Rolling bearing fault feature extraction method based on variable scale multipoint kurtosis deconvolution

Abstract

A rolling bearing fault feature extraction method based on variable scale multipoint kurtosis deconvolution comprises the following steps: a, sampling a rolling bearing vibration acceleration signal to obtain a rolling bearing fault vibration signal; b, constructing a toeplitz autocorrelation matrix; c, constructing a variable-scale multipoint kurtosis deconvolution filter; d, filtering the rolling bearing fault vibration signal by using an optimal deconvolution filter to obtain a fault impact signal; e, performing envelope demodulation processing on the fault impact signal, extracting an envelope of the fault impact signal, and obtaining an envelope spectrum through spectral analysis; and f, judging the fault type of the rolling bearing according to the envelope spectrum. According to the method, the optimal target vector of the deconvolution is searched by constructing the variable-scale multipoint kurtosis index, the optimal filter is constructed, the fault impact signal is extracted through the deconvolution, the fault impact signal is subjected to envelope analysis, the fault feature frequency of the rolling bearing is extracted, and the fault feature of the rolling bearing can be accurately extracted.

Description

Technical Field

[0001] The present invention relates to a method for extracting fault features of rolling bearings based on Variable Scale Multipoint Kurtosis Deconvolution (VSMKD), belonging to the technical field of fault diagnosis. Background Art

[0002] Rolling bearings are one of the essential key components in various rotating mechanical equipment and have been widely used in fields such as mechanical manufacturing, aerospace, and energy production. The operating state of rolling bearings plays a decisive role in the safe and reliable operation of mechanical equipment. However, in the actual operating environment, rolling bearings are in long-term service under complex working conditions such as transient speeds and time-varying loads, inevitably resulting in irreversible damages such as fatigue, wear, and corrosion, directly affecting the safe operation of the equipment and even leading to accidents. Therefore, accurately extracting the fault features contained in the vibration signals of rolling bearings is of great significance for the diagnosis and research of mechanical equipment faults.

[0003] After a local fault occurs in a rolling bearing, the obtained vibration signal is the result of the convolution of the original vibration signal and the transmission path. Deconvolving the signal to restore the source impact signal and extracting fault features is one of the important research ideas for rolling bearing fault diagnosis. Minimum Entropy Deconvolution (MED), Maximum Correlated Kurtosis Deconvolution (MCKD), and Multipoint Optimal Minimum Entropy Deconvolution Adjusted (MOMEDA) are representative methods in this field. The deconvolution process of MED is to design an optimal filter to minimize the entropy value of the filter output signal, thereby highlighting the impact features in the signal. The problem with it is that it ignores the periodic characteristics of fault impacts. MCKD is based on MED to output periodic impact components related to faults through deconvolution, but this method has very strict requirements for parameters and requires accurate pre-setting of parameters such as the fault period and filter length. MOMEDA uses a target vector to define the position of periodic pulses and can obtain the best filter without iteration, which can be applied to the extraction of periodic fault features.

[0004] The vibration signals of rolling bearing faults have significant periodic impact characteristics. Currently, in deconvolution methods, a fixed fault period is used as the basis for periodic selection. For example, the invention patent with the publication number CN118294008A discloses a method for extracting early fault characteristics of rolling bearings, including: inputting the original vibration signal; optimizing the parameters of variational mode decomposition with the envelope entropy as the fitness function; decomposing the original vibration signal by the optimized VMD to obtain the corresponding number of sub-signals IMF; screening out the best sub-signal IMF using the envelope spectrum peak factor as an index; reading the optimal sub-signal IMF obtained in the above steps, and performing noise reduction processing on the IMF through MCKD to further filter out the interference caused by background noise to feature extraction; performing Fourier transform on the envelope after noise reduction to extract fault characteristics. The method for extracting early fault characteristics of rolling bearings in the present invention performs noise reduction processing on it through maximum correlation kurtosis deconvolution, and performs Fourier transform on its envelope sequence, which can greatly increase the amplitude of the corresponding characteristic frequency in the spectrum. However, during the actual operation of rolling bearings, slippage inevitably occurs, resulting in disturbances in the fault impact period; at the same time, double impacts or even multi-impact processes will occur when the rolling bearing passes through the defect area, leading to changes in the fault impact interval. Therefore, the periodic impact of rolling bearing fault signals is actually an approximate periodicity, that is, the fault period has random fluctuations. The random fluctuations of the rolling bearing fault period seriously affect the effect of the deconvolution algorithm. Therefore, considering the random fluctuation characteristics of the fault period, studying the variable-scale deconvolution method is of great significance for accurately extracting rolling bearing fault characteristics. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for extracting rolling bearing fault characteristics based on variable-scale multi-point kurtosis deconvolution to accurately extract rolling bearing fault characteristics in view of the drawbacks of the prior art.

[0006] The problems of the present invention are solved by the following technical solutions:

[0007] A method for extracting rolling bearing fault characteristics based on variable-scale multi-point kurtosis deconvolution, the method includes the following steps:

[0008] a: Sampling the vibration acceleration signal of the rolling bearing by using an acceleration sensor installed on the bearing seat to obtain the rolling bearing fault vibration signal x;

[0009] b: Constructing a Toeplitz autocorrelation matrix X0:

[0010] Using N to represent the data length of the rolling bearing fault vibration signal x, dividing the rolling bearing fault vibration signal x starting from one point, setting the length of each segment of the signal to L, and taking (N - L + 1) segments to form the Toeplitz autocorrelation matrix X0:

[0011]

[0012] Where x i represents the i-th data in the rolling bearing fault vibration signal x, where i = 1, 2, …, N;

[0013] c: Construct a variable-scale multi-point kurtosis deconvolution filter:

[0014] ① Introduce a multi-point D-norm to find the maximum formula:

[0015] VSMKD:

[0016] Where: y is the fault impact signal, is the target vector, the weight at the impact location is 1, and the weights at other locations are 0;

[0017] ② Construct a variable-scale multi-point kurtosis index as the objective function:

[0018]

[0019] Where: is the weight value corresponding to each sampling point from 0 to N - L, y n is the signal amplitude corresponding to each sampling point from 0 to N - L, and the subscript n represents each sampling point of the signal from 0 to N - L;

[0020] ③ Calculate the target vector

[0021] There are random fluctuations in the rolling bearing fault period. Define S i as the array composed of the j-th weight w ij in the i-th fluctuation range, w ij takes values of 0 or 1, and all weights w ij constitute the weight vector These weight vectors of the set are defined as W:

[0022]

[0023] Substitute the weight vector into the objective function to calculate the multi-point kurtosis value, and find the weight vector when the VSMK index reaches the maximum value, which is the target vector

[0024]

[0025] Where: w i , i = 1, 2, … represent the position and weight of the target pulse, which are elements in;

[0026] ④ Substitute the target vector into the constructed multi - point D - norm maximum - value calculation formula to solve for the maximum value, and obtain the expression of the optimal filter f of VSMKD:

[0027]

[0028] d: Filter the rolling bearing fault vibration signal x with the optimal solution convolution filter f, and the obtained deconvolution signal is the fault impact signal y:

[0029]

[0030] e: Perform envelope demodulation processing on the fault impact signal y, extract the envelope of the fault impact signal, and obtain the envelope spectrum through spectrum analysis;

[0031] f: Judge the rolling bearing fault type according to the envelope spectrum.

[0032] The specific method for obtaining the envelope spectrum through spectrum analysis in the above - mentioned rolling bearing fault feature extraction method based on variable - scale multi - point kurtosis deconvolution is as follows:

[0033] ① Perform Hilbert transform on the fault impact signal y to obtain the analytic signal y a (t), and the amplitude of the analytic signal y a (t) is the envelope signal e(t):

[0034] y a (t) = y + j·Hy

[0035]

[0036] where: H represents the Hilbert transform;

[0037] ② Perform fast Fourier transform on the obtained envelope signal e(t) to obtain the spectrum of the envelope signal, that is, the envelope spectrum.

[0038] The specific steps for judging the rolling bearing fault type according to the envelope spectrum in the above - mentioned rolling bearing fault feature extraction method based on variable - scale multi - point kurtosis deconvolution are as follows:

[0039] ① Calculate the characteristic frequencies of various faults according to the geometric dimensions and rotational speed information of the faulty bearing:

[0040]

[0041] where: BPFI is the inner - race fault characteristic frequency; BPFO is the outer - race fault characteristic frequency; BSF is the rolling - element fault characteristic frequency; N b is the number of rolling elements; B d is the rolling - element diameter; Pd is the bearing pitch diameter; φ is the contact angle; f r is the rotational frequency;

[0042] ② Judging the fault type and fault location according to the characteristic frequencies in the envelope spectrum and the characteristic frequencies of various fault components:

[0043] If the envelope spectrum is the characteristic frequency BPFO of the outer race fault and its multiple frequencies, then there is an outer race fault in the rolling bearing; if the envelope spectrum is the characteristic frequency BPFI of the inner race fault and its multiple frequencies, and there are sidebands on both sides of the characteristic frequency, with an interval of the rotational frequency f r of the shaft, then there is an inner race fault in the rolling bearing; if the envelope spectrum is the characteristic frequency BSF of the rolling element fault and its multiple frequencies, then there is a rolling element fault in the rolling bearing.

[0044] In the above rolling bearing fault feature extraction method based on variable scale multi-point kurtosis deconvolution, in the process of calculating the target vector , when the fluctuation range takes a relatively large value, to simplify the algorithm, by finding the positions and weights of the pulses corresponding to the maximum amplitudes in each random fluctuation range, the target vector

[0045]

[0046] where: m ij is any position within the i-th fluctuation range S i .

[0047] The present invention constructs a variable scale multi-point kurtosis index to find the optimal target vector for deconvolution, constructs an optimal filter, then extracts the fault impact signal through deconvolution, performs envelope analysis on the fault impact signal, extracts the fault characteristic frequencies of the rolling bearing, and can accurately extract the fault characteristics of the rolling bearing. Description of the Drawings

[0048] The following further details the present invention in conjunction with the drawings.

[0049] Figure 1 is the flowchart of the rolling bearing fault feature extraction provided by the present invention;

[0050] Figure 2 is the schematic diagram of the weight vector;

[0051] Figure 3(a) - Figure 3(b) are the time domain waveform and envelope spectrum of the outer race fault signal, where Fig. 3(a) is the time domain waveform of the outer race fault signal; Fig. 3(b) is the envelope spectrum of the outer race fault signal;

[0052] Figure 4(a) - Figure 4(b)It is a comparison of the processing results of the VSMKD method and the MOMEDA method in Case 1. Among them, Figure 4(a) is the envelope spectrum of the signal after VSMKD deconvolution; Figure 4(b) is the envelope spectrum of the signal after MOMEDA deconvolution;

[0053] Figure 5(a) - Figure 5(b) It is the time-domain waveform and envelope spectrum of the rolling element fault signal. Among them, Figure 5(a) is the time-domain waveform of the rolling element fault signal; Figure 5(b) is the envelope spectrum of the rolling element fault signal;

[0054] Figure 6(a) - Figure 6(b) It is a comparison chart of the processing results of the VSMKD method and the MOMEDA method in Case 2. Among them, Figure 6(a) is the envelope spectrum of the signal after VSMKD deconvolution; Figure 6(b) is the envelope spectrum of the signal after MOMEDA deconvolution.

[0055] Each symbol in the text is respectively represented as: x is the fault vibration signal of the rolling bearing; h is the transmission path; y is the fault impact signal; e is the random noise; X0 is the Toeplitz autocorrelation matrix; N represents the data length of the rolling bearing fault vibration signal x; L is the length of each segment of the signal; x i represents the i-th data in the rolling bearing fault vibration signal x; is the target vector; VSMK is the variable-scale multi-point kurtosis index as the objective function; is the weight value corresponding to each sampling point from 0 to N-L, y n is the signal amplitude corresponding to each sampling point from 0 to N-L, and the subscript n represents each sampling point of the signal from 0 to N-L; S i is the j-th weight w within the i-th fluctuation range ij constituting the array, is the weight vector; W is the set of these weight vectors ; w i represents the position and weight of the target pulse; f is the optimal filter; y a (t) is the analytic signal; e(t) is the envelope signal; H represents the Hilbert transform; BPFI is the inner race fault characteristic frequency; BPFO is the outer race fault characteristic frequency; BSF is the rolling element fault characteristic frequency; N b is the number of rolling elements; B d is the rolling element diameter; P d is the bearing pitch diameter; φ is the contact angle; f r is the rotational frequency; m ij is any position within the i-th fluctuation range S i ; Specific implementation method

[0056] The present invention provides a method for extracting fault features of rolling bearings based on variable-scale multi-point kurtosis deconvolution. Considering the characteristics of periodic impacts of rolling bearing faults and aiming at the problem of random fluctuations in fault cycles, a variable-scale multi-point kurtosis index is constructed to find the optimal target vector for deconvolution, and an optimal filter is constructed to extract the fault impact signal through deconvolution. Envelope analysis is performed on the obtained deconvolution signal to extract the fault characteristic frequency, realizing the fault diagnosis of rolling bearings.

[0057] This method includes the following steps:

[0058] Step a: Sample the vibration acceleration signal of the rolling bearing using an acceleration sensor installed on the bearing housing to obtain the rolling bearing fault vibration signal x. The original rolling bearing fault vibration signal x can be expressed as:

[0059] x = h * y + e (1)

[0060] Where: x is the original rolling bearing fault vibration signal actually collected during the rolling bearing fault; h is the transmission path; y is the fault impact signal; e is the random noise; * is convolution, and both the transmission path h, the fault impact signal y, and the random noise e are unknown.

[0061] During the acquisition process of the original vibration signal of the rolling bearing, an acceleration sensor and a data acquisition device are required. The specific process is as follows: Install the acceleration sensor on the top or side of the bearing housing by gluing or magnetic attraction, and ensure that the sensor is in close contact with the surface of the bearing housing during installation; Connect the acceleration sensor to the data acquisition system through a cable, ensuring a firm connection to avoid interference or interruption during signal transmission; Check the parameter settings of the acquisition system, such as the sampling frequency, the number of sampling points, etc.; After completing the settings of the acquisition system, start the acquisition system to collect signals. The collected signals can be displayed on the computer screen in real time or stored in storage devices such as hard disks for subsequent analysis and processing.

[0062] The transmission path refers to the physical transfer process of the fault source signal to the sensor. For example, the transmission path of an outer ring fault of a rolling bearing is: ① The rolling element passes through the outer ring defect, resulting in periodic impact vibration; ② The vibration is transmitted to the bearing housing through the contact surface; ③ The vibration is transmitted from the bearing housing to the position where the sensor is installed; ④ The sensor detects the vibration signal.

[0063] Step b: Construct the Toeplitz autocorrelation matrix X0.

[0064] Use N to represent the data length of the rolling bearing fault vibration signal x. Divide the rolling bearing fault vibration signal x starting from one point, set the length of each segment of the signal to L, and take (N - L + 1) segments to form the Toeplitz autocorrelation matrix X0:

[0065]

[0066] where x i represents the i-th data in the rolling bearing fault vibration signal x, where i = 1, 2, …, N.

[0067] After that, by constructing an optimal deconvolution filter f to perform deconvolution filtering on the rolling bearing fault vibration signal x, the fault impact signal y can be restored to the greatest extent:

[0068]

[0069] Step c: Construct a variable-scale multi-point kurtosis deconvolution filter to filter out the noise signal to restore the original impact signal to the greatest extent.

[0070] The VSMKD algorithm is used to extract the periodic impact characteristics when there are random fluctuations in the fault period, construct the best finite impulse response filter, and filter out the noise signal to restore the original impact signal to the greatest extent;

[0071] Essentially, the VSMKD algorithm constructs a filter to restore the original impact signal as much as possible. This algorithm is a deconvolution method for target recognition of variable-position fluctuating periodic pulses, and introduces a multi-point D-norm to find the maximum formula, which is defined as:

[0072]

[0073] where: is the target vector. Since there are random fluctuations in the fault period during the operation of the rolling bearing, that is, the position of the impact is variable, the variable position indicates that the position of the target impact is variable. This vector is composed of 0 and 1, and the weight of the position where the impact is located is 1, and the weights of the other positions are 0.

[0074] VSMKD represents variable-scale multi-point kurtosis deconvolution. The above formula indicates that the VSMKD method is a process of solving the optimal deconvolution filter based on the multi-point D-norm (MDN, Multi D-Norm), and the process of solving the optimal filter is transformed into finding the maximum value of the multi-point D-norm.

[0075] The fault impact signal y and the optimal deconvolution filter f can be rewritten according to formula (3) as:

[0076]

[0077] The multi-point kurtosis can convert the target vector at the controlled position into periodic pulse characteristics, so as to more effectively identify the periodic impact components related to faults in the vibration signal. Considering that there are random fluctuations in the fault period T when the rolling bearing fails, an optimization is carried out within the fluctuation range of the fault period T to determine the optimal corresponding to the fault pulse position Construct the Variable Scale Multipoint Kurtosis (VSMK) index as the objective function:

[0078]

[0079] is the weight value corresponding to each sampling point from 0 to N - L, y n is the signal amplitude corresponding to each sampling point from 0 to N - L, and the subscript n represents each sampling point of the signal from 0 to N - L.

[0080] Calculate the target vector

[0081] The target vector is a one-dimensional array composed of the variable position weights of the deconvolution target pulse. The weight at the impact position is 1, and the weights at other positions are 0, which is used to control the separation of the deconvolution pulse signal. The length of the target vector is the same as the length of the output signal, which is (N - L). Due to the random fluctuation of the rolling bearing fault cycle, define S i as the array composed of the jth weight w ij in the ith fluctuation range, and w ij takes values of 0 or 1. The schematic diagram of the weight vector is as shown in Figure 1 . Combine all the weights w ij to form the weight vector These weight vectors The set of is defined as W:

[0082]

[0083] Substitute the weight vector into Equation (6) to calculate the multipoint kurtosis value, and find the weight vector when the VSMK index reaches the maximum value which is the target vector

[0084]

[0085] In the formula: w i , i = 1, 2,... represent the position and weight of the target pulse, which are elements in, the weight at the impact position is 1, and the weights at other positions are 0.

[0086] When the value of the fluctuation range is relatively large, the above process will bring a huge amount of calculation. Further propose a simplified algorithm, that is, by finding the position and weight of the pulse corresponding to the maximum amplitude in each random fluctuation range, to form the target vector

[0087]

[0088] Where: m ij is any position within the i-th fluctuation range S i inside.

[0089] Target vector Substitute it into formula (4) to solve for the maximum value. Solving for the maximum value of formula (4) is equivalent to solving the derivative value of the optimal filter coefficient, that is:

[0090]

[0091] After derivation and calculation, the expression of the optimal filter f of VSMKD is:

[0092]

[0093] Step d: Filter the rolling bearing fault vibration signal x with the optimal solution convolution filter f to obtain the deconvolution signal, which is the fault impact signal y:

[0094]

[0095] Step e: Perform envelope demodulation processing on the fault impact signal y, extract the envelope of the fault impact signal, and obtain the envelope spectrum through spectrum analysis.

[0096] Envelope demodulation processing is to convert the high-frequency carrier signal into a low-frequency envelope signal by extracting the envelope (amplitude change of the signal) of the vibration signal, and then find the fault characteristic information through spectrum analysis. The specific steps are as follows:

[0097] ① Perform Hilbert transform on the fault impact signal y to obtain the analytic signal y a (t), and the amplitude of the analytic signal y a (t) is the envelope signal e(t):

[0098] y a (t) = y + j·Hy (15)

[0099]

[0100] Where: H represents the Hilbert transform.

[0101] ② Perform fast Fourier transform (FFT, Fast Fourier Transform) on the obtained envelope signal e(t) to obtain the spectrum of the envelope signal, that is, the envelope spectrum;

[0102] Step f: Judge the rolling bearing fault type according to the envelope spectrum:

[0103] ① Calculate the characteristic frequencies of various faults based on the geometric dimensions and rotational speed information of the faulty bearing:

[0104]

[0105] Where: BPFI is the characteristic frequency of inner race fault; BPFO is the characteristic frequency of outer race fault; BSF is the characteristic frequency of rolling element fault; N b is the number of rolling elements; B d is the diameter of the rolling element; P d is the pitch diameter of the bearing; φ is the contact angle; f r is the rotational frequency.

[0106] ② Determine the fault type and fault location based on the characteristic frequencies in the envelope spectrum and the characteristic frequencies of various fault components:

[0107] If the envelope spectrum is the characteristic frequency BPFO of the outer race fault and its multiples, then there is an outer race fault in the rolling bearing; if the envelope spectrum is the characteristic frequency BPFI of the inner race fault and its multiples, and there are sidebands on both sides of the characteristic frequency, with an interval of the rotational frequency f of the shaft r , then there is an inner race fault in the rolling bearing; if the envelope spectrum is the characteristic frequency BSF of the rolling element fault and its multiples, then there is a rolling element fault in the rolling bearing.

[0108] The following lists two specific implementation cases of the method of the present invention, and compares with the commonly used deconvolution method MOMEDA to prove the effectiveness of this method in the fault feature extraction and analysis of rolling bearings.

[0109] Example 1:

[0110] To verify the effectiveness of the proposed method, the publicly available rolling bearing dataset provided by Case Western Reserve University (CWRU) was used for analysis. The test bench consists of a 2HP (1.5KW) motor, a torque sensor, a power tester, and an electronic controller. In this case, the outer race fault signal of the bearing was selected. The bearing model is SKF 6203 deep groove ball bearing, the fault diameter is 0.014 inches, and the fault location is at the outer race of the fan-end bearing and at the 6 o'clock direction relative to the loading area. The experimental parameters are the motor speed of 1730 r / min, the motor load of 3HP (2.2KW), the sampling frequency of 48000 Hz, and the calculated characteristic frequency of the outer race fault of the bearing according to the theoretical formula is 103.4 Hz.

[0111] The time-domain waveform and envelope spectrum of the outer-race fault signal are shown in Fig. 3. No periodic impact component can be identified in the time-domain waveform shown in Fig. 3(a). The prominent peaks in the envelope spectrum shown in Fig. 3(b) do not correspond to the outer-race fault characteristic frequency BPFO. The fault characteristic frequency and its harmonics cannot be found, and the interference frequency components are obvious. The method VSMKD and MOMEDA proposed in the present invention are used for analysis, and the results are shown in Fig. 4. Fig. 4(a) is the envelope demodulation analysis result of the deconvolution signal obtained by VSMKD processing. From the figure, the 1st to 4th harmonics of the outer-race fault characteristic frequency can be accurately obtained, the messy interference frequencies are significantly suppressed, and the fault characteristics are clearly extracted; Fig. 4(b) is the envelope spectrum of the deconvolution signal obtained by the MOMEDA method. After filtering by the deconvolution method, the 1st to 3rd harmonics of the fault characteristic frequency can be found, but the fundamental frequency and the 3rd harmonic of the fault characteristic frequency are submerged in other interference frequencies, and there are also obvious interference frequencies near the 2nd harmonic, and the fault characteristic frequency cannot be accurately extracted. Through the comparative analysis of the two methods, it can be seen that the method proposed in the present invention can not only accurately extract the fault characteristic frequency, but also effectively suppress the messy interference frequency components, which proves that this method has significant advantages in the extraction of rolling bearing fault characteristics.

[0112] Embodiment 2:

[0113] The experimental signal used in this case is from the bearing fault experimental data of North China Electric Power University. The used rolling bearing fault test bench consists of a variable-frequency motor - speed-up gearbox - rotating shaft - faulty bearing - loader. In this case, the rolling element fault signal of the bearing is selected. The bearing signal is a deep groove ball bearing of type LYC6205E, and the fault is a single-point defect of the rolling element. The experimental parameters are that the motor speed is 1800 r / min, the motor load is 0, the sampling frequency is 12800 Hz, and the fault characteristic frequency of the rolling element is calculated to be 70.7 Hz according to the theoretical formula.

[0114] The time-domain waveform and envelope spectrum of the rolling element fault signal are shown in Fig. 6. It can be found that there are impact components in the time-domain waveform shown in Fig. 6(a). In the envelope spectrum shown in Fig. 6(b), the frequency components are relatively complex, and the fault characteristic frequency is submerged by the interference frequency, and the fault characteristic frequency cannot be clearly distinguished. The method VSMKD proposed in the present invention is compared with the MOMEDA method. In the envelope spectrum of the signal after VSMKD deconvolution, the 1st to 5th harmonics of the rolling element fault characteristic frequency are clearly visible, and the interference frequency is effectively suppressed; in the envelope spectrum of the signal after MOMEDA deconvolution, the rolling element fault characteristic frequency and its harmonics are not prominent, and it is difficult to accurately extract the fault characteristics. Through the comparison of the two methods, the superiority of the proposed method in extracting the rolling bearing fault characteristics is proved.

[0115] The method of the present invention can accurately identify the fault characteristic frequencies that cannot be detected by direct envelope analysis, and the effect is significantly better than that of MOMEDA, which proves that the method of the present invention has a significant promoting effect on the extraction of rolling bearing fault characteristics.

Claims

1. A rolling bearing fault feature extraction method based on variable-scale multi-point kurtosis deconvolution, characterized in that, The method includes the following steps: a: Sampling the vibration acceleration signal of the rolling bearing by using the acceleration sensor installed on the bearing seat to obtain the rolling bearing fault vibration signal x; b: Constructing the Toeplitz autocorrelation matrix X0: Using N to represent the data length of the rolling bearing fault vibration signal x, dividing the rolling bearing fault vibration signal x from a point, setting the length of each segment of the signal to L, and taking (N - L + 1) segments to form the Toeplitz autocorrelation matrix X0; where x i represents the i-th data in the rolling bearing fault vibration signal x, where i = 1, 2, …, N; c: Constructing a variable-scale multi-point kurtosis deconvolution filter: ① Introducing the multi-point D-norm to find the maximum value formula; Where: y is the fault impact signal, is the target vector, the weight at the impact location is 1, and the weights at the remaining locations are 0; ② Constructing a variable-scale multi-point kurtosis index as the objective function; Wherein: is the weight value corresponding to each sampling point from 0 to N-L, and y n is the signal amplitude corresponding to each sampling point from 0 to N-L, and the subscript n represents each sampling point of the signal from 0 to N-L; ③ Calculate the target vector There are random fluctuations in the failure cycle of rolling bearings. Define S i as the j-th weight w within the i-th fluctuation range ij that makes up the array, where w ij takes a value of 0 or 1. Combine all the weights w ij to form the weight vector These weight vectors are defined as the set W: Substitute the weight vector into the objective function to calculate the multi-point kurtosis values, and find the weight vector when the maximum value of the VSMK index is obtained which is the target vector Where: w i , i = 1, 2, … represent the positions and weights of the target pulses, and are elements in; ④ Substitute the target vector into the constructed multi-point D-norm maximum value calculation formula to solve for the maximum value, and obtain the expression of the optimal filter f of VSMKD: d: Filtering the rolling bearing fault vibration signal x with the optimal deconvolution filter f, and the obtained deconvolution signal is the fault impact signal y; e: Performing envelope demodulation processing on the fault impact signal y, extracting the envelope of the fault impact signal, and obtaining the envelope spectrum through spectrum analysis; f: Judging the rolling bearing fault type according to the envelope spectrum.

2. A rolling bearing fault feature extraction method based on variable-scale multi-point kurtosis deconvolution according to claim 1, characterized in that, The specific method for obtaining the envelope spectrum through spectrum analysis is: ①Perform Hilbert transform on the fault impact signal y to obtain the analytic signal y a (t). The amplitude of the analytic signal y a (t) is the envelope signal e(t): y a (t) = y + j·Hy In the formula: H represents the Hilbert transform; ② Performing a fast Fourier transform on the obtained envelope signal e(t) to obtain the spectrum of the envelope signal, that is, the envelope spectrum.

3. A rolling bearing fault feature extraction method based on variable-scale multi-point kurtosis deconvolution according to claim 1 or 2, characterized in that, The specific steps for judging the rolling bearing fault type according to the envelope spectrum are as follows: ① Calculating the characteristic frequencies of various faults according to the geometric dimensions and rotational speed information of the faulty bearing: In the formula: BPFI is the inner race fault characteristic frequency; BPFO is the outer race fault characteristic frequency; BSF is the rolling element fault characteristic frequency; N b is the number of rolling elements; B d is the diameter of the rolling elements; P d is the pitch diameter of the bearing; φ is the contact angle; f r is the rotational frequency; ② Judging the fault type and fault location according to the characteristic frequencies in the envelope spectrum and the characteristic frequencies of various fault components If the envelope spectrum is the outer race fault characteristic frequency BPFO and its multiples, then there is an outer race fault in the rolling bearing; if the envelope spectrum is the inner race fault characteristic frequency BPFI and its multiples, and there are sidebands on both sides of the characteristic frequency, with an interval of the shaft rotation frequency f r , then there is an inner race fault in the rolling bearing; if the envelope spectrum is the rolling element fault characteristic frequency BSF and its multiples, then there is a rolling element fault in the rolling bearing.

4. A rolling bearing fault feature extraction method based on variable-scale multi-point kurtosis deconvolution according to claim 3, characterized in that, in Calculating the target vector During the process, when the value range of fluctuations is relatively large, to simplify the algorithm, the position and weight of the pulse corresponding to the maximum amplitude within each random fluctuation range are found to form the target vector where: m ij is any position within the i-th fluctuation range S i ​

Images