Precise bearing life cycle fault early warning method based on multi-source vibration signal decoupling

Abstract

This invention belongs to the field of bearing fault monitoring technology, specifically relating to a precision bearing full-lifecycle fault early warning method based on multi-source vibration signal decoupling. The method includes: acquiring vibration signals from three channels of the bearing; calculating the impact activity level for the current time window based on the local energy of all micro-segments of the spatial vector mode sequence and the number of times the spatial vector mode in the spatial vector mode sequence is greater than the mean of the spatial vector mode sequence; calculating the autocorrelation coefficient and skewness coefficient of the spatial vector mode sequence and obtaining the background noise coupling strength; obtaining the envelope spectral entropy of the spatial vector mode sequence and weighting it using the impact activity level and the background noise coupling strength to obtain a fault characteristic index; and outputting a graded early warning based on the fitting slope of the fault characteristic index for multiple consecutive time windows and an early warning threshold. This invention overcomes the masking of weak features by strong background noise, significantly reduces early missed and false alarms, and achieves high-precision monitoring throughout the entire lifecycle.

Description

Technical Field

[0001] This invention relates to the field of bearing fault monitoring technology. More specifically, this invention relates to a method for early warning of precision bearing faults throughout their entire life cycle based on the decoupling of multi-source vibration signals. Background Technology

[0002] With the rapid development of high-end equipment manufacturing, precision bearings, as the core components of rotating machinery, directly determine the processing accuracy and operational safety of the equipment in their health condition. Throughout their life cycle, precision bearings typically undergo a long break-in period, a stabilization period, and a degradation period. In the early stages of degradation, bearings often only exhibit extremely weak spalling or pitting. These weak fault characteristic signals have extremely low energy and are often submerged in the strong background noise generated by equipment operation and the coupling interference of multi-source vibration signals. If these weak characteristics cannot be accurately identified and decoupled for early warning in the early stages, the equipment life will be shortened once the late stage of severe wear is reached.

[0003] Currently, bearing fault monitoring mainly relies on vibration signal analysis and uses fixed thresholds to determine the bearing condition. The basic logic of these methods is to assume that the background noise is Gaussian white noise, filter out high-frequency or low-frequency interference through frequency domain filtering, and identify obvious fault characteristic frequencies.

[0004] Because the energy of early fault impacts is extremely weak, and multi-source vibration signals undergo complex nonlinear coupling in the transmission path, traditional frequency domain filtering cannot effectively separate deterministic mechanical interference from sudden fault impacts. Strong background noise such as gear meshing is often non-Gaussian colored noise with strong periodicity and complex modulation coupling with bearing fault frequencies. This makes it difficult for existing technologies to miss or falsely report in the early stages of bearing degradation, making it difficult to build an accurate health evolution model and achieve accurate trend early warning throughout the entire life cycle. Summary of the Invention

[0005] To address the technical problem that early-stage weak fault characteristics of precision bearings are easily masked by multi-source strong background noise coupling, leading to untimely life-cycle early warning and high false alarm rate, this invention provides a precision bearing life-cycle fault early warning method based on multi-source vibration signal decoupling. The method includes: acquiring vibration signals from three channels of the bearing and dividing them into multiple time windows; calculating the spatial vector magnitude of the vibration signal at each moment in the current time window to obtain a spatial vector magnitude sequence; dividing the sequence into multiple micro-segments, where the sum of the spatial vector magnitudes within each micro-segment constitutes the local energy of each micro-segment; and determining the local energy based on the standard deviation of the local energy of the multiple micro-segments and the sequence... The number of times the spatial vector magnitude in the column is greater than the mean of the sequence is used to calculate the impact activity; the autocorrelation coefficient and skewness coefficient of the sequence are calculated, and the background noise coupling strength is calculated based on the autocorrelation coefficient and the skewness coefficient; the Hilbert transform is performed on the sequence to obtain the envelope spectrum and the envelope spectrum entropy is calculated; the envelope spectrum entropy is weighted using the background noise coupling strength and the impact activity to obtain the fault characteristic index; and graded early warning is performed based on the comparison result between the fault characteristic index and the obtained early warning threshold, and the fitting slope value of the fault characteristic index and the fault characteristic index of multiple consecutive time windows.

[0006] This invention effectively distinguishes between sudden fault impacts and Gaussian colored background noise by separately evaluating the impact activity of local energy and the background noise coupling strength based on autocorrelation and skewness coefficients in the time domain. In the frequency domain, it dynamically corrects the envelope spectral entropy using noise and impact intensity as mutually exclusive weighting factors, and adaptively demodulates a pure fault characteristic index from multi-source coupled signals. By combining the dual judgment of threshold and evolution slope of continuous time window, it significantly reduces the false alarm and false negative rates in the early stage of degradation, and achieves reliable hierarchical early warning.

[0007] Preferably, the impact activity satisfies the expression: In the formula, The impact activity level within the current time window; , The standard deviation and mean of the local energy of all micro-segments within the current time window; The ratio of the number of times the spatial vector magnitude of the spatial vector magnitude sequence is greater than the mean of the spatial vector magnitude sequence within the current time window to the length of the time window; The sensitivity coefficient is denoted as .

[0008] This invention utilizes the physical difference between local energy dispersion and signal pulse density, making the calculation results extremely sensitive to intermittent high-energy bursts. At the same time, by adjusting the sensitivity coefficient, it can adaptively suppress high-frequency dense non-fault random white noise, thereby improving the identification of early weak impact characteristics.

[0009] Preferably, the method for obtaining the autocorrelation coefficient includes: obtaining the number of sampling points for one meshing cycle of the bearing and recording it as... , the first part of the sequence The sampling point is used as the reference subsequence, and the first sampling point is used as the reference subsequence. One to the first Each sampling point is used as a hysteresis subsequence. The number of sampling points in the current time window is given; the Pearson correlation coefficient between the baseline subsequence and the lagged subsequence is calculated to obtain the autocorrelation coefficient.

[0010] This invention utilizes the strong periodicity of background noise such as gear meshing. By extracting sampling points from one meshing cycle, constructing a benchmark and hysteresis subsequences, and calculating the Pearson correlation coefficient, it is possible to obtain the regularity of the vibration waveform after spatial synthesis. This provides a high-confidence data basis for subsequently distinguishing between fault impacts with slippage randomness and deterministic mechanical interference.

[0011] Preferably, the skewness coefficient is obtained by calculating the third-order central moment.

[0012] Preferably, the background noise coupling strength satisfies the expression: In the formula, The background noise coupling strength for the current time window; The autocorrelation coefficient of the spatial vector magnitude sequence within the current time window; The skewness coefficients of the spatial vector modulus sequence within the current time window; These are the weighting coefficients; To take the absolute value; It is a natural exponential function.

[0013] This invention integrates the periodic intensity and symmetric distribution weights of the signal. When encountering strong periodic non-Gaussian noise, it can quickly approach the peak value to indicate strong coupling interference. At the same time, by adjusting the weighting coefficients, the response rate of the skewness coefficient can be flexibly controlled to ensure accurate assessment and quantification of the interference ratio of background noise under complex operating conditions.

[0014] Preferably, the calculation of the envelope spectrum entropy includes: performing Hilbert transform demodulation on the sequence to obtain the envelope signal, and performing fast Fourier transform on the envelope signal to obtain the envelope spectrum; obtaining the probability of the amplitude of each frequency point on the envelope spectrum, and calculating the envelope spectrum entropy using the information entropy formula.

[0015] Preferably, the fault characteristic index satisfies the expression: In the formula, The fault characteristic index for the current time window; The impact activity level within the current time window; The background noise coupling strength for the current time window; The envelope spectral entropy of the current time window; This is the penalty coefficient.

[0016] This invention establishes an adaptive gating weighting mechanism that automatically amplifies features when impact dominates and strongly suppresses interference when noise dominates. Combined with the negative exponential inverse mapping of the envelope spectrum entropy, the energy concentration trend that was originally masked by the high-frequency carrier is nonlinearly amplified exponentially, thereby achieving decoupling and purification of fault features.

[0017] Preferably, the method for obtaining the early warning threshold includes: constructing a benchmark sequence by extracting the fault characteristic index within the initial time window after the precision bearing is put into operation, and using the sum of the products of the mean of the benchmark sequence, the preset safety factor, and the standard deviation of the benchmark sequence as the early warning threshold.

[0018] Preferably, the step of performing graded early warning includes: recording the fault characteristic index of the current time window as... The warning threshold is The fitting slope of the fault characteristic index for the current time window and its preceding consecutive time windows is... ;like and Output a red warning signal; if but or but Output a yellow warning signal; if and No warning signal is triggered.

[0019] Preferably, the fitting slope is obtained by performing a least-squares linear fit on the fault characteristic indices of the current time window and several consecutive time windows preceding it.

[0020] The beneficial effects of this invention are as follows:

[0021] This invention uses the time-domain calculated impact activity and background noise coupling strength as mutual exclusion factors to dynamically correct the frequency domain envelope spectral entropy. This ensures that interference is automatically suppressed when noise dominates throughout the bearing's entire life cycle, and weak impact characteristics are significantly amplified at the initial stage of a fault, thereby extracting pure fault characteristics. Furthermore, by combining the dual verification of the trend fitting slope, this invention solves the pain points of traditional frequency domain filtering, which is prone to missed and false alarms, and achieves high-precision graded early warning throughout the entire life cycle. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating the precision bearing full life cycle fault early warning method based on multi-source vibration signal decoupling in this invention;

[0023] Figure 2This is a schematic diagram illustrating the waveform of the spatial vector mode;

[0024] Figure 3 It is a schematic diagram showing the comparison of early warning indicators. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. 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.

[0026] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0027] This invention discloses a method for early warning of full-life-cycle faults in precision bearings based on decoupling of multi-source vibration signals, referring to... Figure 1 This includes steps S1 to S5:

[0028] S1. Obtain the vibration signals of the bearing through three channels and divide them into multiple time windows.

[0029] It should be noted that early failure impacts in precision bearings are often directional. To prevent a single sensor from missing the fault due to installation angle issues, signals from three orthogonal directions need to be collected simultaneously. Furthermore, to capture early failure impacts, high-frequency continuous sampling and an overlapping sliding window method are required to ensure that the fault features are not truncated at the window edge, thus providing benchmark data for subsequent analysis.

[0030] Specifically, a triaxial high-frequency accelerometer is arranged on a precision bearing housing to synchronously collect continuous vibration signals from three channels: vertical, horizontal, and axial, according to a preset sampling frequency.

[0031] In precision vibration monitoring, in order to obtain an effective analysis bandwidth, the sampling frequency is usually set to 2.56 times the effective bandwidth. In this embodiment, in order to cover the high-frequency resonance region triggered by early failure of precision bearing, an effective analysis bandwidth of at least 10kHz is required. Therefore, in this embodiment, the preset sampling frequency is preferably 25.6kHz. The implementer can adjust it according to the actual working conditions.

[0032] Simultaneously, the three data streams are intercepted according to the preset window length and sliding step size to obtain the vibration signal sequence of the three channels within each time window. To ensure the real-time performance of online monitoring, this invention uses the radix-2 FFT algorithm with the highest computational efficiency for frequency domain conversion, requiring the number of input data points to be an integer power of 2. The frequency resolution for each data point value is calculated. Since the frequency resolution of precision bearings is usually between 10Hz and 20Hz, 2048 is preferred to ensure that the frequency resolution is within the specified range. The sliding step size is selected as 1 / 4 of the window length, i.e., 512, so that the window overlap rate is (2048-512) / 2048=75%. The window overlap rate can ensure that the edge attenuation caused by the Hanning window is compensated by the high-weight area in the center of the next window, and also ensure that the data throughput is moderate.

[0033] Thus, the three-channel vibration signal sequence within each time window was obtained.

[0034] S2. For the current time window, obtain the local energy of all micro-segments of the spatial vector mode sequence; calculate the impact activity based on the standard deviation of the local energy of all micro-segments and the number of times the spatial vector mode in the spatial vector mode sequence is greater than the mean of the spatial vector mode sequence.

[0035] It should be noted that, in order to comprehensively capture impact energy from any direction, the randomness of the fault location leads to significant differences in the projected components of impact energy in different orthogonal directions. If a single channel or independent channel analysis is used, it is easy to miss features due to the mismatch between the sensor installation angle and the main impact direction. Therefore, by constructing a spatial vector model, we can achieve comprehensive capture of impact energy across the entire space. In addition, the transient impact triggered by early weak bearing faults is essentially a non-stationary abrupt signal, which will cause extremely high dispersion in the energy distribution on the local time scale. Conversely, background noise such as gear meshing usually exhibits a generalized stationary random process with a relatively uniform energy distribution. Based on this physical difference, by analyzing the instantaneous energy fluctuation rate and pulse sparsity of the spatial vector model sequence within the microscopic time window, we can decouple it from strong background noise and identify potential damage impact features.

[0036] Specifically, the spatial vector modulus sequence within the current time window is obtained by calculating the spatial vector modulus at each moment based on the three-channel vibration signal at each moment within the current time window. , , , , The first in the current time window The vertical, horizontal, and axial vibration signals at each moment are obtained; the spatial vector mode sequence within the current time window is acquired.

[0037] The current time window is divided into multiple equal-length micro-segments, and the sum of the spatial vector magnitudes at all times within each micro-segment is taken as the local energy of each micro-segment.

[0038] It should be added that the number of micro-segments is used to define the resolution of the micro-timescale. If the value is too large, each micro-segment contains too few data points, and the calculated energy value is easily affected by random noise, resulting in unstable statistical results. If the value is too small, the micro-segment length is too long, and the energy of the brief fault impact will be averaged by other noise in the segment, resulting in diluted features. Therefore, the value is even and the range is usually 8 to 32. In this embodiment, 16 is used. Each micro-segment contains about 5ms of data, which can completely cover a bearing high-frequency resonance impact. The implementer can adjust it according to the decay time constant of the bearing resonance frequency.

[0039] Traverse the spatial vector modulus sequence within the current time window, calculate the mean of the spatial vector modulus sequence, and count the number of times each spatial vector modulus in the spatial vector modulus sequence is greater than the mean.

[0040] The impact activity of the current time window is calculated based on the number of times the spatial vector magnitude sequence exceeds the mean and the standard deviation of the local energy of all micro-segments. The specific calculation formula is as follows:

[0041]

[0042] In the formula, The impact activity level within the current time window; , The standard deviation and mean of the local energy of all micro-segments within the current time window; The ratio of the number of times the spatial vector magnitude of the spatial vector magnitude sequence is greater than the mean of the spatial vector magnitude sequence within the current time window to the length of the time window; The sensitivity coefficient is denoted as .

[0043] in, It reflects the relative dispersion of local energy distribution in the micro-segment within the current time window. The larger the value, the more uneven the energy distribution on the time axis within the current time window, that is, there are intermittent high energy bursts, which means that the vibration signal within the current time window is more likely to contain fault characteristics. This reflects a noise suppression factor based on pulse density. It utilizes the characteristic that mechanical impact signals typically have higher sparsity than Gaussian white noise. Under normal conditions, the energy distribution of Gaussian white noise is relatively continuous, but during early faults, the energy is highly concentrated on a very small number of transient impact points, exhibiting high sparsity. Therefore... The smaller the value, the more concentrated the vibrational energy is on a few peak pulses, indicating a greater likelihood that the signal contains non-stationary impact components. The value is effectively amplified; considering the possibility of equipment not operating or sensors going offline throughout the entire lifecycle, i.e. When it is 0, let Take 0.

[0044] It should be added that the sensitivity coefficient is used to adjust the degree to which sparsity suppresses activity. When the sensitivity coefficient is large, the suppression effect on high-frequency dense random noise is enhanced, and non-fault high-frequency interference can be filtered out, but it may mistakenly identify overly dense fault features. When the sensitivity coefficient is small, the sensitivity to impulses is increased, but the noise immunity is reduced. Therefore, the sensitivity coefficient range is 0.5. 5. In this embodiment, we take 2. In order to retain the low-frequency repetitive impact characteristics unique to bearing failure while effectively suppressing the interference of broadband white noise, the implementers can make adjustments according to the frequency distribution characteristics of the background noise on site.

[0045] At this point, the impact activity level for the current time window has been obtained.

[0046] S3. Calculate the autocorrelation coefficient and skewness coefficient of the spatial vector mode sequence, and calculate the background noise coupling strength based on the autocorrelation coefficient and skewness coefficient.

[0047] It should be noted that background noise such as bearing gear meshing and shaft imbalance is usually similar to normal conditions and has a strong periodic autocorrelation. Moreover, its vibration waveform still maintains good regularity after spatial synthesis. However, the impact of early bearing failure is sudden and decays rapidly, and will cause a serious skewness in the distribution of synthesized amplitude, exhibiting obvious non-Gaussian characteristics. Therefore, by analyzing these differences, the proportion of background noise can be calculated, thereby providing a basis for subsequent fault monitoring.

[0048] Specifically, the autocorrelation coefficient of the spatial vector magnitude sequence within the current time window is calculated. The calculation method includes obtaining the number of sampling points for one meshing cycle of the bearing. , ,in To preset the sampling frequency, For the target speed, Number of teeth; To round down; extract the first few bits of the spatial vector modulus sequence within the current time window. A baseline subsequence is constructed using sampling points; the spatial vector modulus sequence within the current time window is then extracted. The sampling point, i.e. the th sampling point To the Sampling points are used to construct a lagged subsequence; the Pearson correlation coefficient between the baseline subsequence and the lagged subsequence is calculated to obtain the autocorrelation coefficient of the spatial vector magnitude sequence within the current time window; This represents the number of sampling points within the current time window.

[0049] The skewness coefficients of the spatial vector mode sequence within the current time window are calculated using the third-order central moments. The method for calculating the skewness coefficients using the third-order central moments is a well-known technique and will not be described in detail here.

[0050] The background noise coupling strength is calculated based on the autocorrelation coefficient and skewness coefficient of the spatial vector mode sequence within the current time window; the background noise coupling strength satisfies the expression:

[0051]

[0052] In the formula, The background noise coupling strength for the current time window; The autocorrelation coefficient of the spatial vector magnitude sequence within the current time window; The skewness coefficients of the spatial vector modulus sequence within the current time window; These are the weighting coefficients; To take the absolute value; It is a natural exponential function.

[0053] in, The periodic autocorrelation intensity reflects the vibration signal within the current time window. Due to the strong periodicity of the background noise generated by gear meshing, the waveforms remain highly similar even after a lag of one meshing cycle. Approaching 1; however, the impact signal generated by early bearing failure has sudden and random slip characteristics. After a lag of one meshing cycle, the impact waveforms cannot overlap, leading to... Significant decrease; The symmetry weight reflects the distribution of vibration signals within the current time window. When the vibration signal follows a Gaussian distribution, the skewness coefficient is close to 0, and this term approaches 1, preserving the noise intensity. When the signal is severely skewed, this term approaches 0, reducing the noise judgment value. In summary, if the periodic autocorrelation intensity of the vibration signal within the current time window is stronger and the symmetry weight is higher, it indicates that the current signal is dominated by strong background noise coupling, and it needs to be suppressed in subsequent fault assessment to reduce false alarms.

[0054] It should be added that the weighting coefficient is used to adjust the response speed of the skewness coefficient to the coupling strength of background noise. When the weighting coefficient is large, even a slight asymmetry in the signal will cause the exponential term to decrease rapidly, thereby reducing the background noise coupling strength and making it easier to detect minor faults. When the weighting coefficient is small, the judgment process is more conservative, assuming that only extreme skewness is not noise. Therefore, the value range of the weighting coefficient is 0.5. 2. In this embodiment, we take 1.5 to balance the tolerance to non-Gaussian background noise and the sensitivity to fault impact. The implementer can adjust it according to the degree of non-Gaussian vibration during normal equipment operation.

[0055] At this point, the background noise coupling strength of the current time window is obtained.

[0056] S4. Perform Hilbert transform on the spatial vector mode sequence to obtain the envelope spectrum and calculate the envelope spectrum entropy; use the background noise coupling strength and impact activity to weight the envelope spectrum entropy and obtain the fault characteristic index.

[0057] It should be noted that while time-domain indicators alone can capture the impact, they are easily misled by non-faulty random interference, leading to false alarms. Therefore, introducing Hilbert transform to obtain the envelope spectrum of the signal can effectively demodulate the low-frequency fault characteristic frequencies carried by the high-frequency carrier. However, since fault impacts and background noise often overlap in the frequency domain, and strong background noise forms significant interference peaks in the envelope spectrum, simple frequency domain feature extraction is difficult to accurately distinguish the fault source. Based on this, the impact activity and background noise coupling strength obtained in the previous steps are used as mutually exclusive weighting factors to dynamically correct the envelope spectrum entropy, which reflects the frequency domain energy concentration of the signal. This automatically suppresses the contribution of entropy values ​​when strong noise dominates, while significantly enhancing the sensitivity of entropy values ​​when fault impacts dominate. This decouples a pure fault characteristic index that can truly characterize the degradation state of the bearing throughout its entire life cycle from the complex mixed signal.

[0058] Specifically, the spatial vector modulus sequence within the current time window is demodulated using a Hilbert transform to obtain the envelope signal. A fast Fourier transform is then performed on the envelope signal to obtain the envelope spectrum. The probability of the amplitude at each frequency point in the envelope spectrum is obtained, where the probability is the ratio of the amplitude at each frequency point to the sum of the amplitudes at all frequency points. The envelope spectrum entropy is calculated using the principle of information entropy. The envelope spectrum entropy satisfies the expression:

[0059]

[0060] In the formula, The envelope spectral entropy of the current time window; The first in the envelope spectrum within the current time window The probability of the amplitude occurring at a given frequency point; It is a logarithmic function with base 2; , This represents the index value and total number of frequency points in the envelope spectrum within the current time window.

[0061] in, This value reflects the uncertainty of the frequency domain distribution of the envelope spectrum energy within the current time window. A larger value indicates that the energy is evenly distributed across all frequency points, meaning the bearing is in good condition or the signal is dominated by broadband random noise. Since faults can produce strong impacts at specific frequencies, a smaller value indicates a high concentration of energy in the envelope spectrum, suggesting periodic fault impact characteristics and increasing the orderliness of the envelope spectrum energy within the current time window. Specifically, when... season .

[0062] Furthermore, the envelope spectral entropy is weighted using the background noise coupling strength and impact activity within the current time window to obtain a fault characteristic index; the fault characteristic index satisfies the expression:

[0063]

[0064] In the formula, The fault characteristic index for the current time window; The impact activity level within the current time window; The background noise coupling strength for the current time window; The envelope spectral entropy of the current time window; This is the penalty coefficient.

[0065] in, This reflects the adaptive gating weights based on time-domain feature fusion. If the vibration signal within the current time window exhibits strong impact and weak correlation, i.e. Increase and When decreasing Approaching 1, the gating is opened to preserve frequency domain characteristics; if the vibration signal in the current time window shows strong correlation and weak impact, Approaching 0, the gating is turned off to suppress interference; utilizing the monotonically decreasing property of the negative exponential function, the trend of the envelope spectral entropy is reversed. When the bearing is in a healthy state or when background noise dominates: the spectral energy distribution is discrete and chaotic. Approaching 0, combined with the suppression effect of gating weights, the fault characteristic index is kept at an extremely low level, avoiding false alarms; when a bearing fails: the fault impact causes a high concentration of spectral energy, and the entropy value decreases significantly. It increases sharply and nonlinearly, and combined with the conduction effect of the weights, the fault characteristic index is significantly amplified; when there is A value of 0 indicates that the system is currently in a shutdown state. Assign zero.

[0066] It's important to note that the penalty coefficient is used to adjust the strength of the penalty against background noise. A larger penalty coefficient results in very strong suppression of background noise, increasing the weight denominator and decreasing the overall performance index. However, this might lead to over-suppression and missed detections in the very early stages of a fault. Conversely, a smaller penalty coefficient preserves more detail but reduces noise immunity. Therefore, the penalty coefficient is typically set between 1.5 and 1. 5. In this embodiment, we take 3. In order to suppress background noise to the greatest extent and highlight the weak bearing faults under strong gear meshing interference, the implementer can adjust it according to the signal-to-noise ratio on site.

[0067] At this point, the fault characteristic index for the current time window has been obtained.

[0068] S5. Based on the comparison results between the fault characteristic index and the obtained early warning threshold, and the fitting slope values ​​of the fault characteristic index and the fault characteristic index of multiple consecutive time windows, graded early warning is carried out.

[0069] It should be noted that the vibration level of a precision bearing in its health state is not a fixed value throughout its entire life cycle, but fluctuates with different manufacturing processes and load conditions. Therefore, a benchmark correction mechanism is introduced to extract the fault characteristic index at the beginning of stable bearing operation as the health baseline. The alarm boundary is dynamically set using statistical principles. At the same time, considering the nonlinearity of bearing fault development, a double confirmation is performed in combination with the deterioration trend to avoid false alarms.

[0070] Specifically, a baseline sequence is constructed by extracting fault characteristic indices within the initial time window after the precision bearing is put into operation, and an early warning threshold is calculated. , ,in , The mean and standard deviation of the baseline sequence are given. This is a preset safety factor.

[0071] It should be added that the safety factor is used to define the tolerance range for fault deviations. If the value is large, it can avoid false alarms but will lead to a lag in early warning; if the value is small, the method will be too sensitive and will lead to frequent false alarms. According to the statistical law of normal distribution, The value range is usually 3. 5. In this embodiment, A value of 4 is used to ensure that even at extremely high confidence levels, subtle early signs of degradation exceeding the normal fluctuation range can be detected, while also providing sufficient buffer margin for emergency shutdown. Implementation personnel can adjust this value according to the importance level of the equipment.

[0072] Furthermore, least squares linear fitting is performed on the fault characteristic indices of the current time window and several consecutive time windows prior to it to obtain the slope values; then, the hierarchical early warning logic is executed.

[0073] If the fault characteristic index of the current time window is greater than the warning threshold and the slope value is greater than 0, the bearing is determined to have entered the fault period, a red warning signal is output, and it is recommended to stop the machine immediately for inspection; if the fault characteristic index of the current time window is less than or equal to the warning threshold and the slope value is less than or equal to 0, the bearing is determined to be working normally and no warning signal is triggered; if the fault characteristic index of the current time window is less than or equal to the warning threshold, but the slope value is greater than 0, or the fault characteristic index is greater than the warning threshold, but the slope value is less than or equal to 0, the bearing may be in the early deterioration incubation period, a yellow warning signal is output, and it is recommended to closely monitor subsequent changes.

[0074] Among them, multiple time windows are taken as 5 to 10 consecutive time windows, and in the embodiment of the present invention, 6 are preferred to ensure that the slope value obtained by fitting is both statistically significant and has sufficient real-time response speed; the implementer can adjust it according to the sampling frequency under actual working conditions and the specific requirements for early warning sensitivity.

[0075] For example, Figure 2 The waveform is a spatial vector waveform, with time on the horizontal axis and the module length on the vertical axis. During the normal operation phase from 0s to 1s, the waveform is dominated by strong gear meshing and exhibits a relatively stable, high-amplitude broadband fluctuation. During the early weak fault phase from 1s to 2s, the weak fault impact is completely masked by strong background noise, and the waveform shows no obvious visual changes. During the strong fault phase from 2s to 3s, the waveform exhibits obvious high-amplitude impact peaks.

[0076] Figure 3 The chart shows a comparison of early warning indicators. The horizontal axis represents time, the left vertical axis represents the envelope spectral entropy, and the right vertical axis represents the fault characteristic index. In the early, weak fault stage (1s to 2s), the envelope spectral entropy only shows a slight decrease that is easily overwhelmed by fluctuations in operating conditions. However, this invention captures the disruption of autocorrelation caused by weak impacts, leading to an instantaneous increase in the activity of non-stationary impacts and activation of adaptive gating. This causes the fault characteristic index to rapidly jump to 0.5, achieving high-sensitivity identification and early warning of early weak faults. In the strong fault stage (2s to 3s), as periodic fault impacts lead to a high concentration of frequency domain energy, the substantial decrease in envelope spectral entropy triggers a strong nonlinear exponential amplification effect, causing the fault characteristic index to rapidly rise and approach 1, significantly improving the accuracy of fault early warning.

Claims

1. A precision bearing life-cycle fault early warning method based on multi-source vibration signal decoupling, characterized in that, include: The vibration signals of the bearing in three channels—vertical, horizontal, and axial—are acquired and divided into multiple time windows. For the current time window, the spatial vector magnitude of the vibration signal at each moment is calculated to obtain a spatial vector magnitude sequence; the sequence is divided into multiple micro-segments, and the sum of the spatial vector magnitudes within each micro-segment constitutes the local energy of each micro-segment; based on the standard deviation of the local energies of the multiple micro-segments and the number of times the spatial vector magnitude in the sequence is greater than the mean of the sequence, the impact activity is calculated. ,satisfy: ； , The standard deviation and mean of the local energy of all micro-segments within the current time window; The ratio of the number of times the spatial vector magnitude of the spatial vector magnitude sequence is greater than the mean of the spatial vector magnitude sequence within the current time window to the length of the time window; Sensitivity coefficient; Calculate the autocorrelation coefficient and skewness coefficient of the sequence, and calculate the background noise coupling strength based on the autocorrelation coefficient and the skewness coefficient. ,satisfy: ； The autocorrelation coefficient of the spatial vector magnitude sequence within the current time window; The skewness coefficients of the spatial vector modulus sequence within the current time window; These are the weighting coefficients; To take the absolute value; It is a natural exponential function; Perform a Hilbert transform on the sequence to obtain the envelope spectrum and calculate the envelope spectrum entropy; The envelope spectrum entropy is weighted by using the background noise coupling strength and the impact activity to obtain a fault feature index satisfies: ； impact activity for the current time window; background noise coupling strength for the current time window; envelope spectrum entropy for the current time window; penalty coefficient; Based on the comparison results between the fault characteristic index and the obtained early warning threshold, and the fitting slope values ​​of the fault characteristic index and the fault characteristic index of multiple consecutive time windows, a graded early warning is performed.

2. The precision bearing life cycle fault early warning method based on multi-source vibration signal decoupling according to claim 1, characterized in that, The autocorrelation coefficient is obtained in the following ways: Obtain the number of sampling points for one meshing cycle of the bearing and record it as . , the first part of the sequence The sampling point is used as the reference subsequence, and the first sampling point is used as the reference subsequence. One to the first Each sampling point is used as a hysteresis subsequence. The number of sampling points in the current time window is given; the Pearson correlation coefficient between the baseline subsequence and the lagged subsequence is calculated to obtain the autocorrelation coefficient.

3. The method for early warning of precision bearing full life cycle faults based on multi-source vibration signal decoupling according to claim 1, characterized in that, The skewness coefficients are obtained by calculating the third-order central moments.

4. The precision bearing life cycle fault early warning method based on multi-source vibration signal decoupling according to claim 1, characterized in that, The calculation of the envelope spectrum entropy includes: The sequence is demodulated using Hilbert transform to obtain the envelope signal, and the envelope signal is then subjected to Fast Fourier Transform to obtain the envelope spectrum. The probability of the amplitude at each frequency point on the envelope spectrum is obtained, and the envelope spectrum entropy is calculated using the information entropy formula.

5. The precision bearing life cycle fault early warning method based on multi-source vibration signal decoupling according to claim 1, characterized in that, The methods for obtaining the early warning threshold include: The fault feature index in an initial time window after the precision bearing is put into operation is intercepted to construct a benchmark sequence, and a sum of a product of a preset safety coefficient and a standard deviation of the benchmark sequence is taken as a pre-warning threshold ; wherein , is the mean and standard deviation of the reference sequence, is a preset safety factor.

6. The precision bearing life cycle fault early warning method based on multi-source vibration signal decoupling according to claim 1, characterized in that, The aforementioned tiered early warning system includes: Let the fault characteristic index of the current time window be . The warning threshold is The fitting slope of the fault characteristic index for the current time window and its preceding consecutive time windows is... ;like and Output a red warning signal; if but or but Output a yellow warning signal; if and No warning signal is triggered.

7. The precision bearing life cycle fault early warning method based on multi-source vibration signal decoupling according to claim 6, characterized in that, The fitting slope is obtained by performing a least-squares linear fit on the fault characteristic indices of the current time window and several consecutive time windows preceding it.

Images