Fault self-adaptive diagnosis method for low-speed heavy-duty bearing

Abstract

The invention relates to the technical field of bearing fault diagnosis, and discloses a fault self-adaptive diagnosis method for a low-speed heavy-duty bearing, which comprises the following steps: acquiring bearing vibration time domain data under a low-speed heavy-duty working condition, and carrying out data cleaning on the bearing vibration time domain data to obtain cleaned bearing vibration time domain data; performing spectrum analysis on the vibration time domain data of the cleaned bearing to obtain a spectrum diagnosis result; acquiring a noise reduction envelope spectrum of the cleaned bearing vibration time domain data by adopting a clustering weighted envelope spectrum method; performing envelope spectrum analysis on the noise reduction envelope spectrum to obtain a diagnosis result of the noise reduction envelope spectrum; if the frequency spectrum diagnosis result and the noise reduction envelope spectrum diagnosis result are normal, judging that the bearing is normal, otherwise, judging that a bearing fault exists; the method improves the efficiency and accuracy of bearing fault diagnosis.

Description

Technical Field

[0001] The invention relates to the technical field of bearing fault diagnosis, and in particular to a fault adaptive diagnosis method for a low-speed and heavy-loaded bearing. Background Art

[0002] As a key component of rotating machinery, the health status of rolling bearings directly affects the performance and life of the entire equipment. Failure to identify and handle bearing failures in a timely manner may cause the entire system to be paralyzed and cause huge economic losses. Therefore, in-depth research and development of efficient rolling bearing fault diagnosis technology is crucial to ensure the safety and reliability of rotating machinery systems.

[0003] In the field of rolling bearing fault diagnosis technology, there are many diagnostic methods according to different monitoring signals, including vibration analysis, oil sample analysis, noise analysis, acoustic emission analysis, etc. From the perspective of analytical means, these diagnostic methods can be summarized into two categories: one is the traditional signal processing method, which focuses on the extraction and identification of signal features; the other is the emerging deep learning method, which automatically learns and extracts deep fault features by building a complex neural network model to achieve intelligent diagnosis of bearing faults. Since the diagnostic effect of deep learning methods depends largely on the quality and quantity of data and has poor interpretability, they have not been widely used in engineering practice. Signal processing of bearing vibration signals to achieve fault diagnosis has become one of the effective methods for rolling bearing fault diagnosis in industry due to its strong applicability, high interpretability, and reliable diagnostic results. However, when rolling bearings are running under low-speed and heavy-load conditions, it is often difficult to stimulate obvious fault pulses, especially when the fault is still in the early stage, it will be submerged by a large amount of background noise, making its fault diagnosis more difficult than that under normal conditions. In addition, there is also an automatic identification method for rolling bearing faults, which realizes automatic matching of bearing fault frequencies based on the fact that the amplitude of the fault characteristic frequency must be included in the first few values ​​of the envelope power spectrum when the bearing fails. However, the selection of the first few values ​​in this method requires expert experience and will be interfered by random noise. At the same time, it only focuses on the envelope spectrum and omits the fault information of the spectrum. Therefore, this method cannot be applied to the automatic diagnosis of bearing faults under low-speed and heavy-load conditions. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for adaptively diagnosing faults of a low-speed and heavy-loaded bearing, which is used to solve the problems of low diagnostic efficiency and poor accuracy in the prior art.

[0005] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0006] A method for adaptive fault diagnosis of a low-speed heavy-load bearing comprises the following steps:

[0007] S1. Acquire bearing vibration time domain data under low speed and heavy load conditions, and clean the bearing vibration time domain data to obtain cleaned bearing vibration time domain data;

[0008] S2. Perform spectrum analysis on the cleaned bearing vibration time domain data to obtain spectrum diagnosis results;

[0009] S3, using a cluster weighted envelope spectrum method to obtain a denoised envelope spectrum of the cleaned bearing vibration time domain data;

[0010] S4, performing envelope spectrum analysis on the noise reduction envelope spectrum to obtain a diagnosis result of the noise reduction envelope spectrum;

[0011] S5. If the diagnosis results of the spectrum and the noise reduction envelope spectrum are both normal, it is determined that the bearing is normal; otherwise, there is a bearing fault.

[0012] Furthermore, step S2 specifically includes:

[0013] S21, performing Fourier transform on the cleaned bearing vibration time domain data to obtain spectrum data;

[0014] Among them, the spectrum data is the amplitude of the bearing vibration time domain data in the frequency domain;

[0015] S22, using a peak search method to process the spectrum data, to obtain the true rotation frequency of the spectrum data and the multiple frequency of the true rotation frequency of the spectrum data, the true fault characteristic frequency of the spectrum data and the multiple frequency of the true fault characteristic frequency of the spectrum data;

[0016] S23, obtaining the amplitude of the real fault characteristic frequency of the spectrum data and the multiple of the real fault characteristic frequency of the spectrum data, calculating the fault energy of the spectrum data, and simultaneously obtaining the amplitude of other spectrum lines of the spectrum data, calculating the fault threshold of the spectrum data;

[0017] S24, determining whether the fault energy of the spectrum data is greater than the fault threshold of the spectrum data, if so, the spectrum diagnosis result is the corresponding bearing fault type, otherwise, the spectrum diagnosis result is that the bearing is normal;

[0018] The corresponding bearing fault types include: if the fault energy of the inner ring of the bearing is greater than the fault threshold of the inner ring of the bearing, it is determined to be an inner ring fault; if the fault energy of the outer ring of the bearing is greater than the fault threshold of the outer ring of the bearing, it is determined to be an outer ring fault; if the fault energy of the rolling element of the bearing is greater than the fault threshold of the rolling element of the bearing, it is determined to be a rolling element fault; if the fault energy of the bearing retainer is greater than the fault threshold of the bearing retainer, it is determined to be a retainer fault;

[0019] S25. Calculate the reliability of the spectrum diagnosis result according to the spectrum diagnosis result.

[0020] Furthermore, step S22 specifically includes:

[0021] S221, setting the harmonic number of the frequency spectrum data conversion, calculating the first allowable value and the first allowable value range between the actual value and the theoretical value of the frequency multiple of the frequency spectrum data conversion, searching for a peak value within the first allowable value range, and if the frequency corresponding to the peak value is not 0, taking the peak value as a candidate frequency conversion, otherwise, selecting a frequency with an amplitude second lower than the peak value as a candidate frequency conversion;

[0022] The calculation formula of the first allowable value and the first allowable value range is:

[0023]

[0024] in, represents the first allowable value range, k represents the harmonic of the spectrum data conversion, N represents the total number of harmonics of the spectrum data conversion, f1 represents the theoretical value of the spectrum data one-fold conversion, and ε represents the first allowable value;

[0025] S222, obtaining 1 times the candidate frequency conversion and 1 / 2 times the candidate frequency conversion from the candidate frequency conversions, and determining whether the amplitude of 1 times the candidate frequency conversion is greater than the amplitude of 1 / 2 times the candidate frequency conversion, if so, executing step S223, otherwise, the candidate frequency conversion and the multiple of the candidate frequency conversion do not exist, and executing step S224;

[0026] S223, determining whether the amplitudes of the candidate frequency conversion and the multiple frequency of the candidate frequency conversion are peak values ​​in the first local interval, if so, the candidate frequency conversion is the real frequency conversion of the spectrum data, otherwise, the real frequency conversion of the spectrum data does not exist;

[0027] The first local interval refers to a region with the candidate frequency conversion frequency as the center and a number of spectrum lines of the spectrum data as the width on both sides;

[0028] S224, setting the harmonic number of the bearing fault characteristic frequency, calculating the second allowable value and the second allowable value range between the real value and the theoretical value of the multiple frequency of the four bearing fault characteristic frequencies, searching for a peak value within the second allowable value range, and if the frequency corresponding to the peak value is not 0, taking the peak value as a candidate fault characteristic frequency, otherwise, selecting a frequency with an amplitude second lower than the peak value as a candidate fault characteristic frequency;

[0029] The calculation formula of the second allowable value and the second allowable value range is:

[0030]

[0031] in, represents the second allowable value range, k′ represents the harmonic of the bearing fault characteristic frequency, M′ represents the total number of harmonics of the bearing fault characteristic frequency, f1′ represents the theoretical value of one times the bearing fault characteristic frequency, and ε′ represents the second allowable value;

[0032] S225, determining whether the amplitude of the candidate fault characteristic frequency is a peak value in the second local interval, if so, executing step S226, otherwise, the real fault characteristic frequency of the spectrum data does not exist;

[0033] The second local interval refers to a region with the candidate fault characteristic frequency as the center and a number of spectrum lines of the spectrum data as the width on both sides;

[0034] S226. Determine whether the candidate rotation frequency and the multiples of the candidate rotation frequency do not exist. If so, the candidate fault characteristic frequency is the true fault characteristic frequency of the spectrum data. Otherwise, calculate the difference between the candidate fault characteristic frequency and the candidate rotation frequency, and compare the difference with the frequency resolution respectively. If the difference is large, the candidate fault frequency is the true fault characteristic frequency of the spectrum data. If the frequency resolution is large, the true fault characteristic frequency of the spectrum data does not exist.

[0035] The frequency resolution is the difference between each frequency in the spectrum data.

[0036] Furthermore, the calculation formula of the fault energy of the spectrum data in step S23 is:

[0037]

[0038] Among them, E fault represents the fault energy of the spectrum data, A(k′×f fault ) represents the amplitude of the real fault characteristic frequency of the spectrum data on the spectrum, f fault Indicates the true fault characteristic frequency of the spectrum data.

[0039] Furthermore, the calculation formula of the fault threshold of the spectrum data in step S23 is:

[0040]

[0041] Among them, ξ1 represents the fault threshold for judging whether the inner ring or rolling element of the bearing is faulty, ξ2 represents the fault threshold for judging whether the outer ring or retainer of the bearing is faulty, μ represents the mean operator, A(f) represents the amplitude corresponding to all frequencies in the range of 0 to the first N′ times the fault characteristic frequency on the spectrum data, excluding the first N′ times the fault characteristic frequency, m and n both represent normal numbers, and σ represents the standard deviation operator.

[0042] Furthermore, the calculation formula for the result reliability of the spectrum diagnosis result in step S25 is:

[0043]

[0044] Among them, R represents the reliability of the spectrum diagnosis result, E other It represents the mean of the amplitudes corresponding to all frequencies except the first N′ times fault characteristic frequencies in the local area containing the first N′ times fault characteristic frequencies, that is, the average energy of other frequencies. A(f)′ represents the amplitudes corresponding to all frequencies in the local area containing the first N′ times fault characteristic frequencies on the frequency spectrum data, excluding the first N′ times fault characteristic frequencies.

[0045] Furthermore, step S3 specifically includes:

[0046] S31, calculating the fast spectral coherence of the cleaned bearing vibration time domain data, that is:

[0047]

[0048] Among them, α represents the cycle frequency reflecting the fault characteristic frequency, f represents the carrier frequency reflecting the spectrum frequency, and γ x represents the spectral coherence function, S x represents the spectral correlation function, x(t) represents the vibration time domain signal of the bearing cleaned at time t, τ represents the time lag, R x represents the instantaneous autocorrelation function of the cleaned bearing vibration time domain signal, e represents the exponential function, j represents the imaginary unit, x(t-τ) represents the result of the cleaned bearing vibration time domain signal after time delay, (·)* represents the complex conjugate operator, and E represents the expectation operator;

[0049] S32, based on the fast spectral coherence of the cleaned bearing vibration time domain data, divide the spectral frequency slices in sequence along the spectral frequency direction, identify the potential candidate fault characteristic frequency of each spectral frequency slice, and use the potential candidate fault characteristic frequency as the label and weight of each spectral frequency slice;

[0050] The specific process of identifying the potential candidate fault characteristic frequency of each spectral frequency slice is as follows:

[0051] Select several local maximum values ​​in each spectrum frequency slice, use the multiple relationship test method to select the frequency array that satisfies the approximate multiple relationship, and obtain the potential candidate fault characteristic frequency and the multiple frequency of the potential candidate fault characteristic frequency;

[0052] Among them, the calculation formula for selecting the frequency array that satisfies the approximate multiple relationship using the multiple relationship test method is:

[0053]

[0054] Among them, Γ k″represents the frequency array that satisfies the approximate multiple relationship, k″ represents the frequency that satisfies the approximate multiple relationship, p1 represents one potential candidate fault feature frequency, M represents the total number of frequencies that satisfy the approximate multiple relationship, and p k″ represents k″ times the potential candidate fault characteristic frequency, f n Indicates the nth carrier frequency, max indicates the maximum value, represents the cycle frequency with an approximate multiple relationship, and λ represents the error tolerance;

[0055] S33, clustering the spectrum frequency slices with similar labels to obtain clusters of the spectrum frequency slices, and sorting all the clusters of the spectrum frequency slices from large to small according to the weights to obtain sorted clusters of the spectrum frequency slices;

[0056] S34. According to the sorted clusters of the spectral frequency slices, a cluster containing at least two spectral frequency slices with similar labels is selected as a cluster, and a weighted average of the spectral frequency slices in each selected cluster is used as a denoised envelope spectrum.

[0057] Furthermore, the calculation formula of the weight in step S32 is:

[0058]

[0059] Among them, W represents weight, M p represents the total number of selected local maxima, m′ represents the local maximum, α m′ represents the cycle frequency with a local maximum value of m′, m1 represents the index value of one potential candidate fault frequency in the cycle frequency, min represents taking the minimum value, and st represents satisfying it.

[0060] Furthermore, the calculation formula of the noise reduction envelope spectrum in step S34 is:

[0061]

[0062] Among them, CWES h represents the denoised envelope spectrum of the h-th cluster, I h represents the number of spectral frequency slices in the hth cluster, ω represents the weight function of the spectral frequency slice, c h (i) represents the i-th spectral frequency slice of the h-th cluster.

[0063] Furthermore, step S4 specifically includes:

[0064] S41, using a peak search method to perform data processing on the noise reduction envelope spectrum, to obtain a true rotation frequency of the noise reduction envelope spectrum and a multiple of the true rotation frequency of the noise reduction envelope spectrum, a true fault characteristic frequency of the noise reduction envelope spectrum and a multiple of the true fault characteristic frequency of the noise reduction envelope spectrum;

[0065] S42, obtaining the amplitude of the real fault characteristic frequency of the noise reduction envelope spectrum and the multiple frequency of the real fault characteristic frequency of the noise reduction envelope spectrum, calculating the fault energy of the noise reduction envelope spectrum, and simultaneously obtaining the amplitude of other spectral lines of the noise reduction envelope spectrum, and calculating the fault threshold of the noise reduction envelope spectrum;

[0066] S43, judging whether the fault energy of the noise reduction envelope spectrum is greater than the fault threshold of the noise reduction envelope spectrum, if so, the diagnosis result of the noise reduction envelope spectrum is the corresponding bearing fault type, otherwise, the diagnosis result of the noise reduction envelope spectrum is that the bearing is normal;

[0067] S44. Calculate the reliability of the diagnosis result of the denoising envelope spectrum according to the diagnosis result of the denoising envelope spectrum.

[0068] The present invention has the following beneficial effects:

[0069] 1. The present invention proposes a method for adaptive fault diagnosis of low-speed and heavy-load bearings. The peak search method is used to design the allowable value in a segmented form. This method can not only automatically identify the rotation frequency and the fault frequency, but also greatly reduce the interference of noise. At the same time, a mechanism for preventing the rotation frequency and the fault characteristic frequency from being confused is set, thereby improving the accuracy and robustness of fault diagnosis.

[0070] 2. A calculation and comparison mechanism for fault energy and fault threshold is introduced. This mechanism can simultaneously characterize the four fault intensities of the bearing, namely the inner ring, outer ring, rolling element and cage, so as to realize the diagnosis of bearing composite faults; and by calculating the energy proportion of the bearing fault in the local frequency band, it can be used to evaluate the reliability of the diagnosis result;

[0071] 3. Using clustering weighted envelope spectrum method, the discrete information bands of multiple faults can be quickly located without prior knowledge and iterative optimization, the characteristics of weak bearing faults can be extracted, and signal noise reduction can be achieved;

[0072] 4. Comprehensive consideration of the frequency spectrum and envelope spectrum can fully extract bearing fault information, thereby realizing early, mid-term and late fault diagnosis of bearings under low-speed and heavy-load conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 A schematic diagram of a flow chart of a method for adaptive fault diagnosis of a low-speed and heavy-load bearing proposed by the present invention;

[0074] Figure 2 A schematic diagram of a calculation process of fault energy of spectrum data and fault threshold of spectrum data;

[0075] Figure 3 A schematic diagram of the calculation process of the reliability of the spectrum diagnosis result;

[0076] Figure 4It is a schematic diagram of the frequency spectrum analysis of the axial direction of the front bearing of the main shaft of the wind turbine in the embodiment;

[0077] Figure 5 Schematic diagram of envelope spectrum analysis of the axial direction of the front bearing of the main shaft of the wind turbine in the embodiment;

[0078] Figure 6 It is a schematic diagram of the spectrum analysis in the horizontal direction of the rear bearing of the main shaft of the wind turbine in the embodiment;

[0079] Figure 7 Schematic diagram of envelope spectrum analysis of the rear bearing of the wind turbine main shaft in the horizontal direction in the embodiment. DETAILED DESCRIPTION

[0080] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0081] like Figure 1 As shown, a fault adaptive diagnosis method for a low-speed heavy-load bearing includes the following steps S1-S5:

[0082] S1. Acquire bearing vibration time domain data under low speed and heavy load conditions, and clean the bearing vibration time domain data to obtain cleaned bearing vibration time domain data.

[0083] In this embodiment, the purpose of data cleaning is to improve data quality.

[0084] S2. Perform spectrum analysis on the cleaned bearing vibration time domain data to obtain spectrum diagnosis results.

[0085] In this embodiment, the principle of spectral analysis is: to perform Fourier transform on the time domain signal to obtain the amplitudes corresponding to various frequencies in the signal in the frequency domain; therefore, the purpose of performing spectral analysis on the cleaned bearing vibration time domain data in this embodiment is to find the bearing fault characteristic frequency in frequency, thereby comparing the fault energy of the bearing fault characteristic frequency with the threshold value in frequency, thereby realizing fault diagnosis.

[0086] Specifically, step S2 specifically includes S21-S25:

[0087] S21. Perform Fourier transform on the cleaned bearing vibration time domain data to obtain spectrum data; wherein the spectrum data is the amplitude of the bearing vibration time domain data in the frequency domain.

[0088] S22. Process the spectrum data using a peak search method to obtain the true rotation frequency of the spectrum data and the multiples of the true rotation frequency of the spectrum data, the true fault characteristic frequency of the spectrum data and the multiples of the true fault characteristic frequency of the spectrum data.

[0089] Specifically, step S22 specifically includes S221-S226:

[0090] S221. Set the harmonic number of the spectrum data frequency conversion, calculate the first allowable value and the first allowable value range between the actual value and the theoretical value of the multiple frequency of the spectrum data frequency conversion, search for the peak value within the first allowable value range, if the frequency corresponding to the peak value is not 0, then the peak value is used as a candidate for frequency conversion, otherwise, the frequency with an amplitude second lower than the peak value is selected as the candidate for frequency conversion.

[0091] The calculation formula of the first allowable value and the first allowable value range is:

[0092]

[0093] in, represents the first allowable value range, k represents the harmonics of the spectrum data conversion, N represents the total number of harmonics of the spectrum data conversion, f1 represents the theoretical value of one-fold conversion of the spectrum data, and ε represents the first allowable value.

[0094] In this embodiment, the harmonic number of the spectrum data conversion is set to 3. In addition, when judging a bearing fault, the spectrum data conversion and its harmonics may be close to the bearing fault frequency, which may cause confusion between the two. Therefore, the harmonic number of the conversion needs to be set in advance.

[0095] S222, obtaining 1 times the candidate frequency conversion and 1 / 2 times the candidate frequency conversion from the candidate frequency conversions, and determining whether the amplitude of 1 times the candidate frequency conversion is greater than the amplitude of 1 / 2 times the candidate frequency conversion; if so, executing step S223; otherwise, neither the candidate frequency conversion nor the multiple of the candidate frequency conversion exists, and executing step S224.

[0096] S223, determining whether the amplitudes of the candidate frequency transition and the multiple of the candidate frequency transition are peak values ​​in the first local interval; if so, the candidate frequency transition is the real frequency transition of the spectrum data; otherwise, the real frequency transition of the spectrum data does not exist.

[0097] The first local interval refers to a region centered on the candidate transfer frequency and having a width of a number of spectral lines of the spectrum data on both sides.

[0098] In this embodiment, the number of spectral lines is generally greater than 1, and is set to 3 in this embodiment, which is an empirical value.

[0099] S224. Set the harmonic number of the bearing fault characteristic frequency, calculate the second allowable value and the second allowable value range between the actual value and the theoretical value of the multiples of the four bearing fault characteristic frequencies, search for the peak value within the second allowable value range, and if the frequency corresponding to the peak value is not 0, use the peak value as a candidate fault characteristic frequency, otherwise, select the frequency with an amplitude second lower than the peak value as the candidate fault characteristic frequency.

[0100] The calculation formula of the second allowable value and the second allowable value range is:

[0101]

[0102] in, represents the second allowable value range, k′ represents the harmonic of the bearing fault characteristic frequency, N′ represents the total number of harmonics of the bearing fault characteristic frequency, f′1 represents the theoretical value of one times the bearing fault characteristic frequency, and ε′ represents the second allowable value.

[0103] In this embodiment, the harmonic number of the bearing fault characteristic frequency is set to 10. In addition, since the bearing fault is mainly judged by the harmonic number of the bearing fault characteristic frequency and the corresponding amplitude, it is necessary to set the harmonic number of the bearing fault characteristic frequency. At the same time, the four bearing fault characteristic frequencies include inner ring fault, outer ring fault, rolling element fault and cage fault.

[0104] S225, determining whether the amplitude of the candidate fault characteristic frequency is a peak value in the second local interval, if so, executing step S226, otherwise, the real fault characteristic frequency of the spectrum data does not exist.

[0105] The second local interval refers to a region with the candidate fault characteristic frequency as the center and a number of spectrum lines of the spectrum data as the width on both sides.

[0106] In this embodiment, the number of spectral lines is generally greater than 1, and is set to 3 in this embodiment.

[0107] S226. Determine whether the candidate rotation frequency and the multiples of the candidate rotation frequency do not exist. If so, the candidate fault characteristic frequency is the true fault characteristic frequency of the spectrum data. Otherwise, calculate the difference between the candidate fault characteristic frequency and the candidate rotation frequency, and compare the difference with the frequency resolution respectively. If the difference is large, the candidate fault frequency is the true fault characteristic frequency of the spectrum data. If the frequency resolution is large, the true fault characteristic frequency of the spectrum data does not exist.

[0108] The frequency resolution is the difference between each frequency in the spectrum data.

[0109] S23, obtaining the amplitude of the real fault characteristic frequency of the spectrum data and the multiple of the real fault characteristic frequency of the spectrum data, calculating the fault energy of the spectrum data, and simultaneously obtaining the amplitude of other spectrum lines of the spectrum data, and calculating the fault threshold of the spectrum data.

[0110] In this embodiment, Figure 2 The calculation process of the fault energy and fault threshold of the spectrum data is shown. The calculation formula is as follows:

[0111] Specifically, the calculation formula of the fault energy of the spectrum data in step S23 is:

[0112]

[0113] Among them, E fault represents the fault energy of the spectrum data, A(k′×f fault ) represents the amplitude of the real fault characteristic frequency of the spectrum data on the spectrum, f fault Indicates the true fault characteristic frequency of the spectrum data.

[0114] Specifically, the calculation formula of the fault threshold of the spectrum data in step S23 is:

[0115]

[0116] Among them, ξ1 represents the fault threshold for judging whether the inner ring or rolling element of the bearing is faulty, ξ2 represents the fault threshold for judging whether the outer ring or retainer of the bearing is faulty, μ represents the mean operator, A(f) represents the amplitude corresponding to all frequencies in the range of 0 to the first N′ times the fault characteristic frequency on the spectrum data, excluding the first N′ times the fault characteristic frequency, m and n both represent normal numbers, and σ represents the standard deviation operator.

[0117] In this embodiment, the values ​​of m and n are usually 0 to 3, and in this embodiment, the values ​​are 2.4 and 2.7.

[0118] S24, determining whether the fault energy of the spectrum data is greater than the fault threshold of the spectrum data, if so, the spectrum diagnosis result is the corresponding bearing fault type, otherwise, the spectrum diagnosis result is that the bearing is normal.

[0119] Among them, the corresponding bearing fault types include: if the fault energy of the inner ring of the bearing is greater than the fault threshold of the inner ring of the bearing, it is determined to be an inner ring fault; if the fault energy of the outer ring of the bearing is greater than the fault threshold of the outer ring of the bearing, it is determined to be an outer ring fault; if the fault energy of the bearing rolling element is greater than the fault threshold of the bearing rolling element, it is determined to be a rolling element fault; if the fault energy of the bearing retainer is greater than the fault threshold of the bearing retainer, it is determined to be a retainer fault.

[0120] S25. Calculate the reliability of the spectrum diagnosis result according to the spectrum diagnosis result.

[0121] Specifically, the calculation formula for the result reliability of the spectrum diagnosis result in step S25 is:

[0122]

[0123] Among them, R represents the reliability of the spectrum diagnosis result, E other It represents the mean of the amplitudes corresponding to all frequencies except the first N′ times fault characteristic frequencies in the local area containing the first N′ times fault characteristic frequencies, that is, the average energy of other frequencies. A(f)′ represents the amplitudes corresponding to all frequencies in the local area containing the first N′ times fault characteristic frequencies on the frequency spectrum data, excluding the first N′ times fault characteristic frequencies.

[0124] In this embodiment, Figure 3 The calculation process of the result reliability of the spectrum diagnosis result is shown, where A(f)′ is different from A(f); in addition, when the result reliability is above 75%, it is considered that the diagnosis result is relatively reliable.

[0125] S3. Use cluster weighted envelope spectrum method to obtain the denoised envelope spectrum of the cleaned bearing vibration time domain data.

[0126] In this embodiment, the purpose of obtaining the denoised envelope spectrum of the cleaned bearing vibration time domain data is: since the envelope spectrum noise of the original signal is relatively large, the characteristic frequency of the bearing fault is not obvious, so noise reduction is required to enhance the bearing fault characteristics.

[0127] Specifically, step S3 specifically includes S31-S34:

[0128] S31, calculating the fast spectral coherence of the cleaned bearing vibration time domain data, that is:

[0129]

[0130] Among them, α represents the cycle frequency reflecting the fault characteristic frequency, f represents the carrier frequency reflecting the spectrum frequency, and γ x represents the spectral coherence function, S x represents the spectral correlation function, x(t) represents the vibration time domain signal of the bearing cleaned at time t, τ represents the time lag, R x represents the instantaneous autocorrelation function of the cleaned bearing vibration time domain signal, e represents the exponential function, j represents the imaginary unit, x(t-τ) represents the result of the cleaned bearing vibration time domain signal after time delay, (·)* represents the complex conjugate operator, and E represents the expectation operator.

[0131] In this embodiment, the purpose of calculating the fast spectral coherence is to quantify the correlation between the spectral frequency and the cyclic frequency.

[0132] S32. Based on the fast spectral coherence of the cleaned bearing vibration time domain data, the spectral frequency slices are divided in sequence along the spectral frequency direction, the potential candidate fault characteristic frequency of each spectral frequency slice is identified, and the potential candidate fault characteristic frequency is used as the label and weight of each spectral frequency slice.

[0133] In this embodiment, by identifying the potential candidate fault characteristic frequency and its multiple frequency of each spectral frequency slice, the potential candidate fault characteristic frequency is used as a label, and the weight is calculated to prepare for the subsequent clustering of spectral frequency slices belonging to the same fault.

[0134] The specific process of identifying the potential candidate fault characteristic frequency of each spectral frequency slice is as follows:

[0135] In each spectrum frequency slice, several local maxima are selected, and a frequency array satisfying an approximate multiple relationship is selected using a multiple relationship test method to obtain potential candidate fault characteristic frequencies and multiples of potential candidate fault characteristic frequencies.

[0136] In this embodiment, the first 40 local maximum values ​​of the spectrum frequency slices are selected, and the frequency arrays that meet the approximate multiple relationship are selected by the multiple relationship test method and are set to 3.

[0137] Among them, the calculation formula for selecting the frequency array that satisfies the approximate multiple relationship using the multiple relationship test method is:

[0138]

[0139] Among them, Γ k″ represents the frequency array that satisfies the approximate multiple relationship, k″ represents the frequency that satisfies the approximate multiple relationship, p1 represents one potential candidate fault feature frequency, M represents the total number of frequencies that satisfy the approximate multiple relationship, and p k″ represents k″ times the potential candidate fault characteristic frequency, f n Indicates the nth carrier frequency, max indicates the maximum value, represents the cycle frequency with an approximate multiple relationship, and λ represents the error tolerance.

[0140] Specifically, the calculation formula of the weight in step S32 is:

[0141]

[0142] Among them, W represents weight, M p represents the total number of selected local maxima, m′ represents the local maximum, α m′ represents the cycle frequency with a local maximum value of m′, m1 represents the index value of one potential candidate fault frequency in the cycle frequency, min represents taking the minimum value, and st represents satisfying it.

[0143] S33, clustering the spectrum frequency slices with similar labels to obtain clusters of the spectrum frequency slices, and sorting all the clusters of the spectrum frequency slices from large to small according to the weights to obtain sorted clusters of the spectrum frequency slices.

[0144] S34. According to the sorted clusters of the spectral frequency slices, a cluster containing at least two spectral frequency slices with similar labels is selected as a cluster, and a weighted average of the spectral frequency slices in each selected cluster is used as a denoised envelope spectrum.

[0145] Specifically, the calculation formula of the noise reduction envelope spectrum in step S34 is:

[0146]

[0147] Among them, CWES h represents the denoised envelope spectrum of the h-th cluster, I h represents the number of spectral frequency slices in the hth cluster, ω represents the weight function of the spectral frequency slice, c h (i) represents the i-th spectral frequency slice of the h-th cluster.

[0148] S4. Perform envelope spectrum analysis on the noise reduction envelope spectrum to obtain a diagnosis result of the noise reduction envelope spectrum.

[0149] In this embodiment, the purpose of performing envelope spectrum analysis is: since the bearing fault may not be obvious in the frequency spectrum, but is obvious in the envelope spectrum, the bearing fault information hidden in the resonance frequency band is demodulated through envelope spectrum analysis, thereby improving the accuracy of the bearing fault diagnosis results.

[0150] Specifically, step S4 includes S41-S44:

[0151] S41. Process the noise reduction envelope spectrum using a peak search method to obtain the true rotation frequency of the noise reduction envelope spectrum and the multiples of the true rotation frequency of the noise reduction envelope spectrum, the true fault characteristic frequency of the noise reduction envelope spectrum and the multiples of the true fault characteristic frequency of the noise reduction envelope spectrum.

[0152] S42, obtaining the amplitude of the real fault characteristic frequency of the noise reduction envelope spectrum and the multiple of the real fault characteristic frequency of the noise reduction envelope spectrum, calculating the fault energy of the noise reduction envelope spectrum, and simultaneously obtaining the amplitude of other spectral lines of the noise reduction envelope spectrum, and calculating the fault threshold of the noise reduction envelope spectrum.

[0153] In this embodiment, the fault energy of the noise reduction envelope spectrum and the fault threshold of the noise reduction envelope spectrum are calculated using the same formula as the fault energy and fault threshold of the spectrum data in step S2.

[0154] S43, determining whether the fault energy of the noise reduction envelope spectrum is greater than the fault threshold of the noise reduction envelope spectrum, if so, the diagnosis result of the noise reduction envelope spectrum is the corresponding bearing fault type, otherwise, the diagnosis result of the noise reduction envelope spectrum is that the bearing is normal.

[0155] S44. Calculate the reliability of the diagnosis result of the denoising envelope spectrum according to the diagnosis result of the denoising envelope spectrum.

[0156] In this embodiment, the calculation formula of the result reliability of the diagnosis result of the noise reduction envelope spectrum is the same as the calculation formula of the result reliability of the spectrum diagnosis result in the above step S2.

[0157] S5. If the diagnosis results of the spectrum and the noise reduction envelope spectrum are both normal, it is determined that the bearing is normal; otherwise, there is a bearing fault.

[0158] In this embodiment, the bearing fault is diagnosed by combining the spectrum analysis with the envelope spectrum analysis, thereby improving the accuracy of the automatic diagnosis of the bearing fault.

[0159] Secondly, an experiment was conducted by collecting the axial vibration signal of the main shaft front bearing and the horizontal vibration signal of the main shaft rear bearing of a wind turbine in a wind farm to verify the effectiveness of the fault adaptive diagnosis method for low-speed and heavy-load bearings proposed in the present invention. The specific operation process is as follows:

[0160] The axial vibration signal of the front main shaft bearing and the horizontal vibration signal of the rear main shaft bearing were collected for a wind turbine in a wind farm. The sampling frequency was 256Hz. The fault characteristic frequencies of the inner ring, outer ring, rolling element and cage of the front main shaft bearing were 5.512Hz, 5.066Hz, 1.758Hz and 0.078Hz respectively. The fault characteristic frequencies of the inner ring, outer ring, rolling element and cage of the rear main shaft bearing were 6.213Hz, 5.740Hz, 1.992Hz and 0.079Hz respectively.

[0161] like Figure 4-Figure 5 As shown, for the front bearing of the main shaft, although no fault information is found in the frequency spectrum of its original axial vibration signal, it can be clearly found in the envelope spectrum of the noise reduction signal that the fault energies of the cage and the rolling element are greater than the fault threshold, and the cage and rolling element faults of the front bearing of the main shaft are accurately identified. The reliability of the envelope spectrum analysis result is greater than 75%, which verifies the effectiveness of the present invention in bearing fault diagnosis under low-speed and heavy-load conditions.

[0162] like Figure 6-Figure 7 As shown, for the rear bearing of the main shaft, its horizontal original vibration signal spectrum and noise reduction signal envelope spectrum can automatically identify inner and outer ring faults, and the reliability of the spectrum and envelope spectrum analysis results are both greater than 75%, verifying the effectiveness of the present invention in bearing fault diagnosis under low-speed and heavy-load conditions.

[0163] In summary, the present invention proposes a method for adaptive fault diagnosis of low-speed and heavy-load bearings. First, the peak search method is used to design the allowable value in a segmented form, which can not only automatically identify the rotation frequency and the fault frequency, but also greatly reduce the interference of noise. At the same time, a rotation frequency and fault characteristic frequency anti-confusion mechanism is set to improve the accuracy and robustness of fault diagnosis; secondly, a calculation and comparison mechanism of fault energy and fault threshold is introduced, which can simultaneously characterize the four fault intensities of the bearing, namely the inner ring, the outer ring, the rolling element and the retaining cage, so as to realize the diagnosis of bearing composite faults; and by calculating the energy proportion of the bearing fault in the local frequency band, it can be used to evaluate the reliability of the diagnosis result; then, the clustering weighted envelope spectrum method is used to quickly locate the discrete information bands of multiple faults without prior knowledge and iterative optimization, extract the weak fault characteristics of the bearing, and realize signal noise reduction; finally, by comprehensively considering the spectrum and the envelope spectrum, the bearing fault information can be fully extracted, thereby realizing the early, mid-term and late fault diagnosis of the bearing under low-speed and heavy-load conditions.

[0164] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

[0165] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A method for adaptive fault diagnosis of low-speed and heavy-load bearings, characterized in that: The following steps are involved: S1. Acquire bearing vibration time domain data under low speed and heavy load conditions, and clean the bearing vibration time domain data to obtain cleaned bearing vibration time domain data; S2. Perform spectrum analysis on the cleaned bearing vibration time domain data to obtain spectrum diagnosis results; S3, using a cluster weighted envelope spectrum method to obtain a denoised envelope spectrum of the cleaned bearing vibration time domain data; S4, performing envelope spectrum analysis on the noise reduction envelope spectrum to obtain a diagnosis result of the noise reduction envelope spectrum; S5. If the diagnosis results of the spectrum and the noise reduction envelope spectrum are both normal, it is determined that the bearing is normal; otherwise, there is a bearing fault.

2. The method for self-adaptive fault diagnosis of a low-speed and heavy-load bearing according to claim 1, characterized in that: Step S2 specifically includes: S21, performing Fourier transform on the cleaned bearing vibration time domain data to obtain spectrum data; Among them, the spectrum data is the amplitude of the bearing vibration time domain data in the frequency domain; S22, using a peak search method to process the spectrum data, to obtain the true rotation frequency of the spectrum data and the multiple frequency of the true rotation frequency of the spectrum data, the true fault characteristic frequency of the spectrum data and the multiple frequency of the true fault characteristic frequency of the spectrum data; S23, obtaining the amplitude of the real fault characteristic frequency of the spectrum data and the multiple of the real fault characteristic frequency of the spectrum data, calculating the fault energy of the spectrum data, and simultaneously obtaining the amplitude of other spectrum lines of the spectrum data, calculating the fault threshold of the spectrum data; S24, determining whether the fault energy of the spectrum data is greater than the fault threshold of the spectrum data, if so, the spectrum diagnosis result is the corresponding bearing fault type, otherwise, the spectrum diagnosis result is that the bearing is normal; The corresponding bearing fault types include: if the fault energy of the inner ring of the bearing is greater than the fault threshold of the inner ring of the bearing, it is determined to be an inner ring fault; if the fault energy of the outer ring of the bearing is greater than the fault threshold of the outer ring of the bearing, it is determined to be an outer ring fault; if the fault energy of the rolling element of the bearing is greater than the fault threshold of the rolling element of the bearing, it is determined to be a rolling element fault; if the fault energy of the bearing retainer is greater than the fault threshold of the bearing retainer, it is determined to be a retainer fault; S25. Calculate the reliability of the spectrum diagnosis result according to the spectrum diagnosis result.

3. The method for self-adaptive fault diagnosis of a low-speed and heavy-load bearing according to claim 2, characterized in that: Step S22 specifically includes: S221, setting the harmonic number of the frequency spectrum data conversion, calculating the first allowable value and the first allowable value range between the actual value and the theoretical value of the frequency multiple of the frequency spectrum data conversion, searching for a peak value within the first allowable value range, and if the frequency corresponding to the peak value is not 0, taking the peak value as a candidate frequency conversion, otherwise, selecting a frequency with an amplitude second lower than the peak value as a candidate frequency conversion; The calculation formula of the first allowable value and the first allowable value range is: in, represents the first allowable value range, k represents the harmonic of the spectrum data conversion, N represents the total number of harmonics of the spectrum data conversion, f1 represents the theoretical value of the spectrum data one-fold conversion, and ε represents the first allowable value; S222, obtaining 1 times the candidate frequency conversion and 1 / 2 times the candidate frequency conversion from the candidate frequency conversions, and determining whether the amplitude of 1 times the candidate frequency conversion is greater than the amplitude of 1 / 2 times the candidate frequency conversion, if so, executing step S223, otherwise, the candidate frequency conversion and the multiple of the candidate frequency conversion do not exist, and executing step S224; S223, determining whether the amplitudes of the candidate frequency conversion and the multiple frequency of the candidate frequency conversion are peak values ​​in the first local interval, if so, the candidate frequency conversion is the real frequency conversion of the spectrum data, otherwise, the real frequency conversion of the spectrum data does not exist; The first local interval refers to a region with the candidate frequency conversion frequency as the center and a number of spectrum lines of the spectrum data as the width on both sides; S224, setting the harmonic number of the bearing fault characteristic frequency, calculating the second allowable value and the second allowable value range between the real value and the theoretical value of the multiple frequency of the four bearing fault characteristic frequencies, searching for a peak value within the second allowable value range, and if the frequency corresponding to the peak value is not 0, taking the peak value as a candidate fault characteristic frequency, otherwise, selecting a frequency with an amplitude second lower than the peak value as a candidate fault characteristic frequency; The calculation formula of the second allowable value and the second allowable value range is: in, Indicates the second allowable value range, k ′ Harmonics representing the characteristic frequency of bearing faults, N ′ The total number of harmonics representing the characteristic frequency of the bearing fault, f1 ′ Represents the theoretical value of one times the characteristic frequency of bearing fault, ε ′ represents the second permissible value; S225, determining whether the amplitude of the candidate fault characteristic frequency is a peak value in the second local interval, if so, executing step S226, otherwise, the real fault characteristic frequency of the spectrum data does not exist; The second local interval refers to a region with the candidate fault characteristic frequency as the center and a number of spectrum lines of the spectrum data as the width on both sides; S226. Determine whether the candidate rotation frequency and the multiples of the candidate rotation frequency do not exist. If so, the candidate fault characteristic frequency is the true fault characteristic frequency of the spectrum data. Otherwise, calculate the difference between the candidate fault characteristic frequency and the candidate rotation frequency, and compare the difference with the frequency resolution respectively. If the difference is large, the candidate fault frequency is the true fault characteristic frequency of the spectrum data. If the frequency resolution is large, the true fault characteristic frequency of the spectrum data does not exist. The frequency resolution is the difference between each frequency in the spectrum data.

4. The method for self-adaptive fault diagnosis of a low-speed and heavy-load bearing according to claim 3, characterized in that: The calculation formula of the fault energy of the spectrum data in step S23 is: Among them, E fault represents the fault energy of the spectrum data, A(k ′ ×f fault ) represents the amplitude of the real fault characteristic frequency of the spectrum data on the spectrum, f fault Indicates the true fault characteristic frequency of the spectrum data.

5. The method for self-adaptive fault diagnosis of a low-speed and heavy-load bearing according to claim 4, characterized in that: The calculation formula of the fault threshold of the spectrum data in step S23 is: Where, ξ1 represents the fault threshold for judging whether the inner ring or rolling element of the bearing is faulty, ξ2 represents the fault threshold for judging whether the outer ring or cage of the bearing is faulty, μ represents the mean operator, and A(f) represents the frequency spectrum of the spectrum data from 0 to the first N. ′ times the amplitude of all frequencies within the fault characteristic frequency range, excluding the first N ′ times the fault characteristic frequency, m and n both represent normal numbers, and σ represents the standard deviation operator.

6. The method for self-adaptive fault diagnosis of a low-speed and heavy-load bearing according to claim 5, characterized in that: The calculation formula for the reliability of the spectrum diagnosis result in step S25 is: Among them, R represents the reliability of the spectrum diagnosis result, E other Indicates that the first N ′ times the fault characteristic frequency in the local area except for the first N ′ The mean of the amplitudes corresponding to all frequencies other than the fault characteristic frequency, that is, the average energy of other frequencies, A(f) ′ Indicates that the frequency of the spectrum data contains the first N ′ The amplitudes corresponding to all frequencies in the local area of ​​the fault characteristic frequency, excluding the first N ′ times the fault characteristic frequency.

7. The method for self-adaptive fault diagnosis of a low-speed and heavy-load bearing according to claim 6, characterized in that: Step S3 specifically includes: S31, calculating the fast spectral coherence of the cleaned bearing vibration time domain data, that is: Among them, α represents the cycle frequency reflecting the fault characteristic frequency, f represents the carrier frequency reflecting the spectrum frequency, and γ x represents the spectral coherence function, S x represents the spectral correlation function, x(t) represents the vibration time domain signal of the bearing cleaned at time t, τ represents the time lag, R x represents the instantaneous autocorrelation function of the cleaned bearing vibration time domain signal, e represents the exponential function, j represents the imaginary unit, x(t-τ) represents the result of the cleaned bearing vibration time domain signal after time delay, (·) * represents the complex conjugate operator, and E represents the expectation operator; S32, based on the fast spectral coherence of the cleaned bearing vibration time domain data, divide the spectral frequency slices in sequence along the spectral frequency direction, identify the potential candidate fault characteristic frequency of each spectral frequency slice, and use the potential candidate fault characteristic frequency as the label and weight of each spectral frequency slice; The specific process of identifying the potential candidate fault characteristic frequency of each spectral frequency slice is as follows: Select several local maximum values ​​in each spectrum frequency slice, use the multiple relationship test method to select the frequency array that satisfies the approximate multiple relationship, and obtain the potential candidate fault characteristic frequency and the multiple frequency of the potential candidate fault characteristic frequency; Among them, the calculation formula for selecting the frequency array that satisfies the approximate multiple relationship using the multiple relationship test method is: Among them, Γ k″ represents the frequency array that satisfies the approximate multiple relationship, k″ represents the frequency that satisfies the approximate multiple relationship, p1 represents one potential candidate fault feature frequency, M represents the total number of frequencies that satisfy the approximate multiple relationship, and p k″ represents k″ times the potential candidate fault characteristic frequency, f n Indicates the nth carrier frequency, max indicates the maximum value, represents the cycle frequency with an approximate multiple relationship, and λ represents the error tolerance; S33, clustering the spectrum frequency slices with similar labels to obtain clusters of the spectrum frequency slices, and sorting all the clusters of the spectrum frequency slices from large to small according to the weights to obtain sorted clusters of the spectrum frequency slices; S34. According to the sorted clusters of the spectral frequency slices, a cluster containing at least two spectral frequency slices with similar labels is selected as a cluster, and a weighted average of the spectral frequency slices in each selected cluster is used as a denoised envelope spectrum.

8. The method for self-adaptive fault diagnosis of a low-speed and heavy-load bearing according to claim 7, characterized in that: The calculation formula of the weight in step S32 is: Among them, W represents weight, M p represents the total number of selected local maxima, m′ represents the local maximum, α m′ represents the cycle frequency with the local maximum value m′, m1 represents the index value of one potential candidate fault frequency in the cycle frequency, min represents taking the minimum value, and st represents satisfying it.

9. The method for self-adaptive fault diagnosis of a low-speed and heavy-load bearing according to claim 8, characterized in that: The calculation formula of the noise reduction envelope spectrum in step S34 is: Among them, CWES h represents the denoised envelope spectrum of the h-th cluster, I h represents the number of spectral frequency slices in the hth cluster, ω represents the weight function of the spectral frequency slice, c h (i) represents the i-th spectral frequency slice of the h-th cluster.

10. The method for self-adaptive fault diagnosis of a low-speed and heavy-load bearing according to claim 9, characterized in that: Step S4 specifically includes: S41, using a peak search method to perform data processing on the noise reduction envelope spectrum, to obtain a true rotation frequency of the noise reduction envelope spectrum and a multiple of the true rotation frequency of the noise reduction envelope spectrum, a true fault characteristic frequency of the noise reduction envelope spectrum and a multiple of the true fault characteristic frequency of the noise reduction envelope spectrum; S42, obtaining the amplitude of the real fault characteristic frequency of the noise reduction envelope spectrum and the multiple frequency of the real fault characteristic frequency of the noise reduction envelope spectrum, calculating the fault energy of the noise reduction envelope spectrum, and simultaneously obtaining the amplitude of other spectral lines of the noise reduction envelope spectrum, and calculating the fault threshold of the noise reduction envelope spectrum; S43, judging whether the fault energy of the noise reduction envelope spectrum is greater than the fault threshold of the noise reduction envelope spectrum, if so, the diagnosis result of the noise reduction envelope spectrum is the corresponding bearing fault type, otherwise, the diagnosis result of the noise reduction envelope spectrum is that the bearing is normal; S44. Calculate the reliability of the diagnosis result of the denoising envelope spectrum according to the diagnosis result of the denoising envelope spectrum.

Images