Weak Fault Diagnosis Method of High-Pressure Roller Mill Rolling Bearing Based on EEMD-FKU-PMCKD under Strong Background Noise

Abstract

The present invention proposes a weak fault diagnosis method for the rolling bearing of a high-pressure roller mill under strong background noise. For the vibration signal containing strong background noise, ensemble empirical mode decomposition is used to decompose the signal to obtain IMF components. Taking the cross-correlation coefficient and kurtosis as indicators, the optimal components are selected to reconstruct the original signal. Secondly, based on the fast spectral kurtosis algorithm, the recombined signal is filtered once, and the envelope spectrum analysis of the filtered signal is performed to determine the optimization range of the parameters in the maximum correlation deconvolution. According to the particle swarm algorithm, the parameters of the MCKD model are optimized adaptively to obtain the optimal combination parameters of the filter length L, deconvolution period T s and the shift number M in PMCKD. The optimal optimization parameter results are substituted back into the PMCKD model to perform adaptive secondary filtering enhancement on the once-filtered signal. Finally, the bearing fault characteristic frequency extracted by the envelope spectrum is compared with the theoretical value to determine the fault type and achieve fault diagnosis.

Description

Technical Field

[0001] The present invention belongs to the field of rotary machinery fault diagnosis, and particularly relates to the weak fault diagnosis of the rolling bearing of a high-pressure roller mill under strong background noise by using EEMD-FKU-PMCKD. Background Art

[0002] The key components of large-scale mining equipment are complex in structure and harsh in working conditions. They work under low speed, heavy load, high impact load and high dust for a long time. If key rotating parts such as rolling bearings in the equipment fail, it will lead to the shutdown of the equipment, causing significant economic losses and even major safety accidents. According to statistics, 50% of mechanical faults in mining equipment are caused by rolling bearings.

[0003] The high-pressure roller mill is a typical large-scale mining equipment with high energy-saving and production-increasing effects, and has been widely used in the fields of building materials, chemical industry, mining, etc. The double-roll rolling bearing of the high-pressure roller mill is a key component to ensure the stable operation of the roller mill. Due to the large load, long operation period and harsh operation environment of the double-roll rolling bearing of the high-pressure roller mill, the rolling bearing is easily damaged, resulting in a shortened service life cycle. After the bearing is damaged, the repair period is long and the cost is high, resulting in significant economic losses. In addition, the operation environment of the high-pressure roller mill is harsh, and the information reflected by the sensor has great uncertainty. Due to the lack of useful information or the influence of interference signals, strong background noise is caused. Therefore, the weak fault diagnosis of the rolling bearing of the high-pressure roller mill under strong background noise by using EEMD-FKU-PMCKD is crucial.

[0004] The early faults of the rolling bearing of the high-pressure roller mill are very weak, and there is strong noise interference in the system. The vibration transmission direction of the rolling bearing is complex and changeable, with great nonlinearity and non-stationarity, which causes great difficulties in extracting fault characteristics. Effectively purifying and reducing the noise of the original fault signal of the rolling bearing and enhancing the weak impact components in the signal are the keys to the fault diagnosis of the rolling bearing of the high-pressure roller mill.

[0005] Traditional rolling bearing fault diagnosis methods, such as wavelet noise reduction, empirical mode decomposition (EMD), local mean decomposition (LMD), etc., are difficult to realize the selection of adaptive wavelet bases and decomposition layers for the collected signals, and lack strict mathematical theories. Ensemble empirical mode decomposition (EEMD) can effectively avoid the problems of mode mixing and endpoint effects generated by signal decompositions such as EMD and LMD by adding Gaussian white noise. Fast kurtosis (FKU) finds the periodic impact component with the largest kurtosis by designing a band-pass filter. Maximum correlated deconvolution (MCKD) uses a blind deconvolution method to design a'special filter' to enhance the periodic pulse components and non-periodic weak impacts swallowed by strong noise, and improve the kurtosis value of the original signal, which is very suitable for extracting the transient impacts of weak fault signals under strong background noise.

[0006] However, due to the existence of strong background noise, it is difficult to obtain good diagnostic results only using the EEMD and FKU methods - the energy of the bearing fault frequency diagnosed is relatively low, the multiple frequency base is small, and the diagnostic effect is not good. The selection of parameters for the MCKD algorithm is relatively vague and overly dependent on artificial empirical knowledge, and the quality of the parameters directly affects the final diagnostic effect.

[0007] The present invention proposes an EEMD-FKU-PMCKD intelligent diagnosis scheme, which solves the problems of difficult parameter determination in the MCKD algorithm and difficult extraction of weak fault features under strong background noise, realizes noise reduction and purification of the original signal, and further enhances the fault impact component, making the fault diagnosis of rolling bearings under strong background noise more accurate. Summary of the Invention

[0008] In view of the defects and deficiencies in the prior art, the present invention proposes a method for diagnosing weak faults of the rolling bearing of a high-pressure roller mill under strong background noise based on EEMD-FKU-PMCKD.

[0009] The technical solution adopted by the present invention to solve its technical problems is as follows:

[0010] A method for diagnosing weak faults of the rolling bearing of a high-pressure roller mill under strong background noise based on EEMD-FKU-PMCKD includes the following steps:

[0011] S1: Obtain the signal

[0012] Use an acceleration sensor to collect data on the double-roller rolling bearing of the high-pressure roller mill to obtain a vibration acceleration signal with strong noise; the collected vibration acceleration signal is in the radial direction.

[0013] S2: Perform multi-scale decomposition on the noise vibration signal

[0014] Decompose and denoise the original input signal using the ensemble empirical mode decomposition technique to generate multi-scale intrinsic mode components;

[0015] S3: Reconstruct the signal. Screen out the sub-signal with the strongest correlation in the IMF components through the cross-correlation coefficient and kurtosis threshold for signal reconstruction;

[0016] S4: Perform fast spectral kurtosis filtering on the reconstructed signal, make a fast spectral kurtosis diagram of the reconstructed signal, use it to determine the center frequency, bandwidth, and number of layers corresponding to the maximum kurtosis value of the signal, implement primary filtering, and obtain the filtered signal.

[0017] S5: Determine the parameter optimization range. Determine the deconvolution period T s range through the prominent frequencies in the envelope spectrum of the filtered signal. Thus, determine the optimization range of the parameters [L, T, M] in the MCKD algorithm, and use the PSO algorithm for adaptive optimization of parameter selection.

[0018] S6: Perform feature enhancement on the single-filtered signal. Record the optimal combination parameters [L, T, M] in PMCKD, substitute them back into the PMCKD model, and perform deconvolution feature enhancement on the single-filtered signal.

[0019] S7: Fault identification. Extract the rolling bearing fault features from the feature-enhanced signal through envelope processing, compare them with the theoretical fault frequencies of the bearings in the transmission system, and obtain the diagnostic results to achieve fault diagnosis.

[0020] The present invention has the following beneficial effects:

[0021] 1. The present invention proposes an EEMD-FKU-PMCKD intelligent diagnosis scheme. The EEMD technology overcomes problems such as mode mixing and end effects generated by traditional decomposition algorithms, and realizes noise reduction and purification of the original signal.

[0022] 2. MCKD and FKU have strong mathematical theoretical support. At the same time, FKU can find the periodic impact component with the largest kurtosis, making the fault components appear. PMCKD further enhances the fault components and highlights the faults. The PSO optimization algorithm adaptively searches for the parameters in the MCKD model, overcoming the problem of difficult determination of MCKD parameters.

[0023] 3. The present invention proposes an EEMD-FKU-PMCKD intelligent diagnosis scheme, realizes adaptive selection of parameters, solves the problems of over-reliance on prior knowledge and difficult extraction of slight fault features under strong background noise, and makes the fault diagnosis of rolling bearings under strong background noise more accurate. Description of the Drawings

[0024] Figure 1 is the flowchart of strong noise signal processing Detailed Embodiment

[0025] The following combines the attached Figure 1 Carry out a detailed description of the specific implementation manner of the present invention.

[0026] A weak fault diagnosis method for the rolling bearings of a high-pressure roller mill under strong background noise based on EEMD-FKU-PMCKD includes the following steps:

[0027] S1: Obtain the signal

[0028] Use an acceleration sensor to collect data on the double-roller rolling bearings of the high-pressure roller mill, and obtain a vibration acceleration signal with strong noise; the collected vibration acceleration signal is in the radial direction.

[0029] S2: Decompose the multi-scale signal

[0030] The original input signal is decomposed and denoised using the Ensemble Empirical Mode Decomposition (EEMD) technique to generate multi-scale Intrinsic Mode Functions (IMFs).

[0031] Step S2 is specifically as follows: The original signal is decomposed into IMF components using the EEMD algorithm, which essentially performs discrete noise reduction on the signal and selects the component with the maximum signal-to-noise ratio. The steps are as follows:

[0032] S21: Add Gaussian white noise n i (t) with an amplitude level of k I times to the original signal x(t) to generate a new signal z i (t).

[0033] z i (t) = x(t) + kn i (t) (1)

[0034] Where: i ∈ [1, I], k takes values in (0.05 - 0.25), and I takes values in (100 - 200).

[0035] S22: Use EMD to decompose z i (t) respectively to obtain a set of H IMFs and a residue r i (t).

[0036]

[0037] Where: c ih (t) — the h-th IMF obtained after the i-th EMD decomposition, h ∈ [1, I].

[0038] S23: Calculate the average of the corresponding IMFs in the I groups to obtain the IMF and residue r i (t) after EEMD decomposition of the original signal.

[0039]

[0040] Where: c h (t) — the h-th IMF obtained after EEMD decomposition of the original signal x(t).

[0041] S24: Finally, obtain the EEMD decomposition result of the original signal:

[0042]

[0043] S3: Reconstruct the signal. Select the sub-signal with the strongest correlation in the IMF components through the cross-correlation coefficient and kurtosis threshold for signal reconstruction;

[0044] S31: Calculate the kurtosis values of each sub-signal. When the bearing is operating normally, the kurtosis value of the signal is small. When it is greater than 3, it is considered that there are more impact components. The formula is as follows

[0045]

[0046] Where: N is the number of signal segments, d is the signal length, u is the average value of the signal, and σ is the standard deviation of the signal.

[0047] S32: Calculate the cross-correlation coefficient values between the sub-signal and the original signal. The cross-correlation coefficient is used to calculate the linear similarity between two variables and characterize the correlation degree between two signals. The greater the cross-correlation coefficient, the stronger the correlation degree. The formula is as follows

[0048]

[0049] Where: x and y are time-domain signals, x and y are the means of signals x and y; j is the signal length.

[0050] S33: Evaluation of the optimal IMF components. Select the IMFs with kurtosis values greater than the kurtosis threshold and cross-correlation coefficient values greater than the cross-correlation threshold from all IMFs as the optimal components, and select the common IMFs of the two to perform signal reconstruction. The combined use of the two makes the reconstructed signal more effectively approximate the original signal and can effectively reduce the influence of noise, highlighting the fault components.

[0051] S4: Perform fast kurtogram (FKU) filtering on the reconstructed signal, make the fast kurtogram of the reconstructed signal, and use it to determine the center frequency, bandwidth, and number of layers corresponding to the maximum kurtosis value of the signal, so as to achieve one-time filtering and obtain the filtered signal.

[0052] Step S4 is specifically: The fast kurtogram (FKU) algorithm performs one-time filtering and purification on the reconstructed signal.

[0053] S41: Establish the theoretical framework of kurtogram

[0054] In the non-stationary case, if the excitation response of signal X(t) is Y(t), then the spectrum Y(t) can be expressed as

[0055]

[0056] Where: H(t,f) is the time-varying transfer function of the complex envelope of signal Y(t) at frequency f, which is usually calculated by STFT transformation.

[0057]

[0058] In the formula: γ(τ) is a time window function with a very small time width.

[0059] The kurtogram based on the fourth-order spectral cumulant is defined as:

[0060]

[0061] S 2nY S(f) = E{|H(t, f)dX(f)| 2n} / df (11)

[0062] Where: S 2Y S(f) is the 2n-order instantaneous moment, and S 2nY represents the energy of the complex envelope signal.

[0063] Therefore, the spectral kurtosis is the fourth-order spectral cumulant normalized by energy, used to measure the peak value of the probability density function of a process at a certain frequency, that is:

[0064]

[0065] S42: Filter design. Design the filter based on the Fast Kurtogram concept of the tower algorithm and improve the calculation efficiency.

[0066] S421: Construct a low-pass filter h0(n) and a high-pass filter h1(n)

[0067] h0(n) = h(n)e jπn / 4 f ∈ [0, 1 / 4] (13)

[0068] h1(n) = h(n)e j3πn / 4 f ∈ [1 / 4, 1 / 2] (14)

[0069] Where: h0(n) is the low-pass filter model with a cut-off frequency of 1 / 8 (normalized frequency), and h1(n) is the high-pass filter model with a cut-off frequency of 1 / 8 (normalized frequency).

[0070] S422: Perform band-pass filtering on the reconstructed signal, perform a 2-fold downsampling process on the filtering result, and iterate to obtain the corresponding filtering result and calculate the corresponding kurtosis value. represents the filtering result generated by the i-th filter in the L-th layer, where i ranges from 0 to 2 k -1, and the purpose of the 2-fold downsampling is to ensure that the data length of the single-layer filter is the same as the length of the reconstructed signal.

[0071] S423: Repeat step S422, summarize all spectral kurtoses. Obtain the two-dimensional graph of (f, Δf) - the Fast Kurtogram. Filter the signal based on the information obtained from the graph to obtain the primary filtered signal.

[0072] S5: Determine the parameter optimization range. Determine the deconvolution period T through the prominent frequencies in the envelope spectrum of the filtered signal sRange. Thus, the optimization range of the parameters [L, T, M] in the MCKD algorithm is determined, and the PSO algorithm is used for adaptive optimization of parameter selection.

[0073] S51: Use the PSO algorithm to globally optimize the parameters in MCKD. The velocity update formula and position update formula in the PSO algorithm are:

[0074]

[0075] where; d is the particle dimension, take d = 2. q = 0, 1, 2, 3..., q is the population size, take q = 10. h is the number of iterations, H is the maximum number of iterations, take H = 30. is the velocity of the d - dimension of particle q in the h - th iteration; represents the minimum weight, is the maximum weight; learning factors c1, c2, learning rate η; is the position of the individual extreme point of particle q in the d - dimension in the h - th iteration, is the current position of particle q in the d - dimension in the h - th iteration, is the position of the global extreme value of the entire population in the d - dimension in the h - th iteration.

[0076] The PSO algorithm selects the global optimal particle by calculating the particle fitness, repeatedly updates the particle velocity and position, and finally reaches the convergence condition to output the best optimized parameters.

[0077] S52: Configure the parameters of the PSO algorithm

[0078] Initialize the particle position and velocity, set the number of swarm particles N = 10, particle dimension d = 2, maximum number of iterations H = 30, learning factors c1 = c2 = 1, learning rate η takes a random number from 0 to 1, maximum inertia weight minimum inertia weight The maximum iteration position is [100, 100], the minimum position is [8, 10], and the particle velocity interval is [-1, 1].

[0079] S53: Determine the optimization range of each parameter of MCKD.

[0080] The PSO algorithm searches the parameter range of MCKD: filter length [20, 200], shift number [1, 7], deconvolution period [T s - 10, T s + 10], the fitness function is the time - domain kurtosis value. When the fitness function is maximum, the best optimized parameters are output.

[0081] S6: Perform feature enhancement on the single-filtered signal. Record the optimal combination parameters [L, T, M] in PMCKD, substitute them back into the PMCKD model, and perform deconvolution feature enhancement on the single-filtered signal.

[0082] S61: Determine the bearing noise model.

[0083] y n = h.x n + n(t) (17)

[0084] In the formula: y n is the response output signal, h is the transfer function of the system, x n = [x1, x2,... x n is a periodic impulsive signal, and n(t) represents the strong background noise of the system.

[0085] S62: Perform feature enhancement

[0086] S621: Process the response output signal with a FIR (finite impulse response filter).

[0087] The inverse filter function can be expressed as:

[0088]

[0089] In the formula: f = [f1, f2, f3... f n are the filter coefficients, and L is the filter length.

[0090] S622: Solve for the filter coefficients f, and the formula is as follows:

[0091]

[0092] Among them: CK M (T s ) is the relevant kurtosis, T S is the convolution period, and M is the number of unit impulse responses.

[0093] Determine the convolution period T through the fault characteristic frequency f 故障 , and the formula is as follows s In the formula: f

[0094]

[0095] In the formula: f s is the sampling frequency, T 故障 is the fault period, and f 故障 is the fault characteristic frequency.

[0096] S623: MCKD calculates the filter coefficient L with the maximum value of the maximum relevant kurtosis of each signal segment as the objective function, and the formula is as follows:

[0097]

[0098] To obtain the filter coefficient f, the partial derivative of f with respect to the formula (19) is calculated.

[0099]

[0100] The f that can be solved by S621 - S622 is as follows:

[0101]

[0102] Where:

[0103]

[0104] S624: Substitute the obtained final iterative filter coefficient into formula (18) to obtain the deconvolution enhanced signal x of the actual acquired signal y.

[0105] S7: Fault identification. Extract the rolling bearing fault characteristics from the feature enhanced signal through envelope processing, compare them with the theoretical fault frequencies of the bearings in the transmission system, and obtain the diagnostic results to achieve fault diagnosis;

[0106] S71: Calculate the fault frequency through the theoretical formula based on conditions such as bearing size parameters, load, and rotational speed information. The specific formula is as follows (idealized usage conditions: 1. The rolling bearing has no slip; 2. The geometric dimensions of the rolling bearing hardly change; 3. The outer ring of the bearing is fixed and does not rotate)

[0107] Inner ring fault formula:

[0108] Outer ring fault formula:

[0109] Cage fault formula:

[0110] Rolling element fault formula:

[0111] Where: D is the pitch diameter of the raceway (the distance between two rolling elements in the radial direction), d is the diameter of the rolling element, z is the number of rolling elements, f r is the rotational frequency.

[0112] S72: Conduct envelope analysis on the deconvolution enhanced signal. Compare the target frequency values extracted from the envelope spectrum with the theoretical fault frequency values to obtain the rolling bearing fault diagnosis results. The specific method is as follows.

[0113] S721: Determine whether the frequency range in the envelope spectrum diagram contains the outer ring fault frequency / inner ring fault frequency / rolling element fault frequency / cage fault frequency and their multiple frequencies.

[0114] S722: If so, it is considered that the bearing has failed.

[0115] S723: If not, it is considered that the bearing is normal.

[0116] S724: Determine the type of failure and output the diagnostic result.

[0117] The above shows and describes the basic principles, main features and advantages of the present invention. The present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A weak fault diagnosis method for the rolling bearing of a high-pressure roller mill under strong background noise based on EEMD-FKU-PMCKD, characterized in that It includes the following steps: S1: Obtain signals Use an acceleration sensor to collect data on the double-roll rolling bearings of a high-pressure roller mill, obtaining a vibration acceleration signal with strong noise. The collected vibration acceleration signal is in the radial direction; S2: Perform multi-scale decomposition on the noise vibration signal Decompose and denoise the original input signal using the ensemble empirical mode decomposition technique to generate multi-scale intrinsic mode components; S3: Reconstruct the signal Screen out the sub-signal with the strongest correlation in the IMF components through the cross-correlation coefficient and kurtosis threshold for signal reconstruction; S4: Perform fast spectral kurtosis filtering on the reconstructed signal, create a fast spectral kurtosis diagram of the reconstructed signal to determine the center frequency, bandwidth, and number of layers corresponding to the maximum signal kurtosis value, perform one-time filtering, and obtain the filtered signal; S5: Determine the parameter optimization range Determine the deconvolution period T by identifying prominent frequencies in the envelope spectrum of the filtered signal s range, thereby determining the optimization range for the parameters [L, T, M] in the MCKD algorithm, and using the PSO algorithm for adaptive optimization of parameter selection; S6: Enhance the features of the singly filtered signal Record the optimal combination parameters [L, T, M] in PMCKD, substitute them back into the PMCKD model, and perform deconvolution feature enhancement on the singly filtered signal; S7: Fault identification Extract the rolling bearing fault features from the feature-enhanced signal through envelope processing, compare them with the theoretical fault frequencies of the bearings in the transmission system, and obtain the diagnostic results to achieve fault diagnosis.

Images