A variable speed rolling bearing fault diagnosis method

Abstract

This invention discloses a method for diagnosing faults in variable-speed rolling bearings. Based on the rotational speed signal, the method sequentially obtains the zero-crossing time of each pulse of the rising edge, the rotational speed pulse sequence, and the iso-angle time sequence. The iso-angle time sequence is used to perform cubic spline interpolation resampling on the vibration signal to obtain a stationary vibration signal in the angular domain. The stationary vibration signal in the angular domain is divided into several segments. The discrete points of each segment are summed and averaged to obtain the angular domain average signal. The angular domain average signal is decomposed in the frequency band to obtain the signal of each node and the square envelope of each node signal. The unbiased autocorrelation of the square envelope of each node is calculated, along with a comprehensive index of unbiased autocorrelation composed of the sum of kurtosis, power spectral entropy, and correlation coefficient. The order that is prominent and has a domino relationship in the order square envelope spectrum is compared with the theoretical fault order to determine whether the rolling bearing has failed. This method effectively eliminates the interference of non-periodic noise and enhances the stability of the index.

Description

Technical Field

[0001] This invention belongs to the field of fault diagnosis of mechanical products, specifically a fault diagnosis method for rolling bearings that play a supporting and lubricating role in rotating machinery. Background Technology

[0002] Rolling bearings are one of the essential key components for the normal operation of machinery. Failure of rolling bearings may cause concurrent failures in the entire system, resulting in huge economic losses and safety problems. Moreover, rolling bearings are also the most vulnerable rotating components in rotating machinery. Therefore, condition monitoring and fault diagnosis of rolling bearings are of great significance for the normal operation of the entire machine and industrial safety production.

[0003] Fault diagnosis of rolling bearings has been widely used in industry. These diagnostic techniques can be basically divided into time-domain analysis, frequency-domain analysis, and time-frequency-domain analysis methods. These methods are effective in processing stationary signals, but when used to process non-stationary signals, the spectrum will show a phenomenon of "frequency ambiguity," making effective fault diagnosis impossible. Since the vibration signals of rolling bearings are often non-stationary, these methods are not applicable.

[0004] For non-stationary signals, industrial processes typically employ order analysis to process them. This involves first converting the non-stationary signal into a stationary signal in the angular domain, followed by further analysis using traditional time-frequency domain methods. While order analysis is an effective method for handling non-stationary vibration signals, the complex environment of industrial production sites, characterized by strong noise and interference, makes it susceptible to interference from non-order noise from other components. This can lead to reduced diagnostic effectiveness or even failure to diagnose the problem. Generally, order analysis requires signal filtering, using bandpass filtering to remove irrelevant frequencies. However, the selection of filter parameters often relies on manual experience, significantly impacting the analysis results. Therefore, the key to this field is how to adaptively filter the signal and separate the bearing's inherent state information against a strong noise background.

[0005] Chinese Patent Publication No. CN111307460A discloses a method for diagnosing rolling bearing faults based on order tracking and spectral kurtosis. This method performs tree-like decomposition filtering on the time-domain vibration signal to obtain signals for each frequency band, calculates the spectral kurtosis and plots a spectral kurtosis diagram, calculates the center frequency and optimal demodulation bandwidth of the optimal demodulation bandwidth, performs bandpass filtering on the time-domain vibration signal based on the optimal demodulation bandwidth, performs Hilbert envelope demodulation on the filtered signal to obtain an envelope signal containing fault impacts, sets a speed pulse threshold to obtain the pulse occurrence time, solves the coefficients of the quadratic time equation in the angle domain, obtains the stationary signal in the angle domain, performs a fast Fourier transform on the stationary signal in the angle domain, sets the sampling frequency as the angle domain sampling rate, and draws a fault diagnosis conclusion. While this method can extract the order components of bearing fault features under varying operating conditions, the spectral kurtosis calculation is time-consuming, making it inconvenient for engineering applications. Furthermore, it significantly impacts the diagnostic effectiveness when the vibration signal has a low signal-to-noise ratio or contains non-Gaussian noise.

[0006] Chinese Patent Publication No. CN111238843A discloses a method for evaluating the health of wind turbines based on fast spectral kurtosis analysis. First, vibration signals from the wind turbine under different operating conditions are collected to construct an original database. Effective statistical information is extracted from the original vibration signals. For the signal to be tested, the kurtosis value of the signal at different frequency positions is calculated to obtain the center frequency and bandwidth of the optimal demodulation frequency band of the wind turbine vibration signal for signal filtering. The frequency domain information is obtained by performing a Fourier transform on this time-domain information. The vibration information of the wind turbine under test is compared and analyzed with the vibration data under different operating conditions in the database to obtain the health factor of the wind turbine under test, thus reflecting the health level of the wind turbine. However, this evaluation method directly applies fast spectral kurtosis analysis to the time-domain signal and performs a Fourier transform on the filtered signal, resulting in a "frequency ambiguity" phenomenon in the spectrum, making it impossible to effectively extract fault information from the variable speed vibration signal. Furthermore, when the vibration signal contains non-Gaussian noise, it affects the analysis effect of fast spectral kurtosis. Summary of the Invention

[0007] The purpose of this invention is to address the shortcomings of the aforementioned fault diagnosis techniques by providing a fault diagnosis method for variable speed rolling bearings that offers accurate results and high diagnostic efficiency.

[0008] The technical solution of the variable speed rolling bearing fault diagnosis method of the present invention includes the following steps:

[0009] Step 1): Sample the vibration signal and rotation speed signal of the rolling bearing, and obtain the zero-crossing time of each pulse of the rising edge, the rotation speed pulse sequence, and the equal angle time sequence based on the rotation speed signal.

[0010] Step 2): The vibration signal is resampled using cubic spline interpolation using the aforementioned equal-angle time series to obtain a stationary vibration signal in the angular domain;

[0011] Step 3): Divide the angular domain steady vibration signal into several segments with a period of several revolutions, add the discrete points of each segment and take the average value to obtain the angular domain average signal.

[0012] Step 4): Decompose the angular domain average signal in the frequency band to obtain the signal of each node and the squared envelope of each node signal;

[0013] Step 5): Calculate the unbiased autocorrelation of the squared envelope of each node, and the comprehensive index of unbiased autocorrelation composed of the sum of kurtosis, power spectral entropy and correlation coefficient;

[0014] Step 6): Perform a Fourier transform on the node corresponding to the maximum value of the comprehensive index to obtain the order squared envelope spectrum.

[0015] Step 7): Compare the orders that are dominant and prominent in the order square envelope spectrum with the theoretical fault order of the rolling bearing to determine whether the rolling bearing has failed.

[0016] The beneficial effects of adopting the above technical solution in this invention are:

[0017] 1. The present invention uses a diagnostic method that combines order analysis, time-domain averaging techniques, and an improved autocorrelation spectrum kurtosis plot method to better eliminate the phenomenon of "frequency ambiguity".

[0018] 2. The present invention further processes the order signal using a time-domain averaging method, which can effectively eliminate the interference of non-periodic noise, eliminate signal components unrelated to the periodic signal, including noise and random signals, and improve the signal-to-noise ratio of signal analysis.

[0019] 3. This invention employs an improved adaptive spectral kurtosis method. Leveraging the high level of second-order cyclostationarity of bearing signals, it calculates a comprehensive index of the unbiased autocorrelation of the squared envelope of the demodulated signal, rather than simply calculating kurtosis. Compared to using kurtosis as the sole criterion for optimal tuning frequency band, this comprehensive index, combining kurtosis, power spectral entropy, and correlation coefficient, effectively addresses the problem of insufficient detection of different vibration signal characteristics due to a single index. It enhances the stability of the index, better identifies the optimal tuning frequency band, and, combined with order analysis, effectively extracts bearing fault features even in situations with low signal-to-noise ratios or the presence of non-Gaussian noise, thereby improving fault diagnosis accuracy. Attached Figure Description

[0020] Figure 1 This is a flowchart of the method of the present invention;

[0021] Figure 2The waveform of the vibration signal of the rolling bearing is shown in the time domain.

[0022] Figure 3 The waveform diagram shows the rotational speed of the rolling bearing.

[0023] Figure 4 This is a diagram of the rotational speed pulse signal of a rolling bearing.

[0024] Figure 5 This is a resampled image of the time-domain vibration signal in the angular domain.

[0025] Figure 6 The waveform of the output signal after applying the time-domain averaging method is shown.

[0026] Figure 7 A comprehensive index spectrum of the order signal;

[0027] Figure 8 This is the squared envelope spectrum of order. Detailed Implementation

[0028] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments and accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] See Figure 1 This invention first synchronously samples the vibration signal and rotational speed of the rolling bearing. Based on the rotational speed signal, it sequentially obtains the zero-crossing time of each rising pulse, the rotational speed pulse sequence, and the constant-angle time sequence. Utilizing the relationship between rotational angle and time, the time-domain vibration signal is resampled in the angular domain, and cubic spline interpolation is performed for resampling to obtain a stationary angular domain vibration signal. Order analysis is then used to transform the time-domain variable-speed vibration signal into a stationary angular domain vibration signal, yielding an order signal. The order signal is processed using a time-domain averaging method, dividing the stationary angular domain vibration signal into several segments with a period of several revolutions. The discrete points of each segment are summed and averaged to obtain the angular average signal, reducing the influence of non-periodic signals and improving the signal-to-noise ratio of the analyzed signal. Finally, an improved adaptive spectral kurtosis diagram is used to analyze the stationary angular domain vibration signal, decomposing it in the frequency band to obtain the signal at each node and the squared envelope of each node signal. Using a comprehensive index combining the kurtosis, power spectral entropy, and correlation coefficient of each node in the decomposed signal as a standard, fault features in the vibration signal are extracted. Fourier transform is performed on the node corresponding to the maximum value of the comprehensive index to obtain the order squared envelope spectrum. Finally, the order squared envelope spectrum is compared with the theoretical fault order of the rolling bearing, enabling fault diagnosis and fault type determination for variable speed vibration signals. The specific steps are:

[0030] The vibration and rotational speed signals of the rolling bearing were collected using an acceleration vibration sensor and a rotational speed sensor, respectively, with a synchronous sampling frequency of 200 kHz, as shown below. Figure 2 The vibration signal time-domain waveform diagram shown and Figure 3 The shown rotational speed waveforms are as follows: the time-domain waveform of the vibration signal is plotted with time on the x-axis and vibration amplitude on the y-axis; the rotational speed waveform is plotted with time on the x-axis and rotational speed on the y-axis. Figure 3 It can be seen that the rotational speed of the shaft with the rolling bearing installed first decreases and then increases, which is a non-steady vibration signal.

[0031] Based on the structural parameters of the rolling bearing, the theoretical failure order of the rolling bearing can be calculated using the following formula, including the theoretical failure order of the cage, the theoretical failure order of the rolling elements, the theoretical failure order of the outer ring, and the theoretical failure order of the inner ring:

[0032]

[0033]

[0034]

[0035]

[0036] In the formula, O F O B O O and O I These are the theoretical failure orders of the cage, rolling elements, outer ring, and inner ring of the rolling bearing, respectively. n is the number of rolling elements, d is the diameter of the rolling element, D is the pitch circle diameter, and α is the radial contact angle of the rolling element.

[0037] The structural parameters of a certain rolling bearing are shown in the table below:

[0038]

[0039] Based on the structural parameters, the theoretical failure order O of the cage can be obtained. F The value is 0.397, and the theoretical failure order of the rolling element is O. B The value is 2.32, and the theoretical fault order of the outer ring is O. O The value is 3.572, and the theoretical fault order of the inner ring is O. I It is 5.427.

[0040] based on Figure 3 By processing the speed waveform and determining the zero-crossing time t of each pulse on the rising edge of the speed curve, we can obtain the zero-crossing times t1, t2, and t3 of three consecutive speed pulses.

[0041] Based on the rotational speed waveform, the following is obtained: Figure 4 The rotational speed pulse sequence shown can also be directly acquired using this invention as a reference for measuring the vibration phase angle. Based on the rotational speed pulse sequence, the corresponding rotational angle increment is obtained. Then, the undetermined coefficients b0, b1, and b2 are calculated using the following formula:

[0042]

[0043] in, t1 represents the rotation angle increment corresponding to one rotation speed pulse, and t2 and t3 represent the zero-crossing times of three consecutive rotation speed pulses.

[0044] Calculate a rotation angle θ based on the undetermined coefficients b0, b1, and b2. k The corresponding time t k ;

[0045]

[0046] In the formula: t k θ represents the time corresponding to the rotation angle. k For the corner. All the t... k Form a set, t k The set is an equal-angle time series, and thus, the equal-angle time series is obtained based on the relationship between rotation angle and time in the speed signal.

[0047] By using a constant-angle time series to perform cubic spline interpolation resampling on the vibration signal, the time-domain non-stationary vibration signal is transformed into the following angular-domain stationary vibration signal X(θ), where θ is the angle in the angular-domain stationary vibration signal, i.e., the order signal, as shown below. Figure 5 The signal shown;

[0048]

[0049] In the formula, t i <t k <t i+1 , t i and t i+1 For the rotation angle θ k Corresponding time t k In the time-domain vibration signal, two adjacent time points, x(t) i ) and x(t i+1 () represents the time-domain vibration signal at time t i t i+1 The amplitude corresponding to the time is X(θ), which is the angular domain stationary vibration signal obtained after equal angle interpolation and resampling.

[0050] The stationary vibration signal X(θ) in the angular domain is further analyzed using the time-domain averaging method. The signal is divided into several segments (N segments, where N is greater than 2) with a period of m cycles. The discrete points of each segment are summed, and the arithmetic mean is taken to obtain the angular domain average signal Y(θ). In this invention, m is set to 10. Figure 6 As shown. This can eliminate the influence of random interference signals and signals with non-specified periods in the order signal, thereby improving the signal-to-noise ratio of the analyzed signal:

[0051] X(θ)=x1(θ)+x2(θ)+……+x N (θ)+r(θ),

[0052]

[0053] In the formula: x j (θ) represents the stationary vibration signal of each segment of the angular domain, r(θ) represents the signal with less than m remaining cycles, and j = 1, ..., N.

[0054] Based on the binary tree principle, the diagonal-domain averaged signal Y(θ) is decomposed in the frequency band using the maximum overlap discrete wavelet transform. At each decomposition level, a series of signals are generated, each corresponding to a frequency band and a center frequency, called a node y(θ). The squared envelope y(θ) of the decomposed node signals is then obtained. c ), θ c It is the angle of the square envelope.

[0055] Calculate the squared envelope y(θ) of each node in the frequency band. c The unbiased autocorrelation of the signal can suppress the interference of noise and random impulses. In addition, the part directly related to the fault is enhanced.

[0056]

[0057] In the formula: B(c) is the unbiased autocorrelation of the nodal squared envelope, y(θ) c ) represents the squared envelope of the nodal signals obtained from the decomposition; τ = q / f s τ is the delay factor, f s q is the order sampling frequency; q = 0, 1, ..., M-1, M is the length of the angular domain average signal Y(θ).

[0058] Based on the squared envelope y(θ) of each node c The unbiased autocorrelation B(c) of each node is calculated, and the comprehensive index of the unbiased autocorrelation B(c) of each node is composed of three indices: kurtosis K, power spectral entropy h, and correlation coefficient R. The calculation formulas for the three indices are as follows:

[0059]

[0060]

[0061]

[0062]

[0063] In the formula, K is the kurtosis, H is the reciprocal of the power spectral entropy h, R is the correlation coefficient, B(c) is the unbiased autocorrelation of the nodes, and E c is the power spectral energy value of each node, and E is the sum of the energies of all nodes.

[0064] Then, the kurtosis K, the reciprocal H of the power spectral entropy h, and the correlation coefficient R are each normalized to obtain the normalized kurtosis. The reciprocal of the power spectral entropy h and correlation coefficient

[0065]

[0066]

[0067]

[0068] Based on the normalized kurtosis The reciprocal of the power spectral entropy h and correlation coefficient Calculate the sum of the three factors, which will serve as the comprehensive index I:

[0069]

[0070] Based on comprehensive index I, construct a comprehensive index spectrum, such as... Figure 7 As shown, the optimal frequency band used to determine the best modulation, for example, in Figure 7 In the process, at a decomposition level of 5, the maximum value of the comprehensive index is 1.6, the central order is 200, and the optimal bandwidth is 16.

[0071] Perform a Fourier transform on the node corresponding to the maximum value of the comprehensive index I to obtain the squared envelope spectrum of order, as shown below. Figure 8 As shown, the order is 5.402, its second harmonic order is 10.81, and its third harmonic order is 16.21.

[0072] The orders that are dominant and prominent in the order square envelope spectrum are compared with the theoretical fault order O of the rolling bearing. F O B O O O IBy comparing the order of the most prominent, harmonic-related order in the squared envelope spectrum with the theoretical fault order, the error is calculated. Based on the error range, it is determined whether the rolling bearing has failed. The theoretical fault order O is compared... F O B O O O I The fault type can be determined. This invention sets the error range to be less than 2%. If the order is prominent in the square envelope spectrum and has a harmonic relationship, and the error range between the order and the theoretical fault order is within 2%, then the rolling bearing is judged to have a fault, otherwise there is no fault.

[0073] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for diagnosing faults in variable speed rolling bearings, characterized in that: Includes the following steps: Step 1): Sample the vibration signal and rotation speed signal of the rolling bearing, and obtain the zero-crossing time of each pulse of the rising edge, the rotation speed pulse sequence, and the equal angle time sequence based on the rotation speed signal. Step 2): The vibration signal is resampled using cubic spline interpolation using the aforementioned equal-angle time series to obtain a stationary vibration signal in the angular domain; Step 3): Divide the angular domain steady vibration signal into several segments with a period of several revolutions, add the discrete points of each segment and take the average value to obtain the angular domain average signal. Step 4): Decompose the angular domain average signal in the frequency band to obtain the signal of each node and the squared envelope of each node signal; Step 5): Calculate the unbiased autocorrelation of the squared envelope of each node, and the comprehensive index of unbiased autocorrelation composed of the sum of kurtosis, power spectral entropy and correlation coefficient; Step 6): Perform a Fourier transform on the node corresponding to the maximum value of the comprehensive index to obtain the order squared envelope spectrum. Step 7): Compare the orders that are dominant and prominent in the order square envelope spectrum with the theoretical fault order of the rolling bearing to determine whether the rolling bearing has failed.

2. The method for diagnosing faults in variable speed rolling bearings according to claim 1, characterized in that: In step 1), the rotation angle increment corresponding to the corresponding rotation pulse is first obtained based on the aforementioned rotation pulse sequence. Then according to the formula Find the undetermined coefficients b0, b1, and b2, and calculate a rotation angle θ. k The corresponding time All time t k The resulting set is the aforementioned equal-angle time series.

3. The method for diagnosing faults in variable speed rolling bearings according to claim 1, characterized in that: In step 2), the angular domain stationary vibration signal t i and t i+1 For the rotation angle θ k Corresponding time t k In the vibration signal, at two adjacent time points, x(t) i ) and x(t i+1 ) represents the vibration signal at time t i t i+1 The amplitude corresponding to the time.

4. The method for diagnosing faults in variable speed rolling bearings according to claim 1, characterized in that: In step 3), the stationary vibration signal in the angular domain is divided into: X(θ)=x1(θ)+x2(θ)+......+x N (θ)+r(θ), the angular domain average signal x j (θ) represents the stationary vibration signal of each segment of the angular domain, r(θ) represents the signal with fewer than a few remaining revolutions, and j = 1, ..., N.

5. The method for diagnosing faults in variable speed rolling bearings according to claim 1, characterized in that: In step 5), the unbiased autocorrelation of the squared envelope of each node... y(θ c Let τ be the squared envelope of the nodal signal, and τ = q / f s τ is the delay factor, f s It is the sampling frequency of order, q = 0, 1, ..., M-1, and M is the length of the angular domain average signal Y(θ).

6. The method for diagnosing faults in a variable-speed rolling bearing according to claim 1, characterized in that: In step 5), the kurtosis K, power spectral entropy h, and correlation coefficient R are respectively: The reciprocal of the power spectral entropy h E c is the power spectral energy value of each node, and E is the sum of the energies of all nodes.

7. The method for diagnosing faults in variable speed rolling bearings according to claim 6, characterized in that: Normalize each index—kurtosis K, the reciprocal H of power spectral entropy h, and correlation coefficient R—to obtain the normalized kurtosis. The reciprocal of the power spectral entropy h and correlation coefficient Comprehensive indicators 8. The method for diagnosing faults in variable speed rolling bearings according to claim 1, characterized in that: In step 5), the theoretical failure order of the rolling bearing includes the theoretical failure order of the cage, the theoretical failure order of the rolling elements, the theoretical failure order of the outer ring, and the theoretical failure order of the inner ring: O F O B O O and O I These are the theoretical failure orders of the cage, rolling elements, outer ring, and inner ring of the rolling bearing, respectively. n is the number of rolling elements, d is the diameter of the rolling element, D is the pitch circle diameter, and α is the radial contact angle of the rolling element.

9. The method for diagnosing faults in variable speed rolling bearings according to claim 1, characterized in that: The error between the prominent order and the theoretical fault order in the square envelope spectrum is calculated, and the rolling bearing fault is determined based on the error range.

10. A method for diagnosing faults in a variable-speed rolling bearing according to claim 9, characterized in that: The error range is less than 2%.

Images