A Bearing Compound Fault Identification Method Based on Wavelet Transform, Second-Order Difference and 1D-LBP

Abstract

A bearing compound fault identification method based on wavelet transform, second-order difference and 1D-LBP belongs to the technical field of fault diagnosis. The optimal decomposition level of discrete wavelet transform is determined by the correlation coefficient, and the original signal is subjected to discrete wavelet transform according to the determined optimal decomposition level to obtain a series of component signals. The impact characteristics during bearing faults are described based on the second-order difference signals of each component signal after discrete wavelet transform. 1D-LBP is performed on the second-order difference signals with the moving window local average as the criterion to obtain a new signal containing local feature information. Based on the spectral analysis method, the obtained local feature signal is analyzed, its characteristic frequency is extracted, and the extraction of bearing fault feature information and the identification of fault types are realized according to the bearing fault characteristic frequency calculation formula. The invention not only effectively suppresses the noise components in the vibration signal, but also can effectively monitor the running state of the bearing and accurately identify its compound fault types.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault diagnosis, and particularly relates to a bearing compound fault identification method based on the combination of wavelet transform, second-order difference and 1D-LBP. Background Art

[0002] Bearings are widely used in industries, aviation and other fields. They are very important components in rotating machinery, and their health status will directly affect the operation safety of the entire mechanical system. Therefore, it is of great significance to monitor the state and diagnose faults of bearings to ensure the safe and stable operation of equipment. The operating environment of bearings is complex, and the collected vibration signals usually exhibit non-linearity and non-stationarity. Moreover, fault information is easily submerged by irrelevant components and noise, and fault features are often difficult to be accurately extracted. At the same time, in actual engineering, the fault forms of bearings often show compound faults with multiple faults existing simultaneously. Compared with single faults, due to the mutual coupling between faults, the signals corresponding to compound faults are more complex, which further increases the difficulty of fault identification. Therefore, how to accurately extract compound fault features from vibration signals remains a major challenge in the field of bearing fault diagnosis.

[0003] To achieve the accurate extraction of bearing compound fault features and the accurate identification of fault types, a variety of time and frequency domain signal analysis methods have been applied to the field of bearing fault identification. These include research on signal decomposition algorithms such as wavelet transform, empirical mode decomposition, variational mode decomposition, etc.; research on various feature enhancement algorithms such as minimum entropy deconvolution, Teager energy operator, etc.; research on various noise reduction algorithms such as singular value decomposition noise reduction, autocorrelation function noise reduction, and wavelet threshold noise reduction. These signal analysis methods have their own advantages, but also have certain limitations. Especially when extracting features from bearing vibration signals, due to the very complex information contained in the signals, problems such as mode mixing, endpoint effect, need to determine prior parameters, poor anti-noise performance, and difficulty in fully extracting features will occur.

[0004] The above research mainly characterized the bearing fault information from the perspective of time-frequency domain signal analysis. In recent years, some scholars have introduced texture feature extraction methods in the field of image processing into the field of signal processing. Among them, the one-dimensional LBP (one dimension local binary pattern, 1D-LBP) algorithm developed on the basis of the local binary pattern (LBP) is mainly applied to the fields of speech signals and human physiological signals in medicine, and there is relatively little research in the field of fault recognition. However, it shows good local feature extraction ability and helps to extract local fault information in the signal. The 1D-LBP algorithm encodes the local binary signal of the one-dimensional signal into decimal numbers to obtain a series of signals reflecting the local characteristics of the signal within the range of 0-255, thereby representing the feature information in the signal. At present, the research on the 1D-LBP method in the field of fault recognition is mostly directly based on the original signal; due to the influence of component signals and noise unrelated to faults, the fault information in the original signal is weak and complex; in addition, the 1D-LBP method itself is sensitive to noise, which is very likely to lead to insufficient extraction of local features and difficult accurate identification of faults. Summary of the Invention

[0005] To solve the above problems, the technical solution adopted by the present invention is: a bearing compound fault recognition method based on wavelet transform, second-order difference and 1D-LBP, including the following steps;

[0006] Obtain the bearing vibration signal;

[0007] Determine the optimal decomposition level of the discrete wavelet transform based on the correlation coefficient;

[0008] With the obtained optimal decomposition level, perform discrete wavelet transform on the bearing vibration signal to obtain the approximation signal and the detail signal,

[0009] Perform second-order difference operations on the approximation signal and the detail signal to obtain the corresponding second-order difference sequences respectively, and describe the impact characteristics during bearing faults;

[0010] Obtain any one of the second-order difference sequences, and use the local average value of the moving window of the second-order difference sequence as the criterion, and adopt the one-dimensional local binary pattern method to obtain the decimal signal of the second-order difference sequence, which characterizes the local feature information of the bearing vibration signal;

[0011] Perform noise reduction processing on the decimal signal of the obtained second-order difference sequence to obtain the signal after noise reduction;

[0012] For the obtained denoised signal, a spectral analysis method is adopted to extract the prominent frequencies in the spectrum, and the bearing fault characteristic information is extracted and the fault type is identified according to the bearing fault characteristic frequency calculation formula.

[0013] Further, the process of determining the optimal decomposition level is as follows:

[0014] Let X t be the signal to be decomposed, and H t be a component signal after discrete wavelet transform decomposition. Then the correlation coefficient between the two is ρ XH , which can be obtained by formula (1):

[0015]

[0016] where Cov(X t , H t ) is the covariance of X t and H t , and Var(·) represents the variance of the signal;

[0017] Adaptively determine the decomposition level of the discrete wavelet transform (DWT). When reaching the Kth layer, if the correlation coefficient between the component signal and the original signal is less than the correlation coefficient threshold, determine the actual decomposition level as K - 1, that is, update the optimal decomposition level K = K - 1.

[0018] Further, the following second-order difference operation is performed on the detail signal and the approximation signal obtained by decomposition. The second-order difference sequence F k (n i ) of the detail signal d dk (n i ) is expressed as follows:

[0019] F dk (n i ) = d k (t j + 1) - 2d k (t j ) + d k (t j - 1)t j = 2, 3, …, n - 1; n i = t j - 1(2)

[0020] The second-order difference sequence F K (t i ) of the approximation signal a ak (t i ) is expressed as follows:

[0021] Fak (n i ) = a k (t j +1) - 2a k (t j ) + a k (t j -1)t j = 2, 3, …, n - 1; n i = t j -1(3)

[0022] Furthermore, the process of obtaining any second-order difference sequence and using the local average value of the second-order difference sequence moving window as a criterion to obtain the decimal signal of the second-order difference sequence by using the one-dimensional local binary pattern method is as follows:

[0023] Taking the second-order difference sequence F dk (n i ) as an example;

[0024] (a) Let P ji (i = 1, 2, … 8) be the 8 adjacent points of the jth moving window, and calculate the local average value Ed ji of P kj :

[0025]

[0026] (b) Taking the local average value Ed kj as a criterion, binarize the second-order difference sequence F dk (n i ) based on the 1D-LBP algorithm to obtain the 1D-LBP signal;

[0027] Let the binarized sequence corresponding to the jth moving window be F(g ji ), where g ji is the difference between the value of the ith point P i in the jth moving window and the local average value Ed kj . Then the calculation process is shown in formulas (5)-(6).

[0028] g ji = P ji - Ed kj i = 1, 2, …, 8 j = n - 7(5)

[0029]

[0030] (c) Convert the locally binarized sequence F(g ji ) into the decimal signal 1D-LTS(j) that can reflect the signal texture characteristics in sequence.

[0031]

[0032] Further, the window length of the moving window of the moving window is 8, and the moving step size is 1.

[0033] Further, the noise reduction process uses the autocorrelation function for noise reduction.

[0034] A bearing compound fault identification device based on wavelet transform, second-order difference and 1D-LBP, comprising:

[0035] An acquisition module: used to acquire bearing vibration signals;

[0036] A decomposition layer number module: used to determine the optimal decomposition layer number of discrete wavelet transform based on the correlation coefficient;

[0037] A discrete wavelet transform module: used to perform discrete wavelet transform on the bearing vibration signal with the obtained optimal decomposition layer number to obtain an approximation signal and a detail signal.

[0038] A second-order difference operation module: used to perform second-order difference operations on the approximation signal and the detail signal to respectively obtain corresponding second-order difference sequences, and describe the impact characteristics during bearing faults.

[0039] A signal quantization and decimal signal obtaining module: used to obtain any one of the second-order difference sequences, and use the one-dimensional local binary pattern method with the local average value of the moving window of the second-order difference sequence as the criterion to obtain the decimal signal of the second-order difference sequence, which characterizes the local characteristic information of the bearing vibration signal.

[0040] A noise reduction module: used to perform noise reduction processing on the decimal signal of the obtained second-order difference sequence to obtain a noise-reduced signal.

[0041] An extraction and identification module: used to adopt a spectral analysis method for the obtained noise-reduced signal, extract the prominent frequencies in the spectrum, and realize the extraction of bearing fault characteristic information and the identification of fault types according to the bearing fault characteristic frequency calculation formula.

[0042] A computer device, comprising: a processor and a memory, the memory stores program modules, and is characterized in that the program modules run on the processor to implement the method as described above.

[0043] A readable storage medium stores program modules, and the program modules can implement the method when running on a processor.

[0044] A bearing compound fault identification method based on the combination of wavelet transform, second-order difference and 1D-LBP provided by the present invention takes into account that wavelet transform can reduce the influence of noise on the extraction of weak fault information to a certain extent through frequency-domain filtering; at the same time, considering that bearing faults are often accompanied by impact characteristics; the second-order difference of the signal reflects the speed of change of the signal slope, that is, the more obvious the impact characteristics of the signal at the position where the absolute value of the second-order difference of the signal is larger. Therefore, the impact characteristics during bearing faults can be described by using the second-order difference of the signal; in addition, quantifying the second-order difference of the decomposed component signals (instead of the entire vibration signal) by the 1D-LBP method, rather than binarizing the signal itself, can further improve the ability of 1D-LBP to identify weak fault characteristics and the robustness to noise.

[0045] Considering that the correlation coefficient can reflect the correlation between signals, the decomposition level of the discrete wavelet transform is adaptively determined based on the correlation coefficient between the component signals after discrete wavelet transform and the original signal; secondly, considering the characteristics that bearing faults are often accompanied by impact characteristics, and using the sensitivity of the second-order difference signal to impact signals, the impact characteristics during bearing faults are described based on the second-order difference signals of each component signal after discrete wavelet transform; and 1D-LBP is performed on the second-order difference signal based on the local average value to obtain a new signal containing local feature information.

[0046] On this basis, a method combining wavelet transform, second-order difference and 1D-LBP is proposed, which not only effectively suppresses the noise components in the vibration signal, but also highlights the fault feature information; it realizes the effective monitoring of the bearing operation state from the perspective of "texture feature" extraction, accurately extracts the bearing fault feature information, and accurately judges the compound fault type of the bearing, having certain engineering application value.

[0047] (1) Considering that the discrete wavelet transform has good noise reduction effect, the accurate determination of its decomposition level is an important prerequisite for effective signal decomposition; the correlation coefficient reflects the correlation between signals, and the decomposition level of the discrete wavelet transform is adaptively determined by using the correlation coefficient.

[0048] (2) Considering the characteristics that bearing faults are often accompanied by impact characteristics, and using the sensitivity of the second-order difference signal to impact signals, the impact characteristics during bearing faults are described based on the second-order difference signals of each component signal after discrete wavelet transform.

[0049] (3) To further highlight the impact characteristics during bearing faults, 1D-LBP is performed on the second-order difference signal based on the local average value to obtain a new signal containing local information.

[0050] (4) Use the spectral analysis method to extract features from the obtained signal containing local texture information, so as to extract the bearing fault characteristic frequency and accurately identify the fault type. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0052] Figure 1 Block diagram of the method for adaptively determining the decomposition level of discrete wavelet transform in the embodiment of the present invention;

[0053] Figure 2 Block diagram of the bearing compound fault identification method based on wavelet transform, second-order difference and 1D-LBP in the embodiment of the present invention;

[0054] Figure 3 Application of the embodiment of the present invention in the identification of intermediate bearing faults, where (a) is the time domain of the simulation signal, (b) is the time domain of the noise signal, (c) is the mixed signal of (a) and (b), (d) is the spectrogram of (c), (b1) is the component signal after DWT of (c), (b2) is the 1D-LTS signal of (b1), (c1) is the autocorrelation function diagram of D5 in (b2), (d1) is the spectrum of (c1), (c2) is the autocorrelation function diagram of A5 in (b2), (d2) is the spectrogram of (c2), and (d3) is the local enlarged view of (d2). DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other. The following will refer to the drawings and combine the embodiments to detail the present invention.

[0056] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and in no way limits the present invention and its application or use. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0057] A bearing compound fault identification method based on wavelet transform, second-order difference and 1D-LBP, characterized by comprising the following steps:

[0058] Step 1: Obtain the bearing vibration signal;

[0059] Step 2: Based on the correlation coefficient, determine the optimal decomposition level K of the discrete wavelet transform (DWT) of the bearing vibration signal x(t i ), where t i = 1, 2, …, N, and N is the signal length;

[0060] Step 3: With the obtained optimal decomposition level, perform discrete wavelet transform on the bearing vibration signal to obtain the approximation signal a K (t i ) and the detail signal d k (t i ), where k = 1, 2, … K, and t i = 1, 2, …, n;

[0061] Step 4: Randomly perform second-order difference operation on the approximation signal a K (t i ) or the detail signal d k (t i ) to obtain the corresponding second-order difference sequence F dk (n i ) or F ak (t i ) to describe the impact characteristics during bearing faults;

[0062] Step 5: Obtain any one of the second-order difference sequences. Based on the local average value of the moving window of the second-order difference sequence, use the one dimension local binary pattern (1D-LBP) to obtain the decimal signal 1D-LTS(j) of the second-order difference sequence, where j = 1, 2, …, n - 7, for describing the local feature information of the bearing vibration signal; the window length of the moving window is 8, and the moving step is 1;

[0063] Step 6: To further reduce the influence of noise, perform noise reduction processing on the obtained decimal signal of the second-order difference sequence to obtain the noise-reduced signal DE-1D-LTS;

[0064] Step 7: For the obtained noise-reduced signal DE-1D-LTS, use the spectral analysis method to extract the prominent frequencies in the spectrum, and based on the bearing fault characteristic frequency calculation formula, realize the extraction of bearing fault characteristic information and the identification of fault types.

[0065] Steps 1 / Step 2 / Step 3 / Step 4 / Step 5 / Step 6 / Step 7 are executed in sequence;

[0066] Furthermore, the determination process of the correlation coefficient is as follows:

[0067] Let X t be the signal to be decomposed, and H t be a component signal after being decomposed by discrete wavelet transform. Then the correlation coefficient between the two is ρ XH , which can be obtained from formula (1):

[0068]

[0069] where Cov(X t , H t ) is the covariance of X t and H t , and Var(·) represents the variance of the signal;

[0070] Adaptively determine the decomposition level of the discrete wavelet transform. When reaching the Kth layer, if the correlation coefficient between the component signal and the original signal is already less than the correlation coefficient threshold, the actual decomposition level is K - 1, that is, update the optimal decomposition level K = K - 1. As Figure 1 shown. Among them, D1, D2... Dk respectively represent the detail signals of the 1st to Kth layers of the DWT, and Ak represents the approximation signal of the DWT.

[0071] The value of the correlation coefficient threshold is 0.2 (taking 0.2 as an example, not limited to this value).

[0072] Furthermore, the detail signal d k (n i ) performs the following second-order difference operation. The second-order difference sequence F dk (n i ) of the detail signal d k (n i ) has the following expression:

[0073] F dk (n i ) = d k (t j + 1) - 2d k 9t j ) + d k (t j - 1)t j = 2, 3,..., n - 1; n i = t j - 1(2)

[0074] The approximation signal a K (t i) second-order difference sequence F ak (t i ) is expressed as follows:

[0075] F ak (n i ) = a k (t j + 1) - 2a k (t j ) + a k (t j - 1)t j = 2, 3, …, n - 1; n i = t j - 1(3)

[0076] The process of obtaining any second-order difference sequence and getting the decimal signal of the second-order difference sequence by using the one-dimensional local binary pattern method based on the local average value of the second-order difference sequence moving window is as follows:

[0077] Taking the second-order difference sequence F dk (n i ) as an example;

[0078] (a) Let P ji (i = 1, 2, … 8) be the 8 adjacent points of the jth moving window, and calculate the local average value Ed ji of P kj :

[0079]

[0080] (b) Based on the local average value Ed kj , binarize the second-order difference sequence F dk (n i ) by using the 1D-LBP algorithm to obtain the 1D-LBP signal;

[0081] Let the binarized sequence corresponding to the jth moving window be F(g ji ), and g ji be the difference between the value of the ith point P i in the jth moving window and the local average value Ed kj . Then the calculation process is shown in formulas (5)-(6).

[0082] g ji = P ji - Ed kj i = 1, 2, …, 8 j = n - 7(5)

[0083]

[0084] (c) The locally binarized sequence F(g ji ) is sequentially reconverted into a decimal signal 1D-LTS(j) that can reflect the signal texture features.

[0085]

[0086] To further reduce the influence of noise, the autocorrelation function of the obtained 1D-LTS signal is taken for noise reduction (not limited to this noise reduction method), and the signal after noise reduction is denoted as DE-1D-LTS.

[0087] According to the spectrum of the signal DE-1D-LTS after noise reduction, the characteristic information of the combined fault of the intermediate bearing is extracted based on the calculation formula of the fault characteristic frequency of the intermediate bearing in Table 1 (taking the case of the high and low pressure rotors rotating in the same direction), and the type of the combined fault of the bearing is judged.

[0088] Table 1 Parameters represented by each symbol and calculation formula of each characteristic frequency of the intermediate bearing

[0089]

[0090]

[0091] Example 1: To prove the effectiveness of this method, it was verified based on simulation signals. Randomly take the combined fault of the outer ring friction cage and the inner ring friction cage of the intermediate bearing as an example for analysis. The low pressure and high pressure speeds are randomly taken as 3500 r / min and 12000 r / min (the low pressure and high pressure rotation directions are the same rotation direction). Take the intermediate bearing with geometric parameters shown in Table 2 as an example for verification.

[0092] Table 2 Geometric parameters of the intermediate bearing

[0093]

[0094] According to the geometric parameters of the intermediate bearing, the high and low pressure rotation frequencies, and the fault characteristic frequency calculation formula shown in Table 1, the characteristic frequencies of the simulation signal in this state are respectively: F H = 200 Hz, F L = 58.3 Hz, f o = 1945.3 Hz, f i = 2304.7 Hz, f b = 831.8 Hz, F or = 64.8 Hz, F ir = 76.8 Hz.

[0095] Let y ) (t) be the intermediate bearing fault simulation signal; A 1 be the amplitude of the inner ring friction cage fault signal (A1 = 0.2); A 2 is the amplitude of the outer ring friction cage fault signal (A 2 = 0.4); F L , F H are the low - voltage and high - voltage rotation frequencies respectively (the amplitudes are represented by A 3 , A 4 respectively; randomly take A 3 = A 4 = 0.6, greater than the amplitude of the fault signal); f n1 is the natural frequency of the system (f n1 = 1000); g is the damping coefficient (g = 0.2); y 0 is the displacement constant (y 0 = 1); N sampling points (N = 8192 * 3); t 0 is the single - cycle sampling time; t is the sampling time; K is the number of repetitions; the sampling frequency is 3.2 KHz. In this embodiment, the signal amplitudes and parameters are all randomly set. Then the intermediate bearing fault simulation signal y 0 (t) is:

[0096]

[0097] Aiming at the problem that there is a large amount of noise in the vibration signals collected in engineering practice, Gaussian white noise with a signal - to - noise ratio of - 20 is added to the simulation signal y 0 (t) during the intermediate bearing fault. The added noise signal is s(t), and the final fault signal y(t) is:

[0098] y(t)= y 0 (t)+ s(t) (9)

[0099] In this embodiment, the threshold for adaptively determining the decomposition level of DWT is taken as 0.2, the noise reduction method is taken as autocorrelation noise reduction, and the spectral analysis is taken as frequency spectrum analysis. However, in the actual application process, it is not limited to this threshold value, spectral analysis method and noise reduction method. Analyzing the simulation signal of the intermediate bearing compound fault according to the method of the present invention, the results are as Figure 3 shown. Figure 3 (a) corresponds to the simulation signal y ) (t); Figure 3 (b) corresponds to the noise signal s(t); Figure 3 (c)-(d) are the intermediate bearing fault signal y(t) after adding noise and its frequency spectrum. Table 3 is the correlation coefficient between the sub - component signals with different decomposition levels and the original signal obtained according to Figure 1 the method described above. It can be seen from Table 3 that the optimal decomposition level of the discrete wavelet transform in this embodiment is 5 (6 - 1 = 5).

[0100] Figure 3 (b1) are the approximate signals A5 and the detail signals D1 - D5 obtained by performing discrete wavelet transform on Figure 3 (c). Figure 3 (b2) shows the decimal signal 1D - LTS obtained by the proposed method. Due to space limitations, only the autocorrelation function and its spectrum of the component signal with the best performance are shown. Figure 3 (c1)-(c2) are the autocorrelation functions of the 1D - LTS corresponding to D5 and A5; Figure 3 (d1)-(d2) are Figure 3 the spectra of (c1)-(c2).

[0101] Table 3 Correlation coefficients between the detail signals after DWT and the original signal - Example 1

[0102] Detail signal D1 D2 D3 D4 D5 D6 Correlation coefficient 0.696 0.501 0.359 0.263 0.201 0.130

[0103] From Figure 3 the analysis of (d1)-(d3), it can be found that after processing the simulation signal by the method proposed in the present invention, the following obvious characteristic frequency components can be observed:

[0104] 1) In Figure 3 (d1)-(d3), there are significant frequencies of 284 Hz (284 / 2 = 142 ≈ 141.6 = (64.8 + 76.8)), 624.7 Hz ((627.7 + 64.8) / 9 = 76.94), and 1424 Hz (1424 / 10 = 142.4 ≈ 141.6 = (64.8 + 76.8)). These frequencies correspond to the characteristic frequencies of the compound fault of the cage friction inner and outer rings, marked as squares.

[0105] 2) In Figure 3 (d2)-(d3), there are prominent frequencies of 1330 Hz ((1330 - 6*200) / 2 = 65.0), 718 Hz (718 / 11 = 65.3), and 1308 Hz (1308 / 20 = 65.4) corresponding to the characteristic frequencies of the cage friction outer ring fault, and these frequencies are marked as diamonds. At the same time, there are prominent frequencies of 1351 Hz ((1351 - 6*200) / 2 = 75.5) and 1381 Hz (1381 / 18 = 76.7) corresponding to the characteristic frequencies of the cage friction inner ring fault, and these frequencies are marked as circles.

[0106] From the analysis of (1)-(2), it can be found that the cage of the intermediate bearing has a compound fault of friction between the cage and the inner ring and between the cage and the outer ring, which is exactly the same as the set fault type. That is to say, the typical frequencies extracted by the present invention are exactly the same as the fault types of the intermediate bearing set in the simulation signal, and can be used to determine the compound fault of the intermediate bearing.

[0107] It can be seen that: According to a fault identification method based on the combination of wavelet transform, second-order difference and 1D-LBP proposed by the present invention, in the spectral analysis of the obtained signal, the noise components are effectively suppressed. Based on the simulation signal, the typical frequency components in the signal spectrum obtained by the method of the present invention mainly show the characteristic frequencies corresponding to the fault types of the intermediate bearing, realizing the accurate identification of the bearing compound fault types. At the same time, the present invention is also applicable to the fault identification of other bearing types and has excellent engineering application value.

[0108] A bearing compound fault identification device based on wavelet transform, second-order difference and 1D-LBP, comprising:

[0109] Acquisition module: used to acquire the bearing vibration signal;

[0110] Decomposition layer number module: used to determine the optimal decomposition layer number of the discrete wavelet transform based on the correlation coefficient;

[0111] Discrete wavelet transform module: used to perform discrete wavelet transform on the bearing vibration signal with the obtained optimal decomposition layer number to obtain the approximate signal and the detail signal,

[0112] Second-order difference operation module: used to perform second-order difference operation on the approximate signal and the detail signal to obtain the corresponding second-order difference sequences respectively, and describe the impact characteristics during bearing faults;

[0113] Signal quantization and decimal signal obtaining module: used to obtain any one of the second-order difference sequences, and use the local average value of the moving window of the second-order difference sequence as the criterion, and adopt the one-dimensional local binary pattern method to obtain the decimal signal of the second-order difference sequence, characterizing the local characteristic information of the bearing vibration signal;

[0114] Noise reduction module: used to perform noise reduction processing on the decimal signal of the obtained second-order difference sequence to obtain the noise-reduced signal;

[0115] Extraction and identification module: used to adopt the spectral analysis method for the obtained noise-reduced signal, extract the prominent frequencies in the spectrum, and realize the extraction of the bearing fault characteristic information and the identification of the fault type according to the bearing fault characteristic frequency calculation formula.

[0116] A computer device, comprising: a processor and a memory, the memory storing program modules, characterized in that the program modules run on the processor to implement the method as described above.

[0117] A readable storage medium stores program modules, and the program modules can implement the method when running in a processor.

[0118] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than limiting them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A bearing composite fault identification method based on wavelet transform, second-order difference and 1D-LBP, characterized in that: The steps include: Obtain bearing vibration signal; Determine the optimal number of decomposition levels of discrete wavelet transform based on the correlation coefficient; The process of determining the optimal number of decomposition levels is as follows: Let X t is the signal to be decomposed, H t is a component signal after discrete wavelet transform decomposition, then the correlation coefficient between the two is ρ XH , can be obtained by formula (1): Among them, Cov(X t ,H t ) is X t and H t The covariance of , Var(·) represents the variance of the signal; Adaptively determine the number of decomposition layers of discrete wavelet transform. When the Kth layer is reached and the correlation coefficient between the component signal and the original signal is less than the correlation coefficient threshold, the actual number of decomposition layers is determined to be K-1, that is, the optimal number of decomposition layers is updated to K=K-1; With the optimal decomposition layer number obtained, discrete wavelet transform is performed on the bearing vibration signal to obtain the approximate signal and the detail signal; Perform second-order difference operations on the approximate signal and the detail signal to obtain the corresponding second-order difference sequences, and describe the impact characteristics of the bearing failure. Get any second-order difference sequence, take the local average value of the moving window of the second-order difference sequence as the criterion, and use the one-dimensional local binary pattern method to get the decimal signal of the second-order difference sequence to characterize the local characteristic information of the bearing vibration signal; Performing noise reduction processing on the decimal signal of the acquired second-order difference sequence to obtain a noise-reduced signal; The spectrum analysis method is used to extract the prominent frequencies in the spectrum of the obtained denoised signal, and the bearing fault characteristic information is extracted and the fault type is identified based on the bearing fault characteristic frequency calculation formula.

2. The bearing composite fault identification method based on wavelet transform, second-order difference and 1D-LBP according to claim 1 is characterized in that: The decomposed detail signal and the approximate signal are operated by the following second-order difference operation. The detail signal d k (n i ) The second-order difference sequence F dk (n i ) is expressed as follows: F dk (n i )=d k (t j +1)-2d k (t j )+d k (t j -1) t j =2,3,…,n-1; n i =t j -1 (2) The approximate signal a K (t i ) is the second-order difference sequence F ak (t i ) is expressed as follows: F ak (n i )=a k (t j +1)-2a k (t j )+a k (t j -1)t j =2,3,…,n-1;n i =t j -1(3)。 3. The bearing composite fault identification method based on wavelet transform, second-order difference and 1D-LBP according to claim 2 is characterized in that: The process of obtaining any second-order difference sequence, taking the local average value of the moving window of the second-order difference sequence as the criterion, and adopting the one-dimensional local binary pattern method to obtain the decimal signal of the second-order difference sequence is as follows: On the second-order difference sequence F of detail signal dk (n i )as follows; (a) Let P ji (i=1,2,…8) are the 8 adjacent points of the jth moving window, calculate P jj The local mean Ed kj : (b) Taking the local average value Ed kj As the criterion, for the second-order difference sequence F dk (n i ) performing binarization based on the 1D-LBP algorithm to obtain a 1D-LBP signal; Let the binarized sequence corresponding to the jth moving window be F(g ji ), g ji is the i-th point P in the j-th moving window i The value of and the local average value Ed kj The calculation process is shown in formulas (5)-(6). g ji =P ji -Ed kj i=1,2,…,8 j=n-7 (5) (c) The locally binarized sequence F(g ji ), and then reconverted into the decimal signal 1D-LTS(j) which can reflect the signal texture characteristics:

4. The bearing composite fault identification method based on wavelet transform, second-order difference and 1D-LBP according to claim 1 is characterized in that: The window length of the moving window is 8, and the moving step length is 1.

5. The bearing composite fault identification method based on wavelet transform, second-order difference and 1D-LBP according to claim 1 is characterized in that: The noise reduction process uses an autocorrelation function to perform noise reduction.

6. A bearing composite fault identification device based on wavelet transform, second-order difference and 1D-LBP, characterized in that: include: Acquisition module: used to obtain bearing vibration signals; Decomposition layer number module: used to determine the optimal decomposition layer number of discrete wavelet transform based on the correlation coefficient; The process of determining the optimal number of decomposition layers is as follows: Let X t is the signal to be decomposed, H t is a component signal after discrete wavelet transform decomposition, then the correlation coefficient between the two is ρ XH , can be obtained by formula (1): Among them, Cov(X t ,H t ) is X t and H t The covariance of , Var(·) represents the variance of the signal; Adaptively determine the number of decomposition layers of discrete wavelet transform. When the Kth layer is reached and the correlation coefficient between the component signal and the original signal is less than the correlation coefficient threshold, the actual number of decomposition layers is determined to be K-1, that is, the optimal number of decomposition layers is updated to K=K-1; Discrete wavelet transform module: used to implement discrete wavelet transform on the bearing vibration signal to obtain the approximate signal and detail signal in order to obtain the optimal decomposition layer number; Second-order difference operation module: used to perform second-order difference operation on the approximate signal and the detail signal, obtain the corresponding second-order difference sequence, and describe the impact characteristics of the bearing failure; Signal quantization and decimal signal acquisition module: used to obtain any second-order difference sequence, take the local average value of the moving window of the second-order difference sequence as the criterion, and use the one-dimensional local binary pattern method to obtain the decimal signal of the second-order difference sequence to characterize the local characteristic information of the bearing vibration signal; Noise reduction module: used to perform noise reduction processing on the decimal signal of the acquired second-order difference sequence to obtain a noise-reduced signal; Extraction and identification module: It is used to extract the prominent frequencies in the spectrum of the acquired noise-reduced signal by using the spectrum analysis method, and to extract the bearing fault characteristic information and identify the fault type according to the bearing fault characteristic frequency calculation formula.

7. A computer device comprising: A processor and a memory, wherein the memory stores a program module, wherein the program module runs on the processor to implement the method according to any one of claims 1 to 5.

8. A readable storage medium storing a program module, characterized in that: The program module is executed in a processor to implement the method according to any one of claims 1 to 5.

Images