factor, denoted as {τ s ,s=1,2,…}; According to the obtained noise threshold, the semi-soft threshold technology is used to reduce the noise of the selected IMFs to obtain the denoised IMFs rich in fault information, which is denoised as Reconstructing the signal 5. The rolling bearing fault signal denoising and feature extraction method based on ASTENEMD and STFT according to claim 4 is characterized in that: The de-noised signal is converted into a time-frequency image by short-time Fourier transform (STFT) to extract time-frequency features, which specifically includes the following steps: The noise reduction signal of length N is subjected to sliding window overlapping sampling with m data points as a group, the overlapping rate is r, and each sample contains m sampling points; Therefore, the number of samples obtained is The Hanning window is used as the window function to perform STFT on each sample and extract the corresponding time-frequency diagram. The window width is set to 256.

6. The rolling bearing fault signal denoising and feature extraction method based on ASTENEMD and STFT according to claim 4, characterized in that: The steps of the short-time Fourier transform are as follows: STFT divides each sample signal into many short time periods by using a window function, and performs Fourier transform on each short time period, thereby effectively extracting the time domain and frequency domain information of the non-stationary vibration signal. The window function is expressed as ω(t). In order to transform the window function, the time variable τ is introduced to represent the center position of the current window. The transformed window function is expressed as ω(t-τ). The Fourier transform formula is as follows: Where x(t) is a one-dimensional time domain signal, ω is a frequency variable, and exp(-jwt) represents a complex conjugate operation; Based on the above process, the STFT result of the denoised signal sample can be given by the following integral: Where X(t,f) represents the amplitude value at time t and frequency f.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the rolling bearing fault signal denoising and feature extraction method based on ASTENEMD and STFT as described in any one of claims 1 to 6 is implemented.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the rolling bearing fault signal denoising and feature extraction method based on ASTENEMD and STFT as described in any one of claims 1 to 6 is implemented.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the rolling bearing fault signal denoising and feature extraction method based on ASTENEMD and STFT as described in any one of claims 1 to 6 is implemented.

Images