A method of monitoring a rolling bearing for a fault and related devices

Abstract

The application discloses a kind of rolling bearing fault monitoring method and related device.Method includes obtaining healthy benchmark sound signal, using the calculation method of feature row vector to obtain the feature row vector of healthy benchmark sound signal, and using the feature row vector of healthy benchmark sound signal to construct the healthy benchmark model of rolling bearing;Obtain measured sound signal, using the calculation method of feature row vector to obtain the feature row vector of measured sound signal;The feature row vector of measured sound signal is input into the healthy benchmark model of rolling bearing to calculate health index and obtain health index trend curve;After smoothing processing to health index trend curve, combined with criterion and the numerical gradient of health index trend curve, health index alarm threshold is set;Combined with health index trend curve and health index alarm threshold, the fault of rolling bearing is monitored and alarmed.

Description

Technical Field

[0001] This invention belongs to the field of mechanical condition monitoring, and specifically relates to a method and related device for monitoring rolling bearing faults. Background Technology

[0002] When a rolling bearing experiences fatigue pitting failure, the continuous rolling of the rolling elements generates periodic, continuous impacts at the pitting points, thus exciting high-frequency resonance throughout the bearing housing. This results in periodic impact signals within the vibration signal, which are inevitably submerged by irrelevant interference such as impact noise, unrelated periodic components, and background noise in the actual measured signal. Therefore, extracting the periodic impact signal characterizing the fault from the measured signal under the combined effect of multiple interference sources, and constructing a health index sensitive to the periodic impact signal based on this, is crucial for achieving automatic monitoring of rolling bearing faults.

[0003] Mel-scale frequency segmentation is a method used for signal decomposition when calculating Mel-frequency cepstral coefficients. This method first transforms the original linear frequencies into a nonlinear Mel-scale, then divides the Mel-frequency components in the Mel-scale into equally spaced linear segments, and finally uses an inverse transform to convert the aforementioned linear Mel-frequency components back into nonlinear linear frequencies, thus achieving nonlinear frequency segmentation of the signal spectrum. Constructing triangular filters based on Mel-scale frequency segmentation to decompose and extract signal features has become a mainstream method for extracting rolling bearing fault features based on audio signals. However, in practical applications, this method has shortcomings such as not considering the high-frequency resonance characteristics of rolling bearing faults and amplitude suppression at the intersection of adjacent triangular filters.

[0004] Statistical models based on parameter estimation assume that the measured signal follows a specific or mixed distribution, estimate the distribution parameters using training data, and thus construct a baseline distribution model for the signal. The deviation between the current measured signal distribution and the trained baseline distribution is then used as a health indicator for monitoring bearing failures. As a machine learning method, it overcomes the weak fault detection capabilities of traditional methods based on statistical parameters and signal processing, and avoids the difficulties in training and poor interpretability of deep learning methods based on neural networks. However, this method relies on the correct assumption of the signal distribution; incorrect assumptions will lead to erroneous calculations. Furthermore, directly estimating the distribution parameters of the measured signal is susceptible to noise interference in the measured signal.

[0005] Setting health indicator thresholds is a key technology for equipment fault monitoring. Appropriate threshold settings can prevent false alarms and promptly detect equipment faults, providing a reference for maintenance personnel to develop maintenance plans. Currently, health indicator thresholds are mainly set based on manual experience or statistical measurements from historical benchmark data. The former requires a high level of technical expertise from engineers, while the latter only considers the statistical characteristics of historical benchmark data, neglecting the geometric characteristics of health indicator trends. Therefore, thresholds set based on statistical measurements do not fully exploit the performance degradation characteristics inherent in historical benchmark data. Summary of the Invention

[0006] The purpose of this invention is to overcome the problem that the method of constructing a triangular filter based on Mel-scale frequency segmentation to decompose and extract signal features in equipment fault monitoring does not consider the high-frequency resonance characteristics of rolling bearing faults. A rolling bearing fault monitoring method and related device are proposed.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A method for monitoring rolling bearing faults includes the following steps:

[0009] Obtain the health baseline sound signal, which includes the sound signal of the rolling bearing in a healthy state; calculate the feature row vector of the health baseline sound signal using the feature row vector calculation method; and use the feature row vector of the health baseline sound signal to construct the health baseline model of the rolling bearing.

[0010] The measured sound signal is obtained, which includes the rolling bearing sound signal at the current moment. The rolling bearing sound signal at the current moment includes the periodic impact signal of fatigue pitting failure and irrelevant interference noise components. The characteristic row vector of the measured sound signal is calculated using the characteristic row vector calculation method. The characteristic row vector of the measured sound signal is input into the health benchmark model of the rolling bearing to calculate the health index and obtain the health index trend curve.

[0011] The health indicator trend curve is smoothed to obtain a smoothed health indicator trend curve, which is then used... The numerical gradient of the criteria and smoothed health indicator trend curve is used to set the health indicator alarm threshold; the health indicator trend curve and health indicator alarm threshold are used to monitor and alarm for rolling bearing failures.

[0012] The method for calculating feature row vectors includes the following steps:

[0013] The mapping relationship between Mel-scale frequency and line frequency is calculated using Mel-scale frequency curves and symmetric Mel-scale frequency curves; the mapping relationship between Mel-scale frequency and line frequency is segmented and inversely transformed to obtain the line boundary frequency, and a filter is constructed using the line boundary frequency.

[0014] The sound signal is filtered and decomposed through a filter to obtain a narrowband filtered signal. The square envelope of the narrowband filtered signal is calculated, and the envelope Gini coefficient and envelope harmonic ratio of each square envelope signal are calculated. The envelope Gini coefficient and envelope harmonic ratio of each square envelope signal form a feature row vector.

[0015] Furthermore, the symmetric Mel-scale frequency curves with As the axis of symmetry, Let x be the horizontal axis of the Mel-scale frequency curve. The signal sampling frequency;

[0016] The formula for calculating the mapping relationship between Mel-scale frequency and line frequency is shown below:

[0017]

[0018] in, , , and These are the line frequency, Mel-scale frequency, symmetric Mel-scale frequency, and signal sampling frequency, respectively.

[0019] The mapping curve between Mel-scale frequency and line frequency is divided and inversely transformed to obtain the line boundary frequency. Specifically, the mapping curve between Mel-scale frequency and line frequency is divided into N+2 Mel-scale boundary frequencies at equal intervals on the Mel-scale frequency, where N is the frequency band division coefficient, which is always a positive integer. The Mel-scale boundary frequencies are then transformed into the line frequency domain through inverse transformation to obtain 2N+3 line boundary frequencies corresponding to the Mel-scale boundary frequencies.

[0020] The formula for calculating the inverse transform is shown below:

[0021]

[0022] Furthermore, a filter is constructed using the line boundary frequency, specifically as follows:

[0023] Using the Littlewood-Paley wavelet theory, the 2N+3 line boundary frequencies are used as the upper and lower cutoff frequencies of the filter in an overlapping manner to construct 2N+1 filters, including 1 low-pass filter, 2N-1 band-pass filters and 1 high-pass filter.

[0024] The mathematical expressions for each type of filter are shown below:

[0025] Low-pass filter:

[0026]

[0027] Bandpass filter:

[0028]

[0029] High-pass filter:

[0030]

[0031] in, Angular frequency; and These represent the lower cutoff frequency and the upper cutoff frequency, respectively. For transition bandwidth coefficient, ; For function operators, ;

[0032] The filtering process for each type of filter is shown below:

[0033] Low-pass filter:

[0034]

[0035] Bandpass filtering:

[0036]

[0037] High-pass filter:

[0038]

[0039] in, , , and These are the inner product operator, complex conjugate operator, Fourier transform operator, and inverse Fourier transform operator, respectively. , and These are low-pass narrowband filtered signal, band-pass narrowband filtered signal, and high-pass narrowband filtered signal, respectively.

[0040] Furthermore, the formula for calculating the squared envelope is as follows:

[0041]

[0042]

[0043] in, Let be the squared envelope of each narrowband filtered signal. There are 2N+1 narrowband filtered signals. The imaginary unit, For convolution operators, For the first i The analytical signal of a narrowband filtered signal. It is the modulo operator;

[0044] The formulas for calculating the envelope Gini coefficient and envelope harmonic ratio are shown below:

[0045]

[0046]

[0047] in, EGI i and EHR i The first i Envelope Gini coefficient and envelope harmonic ratio of a square envelope signal; The first in ascending order i Discrete sequences of squared envelope signals Num This represents the total number of points in the squared envelope sequence;

[0048] The envelope Gini coefficient and envelope harmonic ratio of each squared envelope signal form a characteristic row vector, as shown in the following equation:

[0049]

[0050] Furthermore, the construction of the health benchmark model for rolling bearings includes the following steps:

[0051] Several sets of sound signals under the healthy state of rolling bearings are collected. The feature row vectors of each set of sound signals under the healthy state of rolling bearings are calculated to obtain several sets of feature row vectors. Several sets of feature row vectors are concatenated to obtain a feature matrix. The normalized probability density function of each column of features in the feature matrix is ​​estimated by a kernel density estimation algorithm based on an iterative algorithm to obtain the probability density function set of the feature matrix. The probability density function set of the feature matrix is ​​used as the health benchmark model of the rolling bearing.

[0052] Furthermore, the formula for calculating health indicators is as follows:

[0053]

[0054] in, HI Indicates health indicators, The first row in the feature vector i Features f i In the health benchmark model i The probability value of a normalized probability density function.

[0055] Furthermore, utilizing The criteria and the numerical gradient of the smoothed health indicator trend curve are used to set the health indicator alarm threshold, specifically:

[0056] Collect health indicators of rolling bearing health status as Threshold benchmark dataset calculation Criterion threshold The formula for calculating the criterion threshold is as follows:

[0057]

[0058] in, for Criterion threshold; and These are the operators for calculating the mean and standard deviation, respectively. HI Indicates health indicators;

[0059] The central difference method is used to calculate the numerical gradient at a certain point in time on the trend curve of health indicators. The formula for calculating the numerical gradient is as follows:

[0060]

[0061] in, and These are the numerical gradient at time point k and the smoothed health index, respectively.

[0062] Will simultaneously satisfy Health indicators whose criterion threshold and numerical gradient are continuously positive are set as health indicator alarm thresholds.

[0063] Smoothing was performed using the Rauch–Tung–Striebel smoother.

[0064] A rolling bearing fault monitoring device includes a reference model construction module for obtaining a health reference sound signal, wherein the health reference sound includes a sound signal indicating the health state of the rolling bearing; the feature row vector of the health reference sound signal is calculated using a feature row vector calculation method; and a health reference model of the rolling bearing is constructed using the feature row vector of the health reference sound signal.

[0065] The health index calculation module is used to obtain the measured sound signal, which includes the rolling bearing sound signal at the current moment. The rolling bearing sound signal at the current moment includes the periodic impact signal of fatigue pitting failure and irrelevant interference noise components. The module calculates the characteristic row vector of the measured sound signal using the characteristic row vector calculation method. The characteristic row vector of the measured sound signal is then input into the health benchmark model of the rolling bearing to calculate the health index and obtain the health index trend curve.

[0066] The detection and alarm module is used to smooth the health indicator trend curve to obtain a smoothed health indicator trend curve. The numerical gradient of the criteria and smoothed health indicator trend curve is used to set the health indicator alarm threshold; the health indicator trend curve and health indicator alarm threshold are used to monitor and alarm for rolling bearing failures.

[0067] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the rolling bearing fault monitoring method described above.

[0068] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the rolling bearing fault monitoring method described above.

[0069] Compared with the prior art, the present invention has the following beneficial technical effects:

[0070] This invention proposes a rolling bearing fault monitoring method. Based on the construction of health indicators, it also provides a method for setting health indicator thresholds, enabling precise monitoring of the rolling bearing's operational health status. The method utilizes Mel-scale-frequency curves and symmetrical Mel-scale-frequency curves to perform frequency segmentation of the original signal, fully considering the characteristics of the high-frequency resonance response of rolling bearing faults. In the high-frequency band, frequency segmentation is performed in a manner equivalent to the original Mel-scale-frequency curve, avoiding noise interference caused by excessively wide frequency segmentation bandwidth in the high-frequency band, which would affect the extraction of rolling bearing fault information. Furthermore, a multi-band feature extraction algorithm is used to calculate the envelope Gini coefficient and envelope harmonic ratio of each narrowband signal, characterizing the periodic impact characteristics of the rolling bearing fault signal.

[0071] Furthermore, this invention replaces the triangular filter used in the original Mel scale-frequency segmentation filter with the Littlewood-Paley wavelet, thus avoiding the amplitude suppression problem at the intersection of the amplitude-frequency response curves of the triangular filter.

[0072] Furthermore, the construction of the health benchmark model for rolling bearings in this invention employs a kernel density estimation algorithm based on an iterative algorithm. This algorithm estimates the probability density function of the multi-band features of the benchmark dataset composed of health data, yielding the probability density function of each dimension of the feature quantity. Next, the health index is constructed by calculating the normalized probability of the multi-band features of the current data in the probability density function model. This method, based on a non-parametric estimation statistical model learning approach, avoids the mismatch between the assumed distribution and the actual distribution, as well as the noise interference problems caused by direct parameter estimation, inherent in parametric estimation-based statistical model health index construction methods.

[0073] Furthermore, the threshold setting proposed in this invention combines... The criteria and health indicator trend analysis, taking into account both the statistical and geometric characteristics of health indicators, can provide more accurate fault alarms. Attached Figure Description

[0074] The accompanying drawings are provided to further understand the invention and constitute a part of this invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0075] Figure 1 This is a flowchart of the health indicator construction method used in the embodiments of the present invention.

[0076] Figure 2(a) is an external view of the rolling bearing full life test bench used in the embodiment.

[0077] Figure 2(b) is a cross-sectional view of the bearing housing of the rolling bearing full life test bench used in the embodiment.

[0078] Figure 3 The diagram shows the unpacking failure of the experimental bearing on the rolling bearing full life test bench used in this embodiment.

[0079] Figure 4(a) is a schematic diagram of Mel-scale frequency segmentation.

[0080] Figure 4(b) is a schematic diagram of symmetrical Mel-scale frequency segmentation.

[0081] Figure 5(a) shows the amplitude-frequency characteristics of the trigonometric function filter.

[0082] Figure 5(b) shows the amplitude-frequency characteristics of the Littlewood-Paley wavelet filter.

[0083] Figure 6 Visualization of the nonparametric estimation statistical model for the example.

[0084] Figure 7 The following is a health indicator trend chart and a magnified view of a portion of the example.

[0085] Figure 8(a) is a square envelope spectrum of alarm time data in the embodiment.

[0086] Figure 8(b) is a square envelope spectrum of the data at the time before the alarm time in the embodiment.

[0087] Figure 8(c) is the square envelope spectrum of the data at the time following the alarm time in the embodiment.

[0088] Figure 9 This is a flowchart of a rolling bearing fault monitoring method according to the present invention.

[0089] Figure 10 This is a structural diagram of a rolling bearing fault monitoring device according to the present invention.

[0090] Figure 11 This is a schematic diagram of an electronic device for a rolling bearing fault monitoring method according to the present invention. Detailed Implementation

[0091] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0092] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0093] Example 1

[0094] See Figure 9 A method for monitoring rolling bearing faults includes the following steps:

[0095] Obtain the health baseline sound signal, which includes the sound signal of the rolling bearing in a healthy state; record the sound signal of the rolling bearing in a normal healthy state as the health baseline; calculate the feature row vector of the health baseline sound signal using the feature row vector calculation method; and use the feature row vector of the health baseline sound signal to construct the health baseline model of the rolling bearing.

[0096] The measured sound signal is obtained, which includes the rolling bearing sound signal at the current moment. The rolling bearing sound signal at the current moment includes the periodic impact signal of fatigue pitting failure and irrelevant interference noise components. The characteristic row vector of the measured sound signal is calculated using the characteristic row vector calculation method. The characteristic row vector of the measured sound signal is input into the health benchmark model of the rolling bearing to calculate the health index and obtain the health index trend curve.

[0097] The health indicator trend curve is smoothed to obtain a smoothed health indicator trend curve, which is then used... The numerical gradient of the criteria and smoothed health indicator trend curve is used to set the health indicator alarm threshold; the health indicator trend curve and health indicator alarm threshold are used to monitor and alarm for rolling bearing failures.

[0098] This embodiment smooths the health indicator trend curve to reduce the impact of noise. Combining preset criteria and the numerical gradient of the health indicator trend curve, a reasonable health indicator alarm threshold is set. The health indicator trend curve is observed in real time. When the health indicator exceeds the alarm threshold, a fault alarm is issued, prompting that the rolling bearing needs to be inspected and maintained.

[0099] The method for calculating feature row vectors includes the following steps:

[0100] The mapping relationship between Mel-scale frequency and line frequency is calculated using Mel-scale frequency curves and symmetric Mel-scale frequency curves; the mapping relationship between Mel-scale frequency and line frequency is segmented and inversely transformed to obtain the line boundary frequency, and a filter is constructed using the line boundary frequency.

[0101] This embodiment uses frequency segmentation to decompose the measured signal into several narrowband signals with different bandwidths and center frequencies. The original signal is frequency segmented using Mel-scale-frequency curves and symmetrical Mel-scale-frequency curves. This fully considers the high-frequency resonance response characteristics generated when the rolling bearing fails. In the high-frequency band, the same frequency division method as the original Mel-scale-frequency curve frequency segmentation method is used to ensure the precision of frequency division and effectively avoid noise interference introduced by excessively wide frequency division bandwidth. This allows for more accurate extraction of rolling bearing fault information.

[0102] The sound signal is filtered and decomposed through a filter to obtain a narrowband filtered signal. The square envelope of the narrowband filtered signal is calculated, and the envelope Gini coefficient and envelope harmonic ratio of each square envelope signal are calculated. The envelope Gini coefficient and envelope harmonic ratio of each square envelope signal form a feature row vector.

[0103] The health indicators and threshold settings constructed in this embodiment can effectively extract fault information from the measured signals, thereby realizing fault monitoring of rolling bearings.

[0104] Example 2

[0105] This embodiment selects a set of sound signals from the entire lifespan of a rolling bearing, from healthy to failure, from a rolling bearing lifespan test bench. Figure 2 shows the arrangement of measurement points on the rolling bearing lifespan test bench, where Figure 2(a) is the overall appearance view of the test bench and Figure 2(b) is a sectional view of the bearing housing. Figure 3 As shown, after the experiment, an inspection revealed damage to the inner raceway of the rolling bearing. Based on the bearing model and rotational speed, the characteristic frequency of the inner raceway fault was determined to be 341.7 Hz. The data in this example was acquired using a microphone, with a sampling frequency, number of sampling points, and sampling interval of 25.6 kHz, 51200, and 30 seconds, respectively. Using this set of full-lifetime sound data as the analysis object, this example is described below:

[0106] A method for monitoring rolling bearing faults employs a method for constructing health indicators for rolling bearing fault monitoring, such as... Figure 1 As shown, it includes the following steps:

[0107] Step 1: Load and read the rolling bearing sound signal;

[0108] Step two, by comparing the Mel scale-frequency curve with the symmetrical Mel scale-frequency curve (using... As the axis of symmetry, The signal sampling frequency, represent (Axis) Establish the mapping relationship between Mel-scale frequencies and line frequencies, and calculate the formula as follows:

[0109]

[0110] in , and These are the linear frequency, Mel-scale frequency, and symmetric Mel-scale frequency, respectively.

[0111] All data in this embodiment ;

[0112] The symmetrical Mel scale-frequency curve represents a compensation for the problem of excessively wide high-frequency band division in the original Mel scale. Symmetry is used to map the low-frequency band division characteristics of the Mel scale to the high-frequency band, ensuring dense frequency band division in both high and low frequency bands. The reason for this is that the fault characteristics of rolling bearings can be understood as information characterizing the fault, which usually manifests as high-frequency resonance characteristics, especially in the early stage of the fault. Therefore, dense frequency band division in both high and low frequency bands can avoid the loss of fault information caused by excessively wide frequency band division.

[0113] Step 3: The Mel-scale frequency to line frequency mapping curve obtained in Step 2 is divided into N+2 equally spaced boundary frequencies at the Mel-scale frequency domain. These boundary frequencies are then transformed back to the line frequency domain via an inverse transform, resulting in 2N+3 corresponding line boundary frequencies. The inverse transform calculation formula is as follows:

[0114]

[0115] In this embodiment, all data N=11. Figures 4(a) and 4(b) show a comparison of frequency segmentation using Mel-scale and symmetric Mel-scale segmentation. It can be seen that the symmetric Mel-scale frequency segmentation method not only increases the density of high-frequency band segmentation but also increases the density of mid-frequency band segmentation to a certain extent, preventing fault information from being submerged in irrelevant interference noise due to excessively wide segmentation bandwidth.

[0116] Step four: Based on the Littlewood-Paley wavelet theory, and using the 2N+3 line frequencies obtained in step three as overlapping upper and lower cutoff frequencies of the filters, construct one low-pass filter, 2N-1 band-pass filters, and one high-pass filter. The mathematical expressions for each type of filter are as follows:

[0117] ①Low-pass filter:

[0118]

[0119] ② Bandpass filter:

[0120]

[0121] ③ High-pass filter:

[0122] In the above formula Angular frequency; and These represent the lower cutoff frequency and the upper cutoff frequency, respectively. The transition bandwidth coefficient should satisfy the following conditions: ; The function operator is shown in the following expression:

[0123]

[0124] Figures 5(a) and 5(b) show the amplitude-frequency characteristics of the triangular filter bank and the Littlewood-Paley wavelet filter bank. It can be observed from the figures that the amplitude-frequency characteristics of each filter constructed by the Littlewood-Paley wavelet do not exhibit amplitude suppression at the interface.

[0125] Step 5: Measure the sound signal obtained in Step 1. The 2N+1 filters constructed in step four are used for filtering decomposition, resulting in a total of 2N+1 narrowband filtered signals. The mathematical expression for the filtering process is as follows:

[0126] ① Low-pass filter:

[0127]

[0128] ② Bandpass filtering:

[0129]

[0130] ③ High-pass filtering:

[0131]

[0132] In the above formula , , and These are the inner product operator, complex conjugate operator, Fourier transform operator, and inverse Fourier transform operator, respectively. , and These are low-pass narrowband filtered signal, band-pass narrowband filtered signal, and high-pass narrowband filtered signal, respectively.

[0133] Step six, denoted as the 2N+1 narrowband filtered signals obtained in step five. According to the definition of square envelope theory, the square envelope of each narrowband filtered signal is calculated sequentially and denoted as . The formula for calculating the square envelope is as follows:

[0134]

[0135]

[0136] In the above formula The imaginary unit, For convolution operators, For the first i The analytical signal of a narrowband filtered signal. It is the modulo operator;

[0137] Step 7: Calculate the envelope Gini coefficient (characterizing the impulsiveness of the signal) and the envelope harmonic ratio (characterizing the periodicity of the signal) for the 2N+1 squared envelopes obtained in Step 6. The formulas for calculating the envelope Gini coefficient and envelope harmonic ratio are shown below:

[0138]

[0139]

[0140] In the above formula EGI i and EHR i The first i Envelope Gini coefficient and envelope harmonic ratio of a square envelope signal; The first in ascending order i Discrete sequences of squared envelope signals Num This represents the total number of points in the squared envelope sequence; the envelope Gini coefficient and envelope harmonic ratio of each squared envelope signal are combined to form a feature row vector, as shown below:

[0141]

[0142] Step 8: Collect L sets of sound signals from rolling bearings under healthy conditions. Calculate the feature row vectors for each data set as described in steps 2 through 7. Concatenate these L sets of feature row vectors to form a feature matrix, which will be used as the training dataset for subsequent steps. The mathematical expression for the feature matrix is ​​shown below:

[0143]

[0144] In this embodiment, L=100;

[0145] Step nine: Treat each column of the feature matrix obtained in step eight as a set of random variables. X A kernel density estimation algorithm based on iterative algorithms is used to estimate... X The probability density function of the feature matrix is ​​used as the health baseline model for rolling bearings, which is then used for subsequent model inference and health index calculation. The specific steps are as follows:

[0146] ① Establish a kernel density estimation model;

[0147] random variable X The kernel density estimation model is shown in the following equation:

[0148]

[0149] In the above formula For random variables X The estimated probability density function; and These represent bandwidth and kernel function, respectively.

[0150] ② Calculate the asymptotic mean square error ( AMISE );

[0151]

[0152] ③Minimize AMISE Find the optimal bandwidth ;

[0153]

[0154] ④ Substitute the kernel function to obtain the optimal bandwidth. expression;

[0155] Using the Gaussian kernel as the kernel function, substitute it into ③ to solve. The expression yields

[0156]

[0157] ⑤ Calculate the optimal bandwidth using an iterative algorithm. ;

[0158] The expression contains h Iterative algorithms can be used to calculate The pseudocode for the algorithm is shown in Table 1 below:

[0159] Table 1. Pseudocode of the Optimal Bandwidth Iteration Algorithm

[0160]

[0161] ⑥ Estimate random variables X probability density function ;

[0162] Gaussian kernel and Substitute Expressions yield random variables X The probability density function is then normalized according to the following formula.

[0163]

[0164] ⑦ Calculate the health benchmark model .

[0165] Repeat steps ①-⑥ for each column of the feature matrix to obtain the healthy baseline model of the training data.

[0166] In this embodiment, Pick ,in , , and These are the length, standard deviation, and interquartile range of the random variable, respectively. It is not a constant value; it varies with the random variable. X It varies depending on the situation. Figure 6The image shown is a visualization of the example model obtained by training a nonparametric estimation statistical model.

[0167] Step 10: Substitute the feature row vectors obtained from Steps 2 to 7 of the current rolling bearing sound signal into the health benchmark model trained in Step 9. Use the probability values ​​of the normalized probability density function of the feature values ​​in each dimension of the health benchmark model as the basis for calculating the health indicators. The calculation formula for the health indicators is as follows:

[0168]

[0169] In the above formula HI Indicates health indicators, The first row in the feature vector i Features f i In the health benchmark model i The probability values ​​of a normalized probability density function;

[0170] Step 11: Use the Rauch-Tung-Striebel smoother to smooth the health indicator trend chart. The Rauch-Tung-Striebel smoother consists of two key steps: forward recursion and backward recursion. Its mathematical expression is shown below:

[0171] ① Forward recursion:

[0172]

[0173] ② Backward recursion:

[0174]

[0175] In the above formula , They are respectively k and k-1 Posterior state estimation at time t; , They are respectively k and k+1 Smooth results over time; , Forward recursion k Prior state estimation and backward recursion at time 1 k+1 Prior state estimation at time step; , Forward recursion at k and k-1 The posterior estimate of the covariance at time t; , They are respectively backward recursion in k andk+1 The posterior estimate of the covariance at time t; , They are respectively k and k+1 Prior estimation of covariance at time points; and They are respectively k Forward Kalman gain and backward Kalman gain at each time step; A, H, Q, R and I These are the state transition matrix, observation matrix, process noise covariance, measurement noise covariance, and identity matrix, respectively.

[0176] In this embodiment, A, H, Q and R The values ​​are 1, 1, 0.1, and 1 respectively;

[0177] Step 12, combined The specific steps for setting thresholds for criteria and health indicator trend analysis are as follows:

[0178] ①Calculation Criterion threshold;

[0179] Collect health indicators of group L under healthy conditions as The threshold benchmark dataset is calculated according to the following formula. Criterion threshold:

[0180]

[0181] The above for Criterion threshold; and These are the operators for calculating the mean and standard deviation, respectively.

[0182] ② Calculation of numerical gradient of health indicator trend curve;

[0183] The central difference method is used to calculate the numerical gradient of the health indicator trend curve at a certain point in time. The calculation formula is as follows:

[0184]

[0185] In the above formula and These are the numerical gradient at time point k and the smoothed health index, respectively.

[0186] ③ Alarm threshold setting;

[0187] The alarm thresholds must be met simultaneously The criterion threshold and numerical gradient are continuously positive, and the pseudocode for their calculation algorithm is shown in Table 2 below:

[0188] Table 2. Pseudocode for Threshold Setting Algorithm

[0189]

[0190] In this embodiment, and The values ​​are 7 and 0.002 respectively.

[0191] Figure 7 The image shows a trend chart and a magnified view of the health indicators and their smoothed curves. Based on the threshold setting algorithm, the data at the time corresponding to file number 404 triggered an alarm.

[0192] Figures 8(a), 8(b), and 8(c) show the square envelope spectra of the data at the alarm time and one time before and after it. As can be seen from Figures 8(a), 8(b), and 8(c), the square envelope spectra of the data at the time corresponding to file number 403 do not show the inner circle fault characteristic frequency, while the square envelope spectra of the data at the time corresponding to file numbers 404 and 405 both show the inner circle fault characteristic frequency, and the amplitude gradually increases, showing a trend of gradual fault evolution.

[0193] The above results show that the rolling bearing health index construction method and threshold setting method proposed in this embodiment can effectively monitor rolling bearing faults and issue timely alarms, providing a basis for maintenance personnel to formulate maintenance plans.

[0194] Example 3

[0195] See Figure 9 A method for monitoring rolling bearing faults includes the following steps:

[0196] Obtain the health reference sound signal, which includes the sound signal of the rolling bearing in a healthy state; calculate the feature row vector of the health reference sound signal using the feature row vector calculation method, and use the feature row vector of the health reference sound signal to construct the health reference model of the rolling bearing;

[0197] The measured sound signal is obtained, which includes the rolling bearing sound signal at the current moment. The rolling bearing sound signal at the current moment includes the periodic impact signal of fatigue pitting failure and irrelevant interference noise components. The characteristic row vector of the measured sound signal is calculated using the characteristic row vector calculation method. The characteristic row vector of the measured sound signal is input into the health benchmark model of the rolling bearing to calculate the health index and obtain the health index trend curve.

[0198] The health indicator trend curve is smoothed to obtain a smoothed health indicator trend curve, which is then used... The numerical gradient of the criteria and smoothed health indicator trend curve is used to set the health indicator alarm threshold; the health indicator trend curve and health indicator alarm threshold are used to monitor and alarm for rolling bearing failures.

[0199] The method for calculating feature row vectors includes the following steps:

[0200] The mapping relationship between Mel-scale frequency and line frequency is calculated using Mel-scale frequency curves and symmetric Mel-scale frequency curves; the mapping relationship between Mel-scale frequency and line frequency is segmented and inversely transformed to obtain the line boundary frequency, and a filter is constructed using the line boundary frequency.

[0201] The sound signal is filtered and decomposed through a filter to obtain a narrowband filtered signal. The square envelope of the narrowband filtered signal is calculated, and the envelope Gini coefficient and envelope harmonic ratio of each square envelope signal are calculated. The envelope Gini coefficient and envelope harmonic ratio of each square envelope signal form a feature row vector.

[0202] The Mel-scale-frequency curve describes the relationship between sound frequency (Hz) and Mel scale (Mel). This curve is not linear but logarithmic; that is, a larger Hertz change in the low-frequency range corresponds to a larger Mel change, while a smaller Hertz change in the high-frequency range corresponds to a smaller Mel change. This relationship better reflects the human ear's perception of pitch, as the human ear has a higher resolution for low-frequency sounds than for high-frequency sounds. At lower frequencies, the Mel value increases rapidly with increasing frequency; at higher frequencies, the rate of increase in the Mel value gradually slows down. This shape ensures that the perceived pitch variation is uniform across the entire audible frequency range.

[0203] Symmetric Mel-scale frequency curves with Let x be the axis of symmetry, and let x be the horizontal axis of the Mel-scale frequency curve. Let be the signal sampling frequency; the formula for calculating the mapping relationship between Mel-scale frequency and line frequency is shown below:

[0204]

[0205] in, , , and These are the line frequency, Mel-scale frequency, symmetric Mel-scale frequency, and signal sampling frequency, respectively.

[0206] The mapping curve between Mel-scale frequency and line frequency is divided and inversely transformed to obtain the line boundary frequency. Specifically, the mapping curve between Mel-scale frequency and line frequency is divided into N+2 Mel-scale boundary frequencies at equal intervals on the Mel-scale frequency, where N is the frequency band division coefficient, which is always a positive integer. The Mel-scale boundary frequencies are then transformed into the line frequency domain through inverse transformation to obtain 2N+3 line boundary frequencies corresponding to the Mel-scale boundary frequencies.

[0207] The formula for calculating the inverse transform is shown below:

[0208]

[0209] Filters are constructed using line boundary frequencies, specifically as follows:

[0210] Using the Littlewood-Paley wavelet theory, the 2N+3 line boundary frequencies are used as the upper and lower cutoff frequencies of the filter in an overlapping manner to construct 2N+1 filters, including 1 low-pass filter, 2N-1 band-pass filters and 1 high-pass filter.

[0211] The Littlewood-Paley wavelet is a wavelet with good characteristics in the frequency domain, but its slow decay rate in the time domain limits its application to some extent. The Littlewood-Paley wavelet theory originated in 1936 with Littlewood and Paley's binary frequency component grouping theory for Fourier series, also known as LP theory. This theory groups Fourier series according to binary frequency components and states that the phase of its Fourier transform does not affect the magnitude and shape of the function.

[0212] The mathematical expressions for each type of filter are shown below:

[0213] Low-pass filter:

[0214]

[0215] Bandpass filter:

[0216]

[0217] High-pass filter:

[0218]

[0219] in, Angular frequency; and These represent the lower cutoff frequency and the upper cutoff frequency, respectively. For transition bandwidth coefficient, ; For function operators, ;

[0220] The filtering process for each type of filter is shown below:

[0221] Low-pass filter:

[0222]

[0223] Bandpass filtering:

[0224]

[0225] High-pass filter:

[0226]

[0227] in, , , and These are the inner product operator, complex conjugate operator, Fourier transform operator, and inverse Fourier transform operator, respectively. , and These are low-pass narrowband filtered signal, band-pass narrowband filtered signal, and high-pass narrowband filtered signal, respectively.

[0228] The formula for calculating the square envelope is as follows:

[0229]

[0230]

[0231] in, Let be the squared envelope of each narrowband filtered signal. There are 2N+1 narrowband filtered signals. The imaginary unit, For convolution operators, For the first i The analytical signal of a narrowband filtered signal. It is the modulo operator;

[0232] The formulas for calculating the envelope Gini coefficient and envelope harmonic ratio are shown below:

[0233]

[0234]

[0235] in, EGI i and EHR i The first i Envelope Gini coefficient and envelope harmonic ratio of a square envelope signal; The first in ascending order i Discrete sequences of squared envelope signals Num This represents the total number of points in the squared envelope sequence;

[0236] The envelope Gini coefficient and envelope harmonic ratio of each squared envelope signal form a characteristic row vector, as shown in the following equation:

[0237]

[0238] The construction of a health benchmark model for rolling bearings includes the following steps:

[0239] Several sets of sound signals under the healthy state of rolling bearings are collected. The feature row vectors of each set of sound signals under the healthy state of rolling bearings are calculated to obtain several sets of feature row vectors. Several sets of feature row vectors are concatenated to obtain a feature matrix. The normalized probability density function of each column of features in the feature matrix is ​​estimated by a kernel density estimation algorithm based on an iterative algorithm to obtain the probability density function set of the feature matrix. The probability density function set of the feature matrix is ​​used as the health benchmark model of the rolling bearing.

[0240] Kernel density estimation (KDE) is a nonparametric method for estimating unknown density functions, combining the ideas of iterative algorithms and kernel density estimation. Proposed by Rosenblatt (1955) and Emanuel Parzen (1962), KDE is a nonparametric method for estimating the probability density function of random variables. It does not require assumptions about the distribution of the data; instead, it directly estimates the density distribution of the data from the data sample using a kernel function. An iterative algorithm is an algorithm that approximates a desired target or result through a repeated feedback process. In each iteration, the algorithm calculates a new value based on the previous result and uses this value as the input for the next iteration. The iterative process continues until a stopping condition is met (such as reaching the maximum number of iterations or result convergence).

[0241] In kernel density estimation algorithms based on iterative algorithms, iteration is mainly used to optimize bandwidth. h or kernel function KThe choice of the kernel function (⋅) is crucial for improving the accuracy and efficiency of density estimation. The convergence and stability of the iterative algorithm are key to its design. It is necessary to ensure that the iterative process consistently converges to the optimal or acceptable solution. The choice of bandwidth h significantly impacts the kernel density estimation results. Excessive bandwidth leads to overly smooth estimation results, losing detailed data information; insufficient bandwidth results in overly coarse estimation results with excessive fluctuations. The choice of the kernel function K(⋅) also affects the estimation results. Different kernel functions have different shapes and properties, suitable for different data distributions and estimation requirements.

[0242] The formula for calculating health indicators is shown below:

[0243]

[0244] in, HI Indicates health indicators, The first row in the feature vector i Features f i In the health benchmark model i The probability value of a normalized probability density function.

[0245] use The criteria and the numerical gradient of the smoothed health indicator trend curve are used to set the health indicator alarm threshold, specifically:

[0246] Collect health indicators of rolling bearing health status as Threshold benchmark dataset calculation Criterion threshold The formula for calculating the criterion threshold is as follows:

[0247]

[0248] in, for Criterion threshold; and These are the operators for calculating the mean and standard deviation, respectively. HI Indicates health indicators;

[0249] The Raida criterion, also known as the standard deviation criterion, is a commonly used method in statistics and signal processing to identify and remove outliers or noise from data. The "3" here refers to the standard deviation (S / S). The mean (μ) is a multiple of the mean (μ), which is a quantitative indicator of the range or dispersion of data. In a normal distribution (Gaussian distribution), approximately 99.73% of the data will fall within a range of the mean (μ) plus or minus 3 standard deviations (μ). This means that if the value of a data point is outside this range, i.e. less than μ-, it is considered to be outside the range. or greater than μ+ If it is an outlier or noise, it will be considered an outlier or noise and may be removed or subject to special processing.

[0250] The central difference method is used to calculate the numerical gradient at a certain point in time on the trend curve of health indicators. The formula for calculating the numerical gradient is as follows:

[0251]

[0252] in, and These are the numerical gradient at time point k and the smoothed health index, respectively.

[0253] Will simultaneously satisfy Health indicators whose criterion threshold and numerical gradient are continuously positive are set as health indicator alarm thresholds.

[0254] Smoothing was performed using the Rauch–Tung–Striebel smoother.

[0255] The Rauch–Tung–Striebel (RTS) smoother is a state smoothing algorithm for nonlinear systems, also known as RTS smoothing or RTS smoother. This algorithm extends the Kalman filter to estimate the system state using the entire observation sequence, thus improving the accuracy of state estimation. The Rauch–Tung–Striebel smoother is based on Bayesian state estimation theory, specifically by performing a posterior estimate of the system state after all observation data is obtained. It utilizes the forward prediction and update steps of the Kalman filter and adds a backward smoothing step to incorporate future observation information to correct past state estimates.

[0256] RTS smoothers typically involve the following steps:

[0257] Forward filtering: The Kalman filter algorithm is used for forward filtering to obtain the state estimate and covariance matrix at each time point.

[0258] Backward smoothing: Starting from the last time point, using the results of forward filtering and future observation information, the smoothed state estimate and covariance matrix of each time point are calculated recursively.

[0259] See Figure 10A rolling bearing fault monitoring device includes a reference model construction module for obtaining a health reference sound signal, wherein the health reference sound includes a sound signal indicating the health status of the rolling bearing; the feature row vector of the health reference sound signal is calculated using a feature row vector calculation method; and a health reference model of the rolling bearing is constructed using the feature row vector of the health reference sound signal.

[0260] The health index calculation module is used to obtain the measured sound signal, which includes the rolling bearing sound signal at the current moment. The rolling bearing sound signal at the current moment includes the periodic impact signal of fatigue pitting failure and irrelevant interference noise components. The module calculates the characteristic row vector of the measured sound signal using the characteristic row vector calculation method. The characteristic row vector of the measured sound signal is then input into the health benchmark model of the rolling bearing to calculate the health index and obtain the health index trend curve.

[0261] The detection and alarm module is used to smooth the health indicator trend curve to obtain a smoothed health indicator trend curve. The numerical gradient of the criteria and smoothed health indicator trend curve is used to set the health indicator alarm threshold; the health indicator trend curve and health indicator alarm threshold are used to monitor and alarm for rolling bearing failures.

[0262] See Figure 11 An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the rolling bearing fault monitoring method described above.

[0263] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the rolling bearing fault monitoring method described above.

[0264] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0265] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method of monitoring a rolling bearing fault, characterized in that, The method comprises the following steps: obtaining a health benchmark sound signal, the health benchmark sound signal comprising a sound signal in a healthy state of a rolling bearing; calculating a feature row vector of the health benchmark sound signal by using a feature row vector calculation method on the health benchmark sound signal, and constructing a health benchmark model of the rolling bearing by using the feature row vector of the health benchmark sound signal; obtaining a measured sound signal, the measured sound signal comprising a current sound signal of the rolling bearing; the current sound signal of the rolling bearing comprising a periodic impact signal of fatigue pitting fault and a noise of irrelevant interference component; calculating a feature row vector of the measured sound signal by using the feature row vector calculation method on the measured sound signal; inputting the feature row vector of the measured sound signal into the health benchmark model of the rolling bearing to calculate a health index to obtain a health index trend curve; The health index trend curve is smoothed to obtain a smoothed health index trend curve, and the health index alarm threshold is set by using the criterion and the numerical gradient of the smoothed health index trend curve. The health index trend curve and the health index alarm threshold are used to monitor and alarm the fault of the rolling bearing. the feature row vector calculation method comprises the following steps: calculating a mapping relationship curve of a mel scale frequency and a line frequency by using a mel scale frequency curve and a symmetric mel scale frequency curve; performing segmentation on the mapping relationship curve of the mel scale frequency and the line frequency and then performing inverse transformation to obtain a line boundary frequency, and constructing a filter by using the line boundary frequency; filtering and decomposing the sound signal by using the filter to obtain a narrowband filter signal, calculating a square envelope of the narrowband filter signal, and calculating an envelope Gini coefficient and an envelope harmonic ratio of each square envelope signal, respectively; the envelope Gini coefficient and the envelope harmonic ratio of each square envelope signal form a feature row vector.

2. A method of monitoring for a fault in an anti-friction bearing according to claim 1, wherein, the symmetric mel-scale frequency curve has a symmetry axis, a horizontal axis of symmetry of the symmetric mel-scale frequency curve, is the signal sampling frequency; The calculation formula of the mapping relationship curve of the mel scale frequency and the line frequency is as shown in the following formula: wherein, , , and are the linear frequency, the mel-scale frequency, the symmetrical mel-scale frequency and the signal sampling frequency, respectively. performing segmentation on the mapping relationship curve of the mel scale frequency and the line frequency and then performing inverse transformation to obtain a line boundary frequency, specifically: equally interval segmenting the mapping relationship curve of the mel scale frequency and the line frequency on the mel scale frequency into N+2 mel scale boundary frequencies, N is a frequency band segmentation coefficient, and is a positive integer; converting the mel scale boundary frequencies into the line frequency domain by inverse transformation to obtain 2N+3 line boundary frequencies corresponding to the mel scale boundary frequencies; the inverse transformation calculation formula is as shown in the following formula:

3. A method of monitoring for a fault in an anti-friction bearing according to claim 2, wherein, constructing a filter by using the line boundary frequency, specifically: taking 2N+3 line boundary frequencies as upper and lower cutoff frequencies of the filter in a mutually overlapping manner according to the Littlewood-Paley wavelet theory to construct 2N+1 filters, including one low-pass filter, 2N-1 band-pass filters and one high-pass filter; the mathematical expressions of the filters of various types are as shown in the following formula: low-pass filter: band-pass filter: high-pass filter: wherein is an angular frequency; and denote a lower and an upper cut-off frequency, respectively; is a transition bandwidth coefficient, ; is a function operator, ; the filtering processes of the filters of various types are as shown in the following formula: low-pass filtering: band-pass filtering: high-pass filtering: wherein, , , and are the inner product operator, the complex conjugate operator, the Fourier transform operator and the inverse Fourier transform operator, respectively; , and are the low-pass narrowband filtered signal, the band-pass narrowband filtered signal and the high-pass narrowband filtered signal, respectively.

4. A method of monitoring for a fault in an anti-friction bearing according to claim 3, wherein, the square envelope calculation formula is as shown in the following formula: wherein is a square envelope of each narrowband filtered signal, is a 2N+1 narrowband filtered signal, is the imaginary unit, is a convolution operator, is an analytical signal of the i nth narrowband filtered signal, is a modulo operator; the calculation formula of the envelope Gini coefficient and the envelope harmonic ratio is as shown in the following formula: wherein EGI i and EHR i are the envelope Gini coefficient and the envelope harmonic ratio of the i th square envelope signal, respectively; is the discrete sequence of the i th square envelope signal sorted in ascending order, Num denotes the total number of points of the square envelope sequence. the envelope Gini coefficient and the envelope harmonic ratio of each square envelope signal form a feature row vector, specifically as shown in the following formula:

5. A method of monitoring for a fault in an anti-friction bearing according to claim 4, wherein, the construction of the health benchmark model of the rolling bearing comprises the following steps: The sound signals of several groups of rolling bearings in healthy states are collected, the feature row vectors of the sound signals of each group of rolling bearings in healthy states are calculated to obtain several groups of feature row vectors, the several groups of feature row vectors are spliced to obtain a feature matrix, the kernel density estimation algorithm based on an iterative algorithm is used to estimate the normalized probability density function of each column feature in the feature matrix to obtain a probability density function set of the feature matrix, and the probability density function set of the feature matrix is taken as a health benchmark model of the rolling bearing.

6. A method of monitoring for a fault in an anti-friction bearing according to claim 5, wherein, The calculation formula of the health index is as follows: in, HI Indicates health indicators, The first row in the feature vector i Features f i In the health benchmark model i The probability value of a normalized probability density function.

7. A method of monitoring for a fault in an anti-friction bearing according to claim 6, wherein, Utilizing The health index alarm threshold is set by using the criterion and the numerical gradient of the smoothed health index trend curve, specifically: Collect health indicators of rolling bearing health status as Threshold benchmark dataset calculation Criterion threshold The formula for calculating the criterion threshold is as follows: wherein, is a criterion threshold value; and are mean and standard deviation operators, respectively, HI denotes a health indicator; The numerical gradient of a point on the health index trend curve is calculated by using the central difference method, and the calculation formula of the numerical gradient is as follows: wherein, and respectively the numerical gradient at the k time point and the smoothed health indicator; will be met simultaneously setting the health indicator that simultaneously satisfies the criteria threshold and the numerical gradient to be continuously positive as a health indicator alarm threshold; The smoothing is performed by using a Rauch-Tung-Striebel smoother.

8. A rolling bearing fault monitoring device, characterized in that, The benchmark model module is used to obtain a health benchmark sound signal, and the health benchmark sound signal includes sound signals of rolling bearings in healthy states; the feature row vector calculation method is used to calculate the feature row vector of the health benchmark sound signal, and the health benchmark model of the rolling bearing is constructed by using the feature row vector of the health benchmark sound signal; The health index calculation module is used to obtain a measured sound signal, and the measured sound signal includes a current time rolling bearing sound signal; the current time rolling bearing sound signal includes a fatigue pitting fault periodic impact signal and an irrelevant interference component noise; The feature row vector calculation method is used to calculate the feature row vector of the measured sound signal; The feature row vector of the measured sound signal is input into the health benchmark model of the rolling bearing to calculate the health index to obtain a health index trend curve; The detection and alarm module is used to smooth the health indicator trend curve to obtain a smoothed health indicator trend curve. The numerical gradient of the criteria and smoothed health indicator trend curve is used to set the health indicator alarm threshold. The health index trend curve and the health index alarm threshold are used to monitor and alarm the fault of the rolling bearing. 9.An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the rolling bearing fault monitoring method in any one of claims 1-7 when executing the computer program. 10.A computer readable storage medium, storing a computer program, wherein the computer program implements the rolling bearing fault monitoring method in any one of claims 1-7 when executed by a processor.

Images