Cam driven bearing quality detection method based on data analysis
By acquiring the detail coefficient pulse singularity of the cam follower bearing and using Visu Shrink threshold denoising, the problems of impure feature extraction and damaged signal energy in early damage detection of cam follower bearings are solved, achieving accurate identification of weak damage and high-fidelity restoration of signal energy.
Patent Information
- Application Number
- CN202610105888.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-27
- Publication Date
- 2026-03-03
AI Technical Summary
Existing wavelet transform-based methods for early damage detection of cam follower bearings suffer from impure feature extraction and signal energy loss after denoising, leading to inaccurate damage identification.
By acquiring multi-layer detail coefficient sequences and approximation coefficient sequences, the enhancement coefficients are obtained using the impulse singularity of the detail coefficients. Denoising is then performed using the Visu Shrink threshold criterion, and the signal is reconstructed to improve detection accuracy.
It achieves directional enhancement and background noise suppression of minor damage to cam follower bearings, improving detection accuracy and signal-to-noise ratio, and ensuring high-fidelity restoration of damage energy.
Smart Images
Figure CN121595207A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of nondestructive testing and signal processing technology. More specifically, this invention relates to a data analysis-based method for quality inspection of cam follower bearings. Background Technology
[0002] Cam follower bearings, as key components in mechanical transmission systems, endure complex intermittent impact loads during operation. When early damage such as micro-scratching or indentation occurs on the bearing surface, it excites vibration response signals with non-stationary and transient characteristics. To ensure the safe operation of equipment, timely and accurate identification of these subtle damage features is crucial. Currently, using piezoelectric accelerometers to collect and analyze vibration signals is the mainstream detection method. Among these methods, Discrete Wavelet Transform (DWT), due to its excellent multi-resolution characteristics and time-frequency localization capabilities, is often used to decompose such non-stationary signals, aiming to separate the high-frequency damage impact components submerged in broadband background noise through multiple layers of detail coefficients.
[0003] However, existing wavelet transform-based detection methods still have limitations in practical applications of cam follower bearings. Traditional methods often lack detailed analysis of waveform geometry when processing detail coefficients, making it difficult to distinguish between the unique bilateral steep features of true damage pulses and the unilateral abrupt changes or random noise caused by load mutations, resulting in impure feature extraction. In the signal denoising stage, commonly used general threshold functions have inherent defects: the discontinuity of hard threshold functions easily introduces pseudo-Gibbs oscillations, destroying waveform smoothness; while soft threshold functions, although ensuring continuity, introduce a constant amplitude deviation. For early damage signals with already weak amplitudes, this constant deviation will severely damage the reconstructed signal energy, failing to accurately reflect the true degree of damage on the bearing surface. Summary of the Invention
[0004] To address the technical problems of inaccurate damage identification caused by impure feature extraction and signal energy loss after denoising in early damage detection of cam follower bearings based on wavelet transform, this invention provides a data analysis-based method for quality inspection of cam follower bearings. The method includes: acquiring a multi-layer detail coefficient sequence and an approximation coefficient sequence based on the vibration acceleration signal of the cam follower bearing; for each detail coefficient in each layer of the detail coefficient sequence, obtaining the impulse singularity of the detail coefficient based on its absolute amplitude and its difference amplitude with adjacent detail coefficients; obtaining the enhancement coefficient of the detail coefficient based on the impulse singularity; calculating the threshold for each layer using the Visu Shrink threshold criterion; obtaining the denoising coefficient of the detail coefficient based on the enhancement coefficient when it exceeds the threshold; reconstructing the signal based on the denoising coefficient and approximation coefficient sequence to obtain a denoised time-domain signal; and completing the quality inspection of the cam follower bearing based on the denoised time-domain signal.
[0005] This invention utilizes an enhancement coefficient to directionally enhance the weak damage characteristics in the vibration signal of a cam follower bearing. Subsequently, a threshold is used to obtain a denoising coefficient and reconstruct the signal. This can suppress background noise while preserving the damage impact characteristics, thereby improving the accuracy of cam follower bearing quality detection.
[0006] Preferably, obtaining the multi-level detail coefficient sequence and the approximation coefficient sequence includes:
[0007] A piezoelectric accelerometer is rigidly mounted on the test station of the cam follower bearing. Vibration acceleration signals are collected during the bearing's rotation. A Symlet wavelet basis is selected, the number of decomposition levels is set, and the vibration acceleration signal is subjected to discrete wavelet decomposition. The output consists of multi-level detail coefficient sequences and the last-level approximation coefficient sequence.
[0008] Preferably, the impulse singularity satisfies the expression: In the formula, Indicates the first Layer Impulse singularity of detail coefficients; Indicates the first Layer Detail coefficients; Indicates the first Layer Detail coefficients; Indicates the first Layer Detail coefficients; This represents absolute value operations.
[0009] This invention uses the product of the detail coefficient and the difference between its preceding and following detail coefficients to measure the steepness of the waveform on both sides. It takes advantage of the fact that the real damage pulse has the characteristic of bilateral abrupt change, while the interference signal usually only has unilateral abrupt change or disordered differences, to effectively distinguish the surface damage impact of the cam follower bearing from load abrupt change or random noise, and provides accurate data support for subsequent feature enhancement.
[0010] Preferably, the enhancement coefficient satisfies the expression In the formula, Indicates the first Layer Enhancement factor for each detail coefficient; Indicates the first Layer Detail coefficients; Indicates the strength enhancement coefficient; Indicates the first Layer Impulse singularity of detail coefficients; Indicates the first The maximum value of all impulse singularities in the layer.
[0011] This invention performs adaptive weighting processing on the detail coefficient sequence, which amplifies the amplitude of the actual damage pulse of the cam follower bearing due to its large pulse singularity, while keeping the background noise amplitude basically unchanged, thereby improving the signal-to-noise ratio and making weak damage impacts easier to identify.
[0012] Preferably, the denoising coefficients satisfy the expression: In the formula, Indicates the first Layer The noise reduction coefficients for each detail coefficient; Indicates the first Layer Enhancement factor for each detail coefficient; Represents a symbolic function; Indicates the first The threshold of the layer; Indicates the gradual adjustment factor; Represents the natural constant; It represents the absolute value.
[0013] This invention ensures the mathematical continuity of the function when the enhancement coefficient is close to the threshold to avoid pseudo-Gibbs oscillations, and utilizes the decay characteristics of the exponential term to eliminate the constant deviation caused by the traditional soft threshold function when the enhancement coefficient is much greater than the threshold. Thus, while removing noise, it restores the original vibration energy of the weak surface damage of the cam follower bearing to the greatest extent.
[0014] Preferably, the signal reconstruction based on the denoising coefficients and the approximation coefficient sequence includes: using discrete wavelet inverse transform to reconstruct the processed denoising coefficients of each layer with the original approximation coefficient sequence to generate a denoised time-domain signal.
[0015] Preferably, the quality inspection of the cam follower bearing based on the denoised time-domain signal includes: performing a Hilbert transform on the denoised time-domain signal to obtain an envelope signal; performing spectral analysis on the envelope signal to obtain an envelope spectrum; and performing quality inspection of the cam follower bearing based on the envelope spectrum.
[0016] Preferably, the quality inspection of the cam follower bearing based on the envelope spectrum includes: searching for the presence of damage characteristic frequencies in the envelope spectrum, where the damage characteristic frequencies include the outer ring damage frequency or the inner ring damage frequency; and determining that the cam follower bearing has surface damage in response to the observation of discrete spectral lines with amplitudes exceeding a preset safety threshold at the damage characteristic frequencies and their harmonics, otherwise determining that it is qualified.
[0017] This invention searches for the outer ring damage frequency or inner ring damage frequency and its harmonics in the envelope spectrum. By observing whether there are discrete spectral lines with amplitudes exceeding a preset safety threshold, the bearing condition is determined. This invention can transform damage information modulated on the high-frequency resonance band into intuitive low-frequency features, thereby achieving accurate determination of whether there is surface damage and the location of damage in the cam follower bearing.
[0018] Preferably, the Symlet wavelet basis is the sym5 wavelet basis.
[0019] Preferably, the number of decomposition layers is set to 4.
[0020] The beneficial effects of this invention are as follows: This invention constructs pulse singularities to obtain pulse singularities of detail coefficients. It utilizes the physical difference that real fault pulses contain bilateral steep features while load abrupt changes and noise lack this feature to perform targeted waveform structure enhancement on the detail coefficient sequence, thereby improving the signal-to-noise ratio of weak damage signals of cam follower bearings.
[0021] This invention utilizes an exponentially adjusted nonlinear function to process the enhancement coefficient. When the amplitude is close to the threshold, the function continuity is maintained to suppress the pseudo-Gibbs phenomenon. When the amplitude is large, the exponential decay characteristic is used to eliminate the constant deviation introduced by the soft threshold method, thereby faithfully restoring the damage impact energy of the cam follower bearing during the noise reduction process.
[0022] This invention combines the multi-resolution characteristics of discrete wavelet transform with envelope analysis technology to separate and demodulate the high-frequency impact components hidden in broadband background noise into low-frequency envelope spectrum features. By searching for the discrete spectral lines at the damage feature frequency and its harmonics, reliable detection of the production quality of cam follower bearings can be achieved. Attached Figure Description
[0023] Figure 1 This is a flowchart illustrating a data analysis-based method for detecting the quality of a cam follower bearing according to the present invention. Figure 2 This is a schematic diagram illustrating the detail coefficient enhancement effect based on impulse singularity; Figure 3 This is a schematic diagram illustrating the envelope spectrum of the time-domain signal after noise reduction. Detailed Implementation
[0024] 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.
[0025] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
[0026] This invention discloses a data analysis-based method for detecting the quality of cam follower bearings, referring to... Figure 1 This includes steps S1 to S4: S1. Obtain multi-layer detail coefficient sequence and approximate coefficient sequence based on the vibration acceleration signal of the cam follower bearing.
[0027] It should be noted that in the production scenarios of electromechanical industries, cam follower bearings typically serve as key transmission components subjected to intermittent impact loads. Early surface damage, such as micro-scratching or indentation, causes vibration responses with significant non-stationary and transient characteristics. This invention utilizes the multi-resolution properties of wavelet transform to map the time-domain vibration acceleration signal onto different frequency bands, thereby separating the weak high-frequency damage impact components hidden in broadband background noise, providing a clean data foundation for subsequent feature enhancement.
[0028] Specifically, a high-frequency response piezoelectric accelerometer is rigidly mounted on the test station of the cam follower bearing, and the sampling frequency is set. The frequency was 25.6 kHz; the acquisition time during the rotation of the cam follower bearing was... Vibration acceleration signal The Symlet wavelet basis sym5, which has high similarity to the mechanical impact waveform, was selected, and the number of decomposition levels was set. The value is 4; for vibration acceleration signals Perform discrete wavelet decomposition to obtain the th Layer detail coefficient sequence and the Approximate coefficient sequence of the layer Among them, the detail coefficient sequence It contains high-frequency abrupt change components of the signal, which directly correspond to the high-frequency resonance waveform caused when the bearing needle impacts the surface damage point, and is a key carrier for detecting bearing surface damage.
[0029] S2. For each detail coefficient in each layer of detail coefficient sequence, obtain the impulse singularity of the detail coefficient based on its absolute amplitude and its difference amplitude with adjacent detail coefficients; obtain the enhancement coefficient of the detail coefficient based on the impulse singularity of the detail coefficient.
[0030] It should be noted that surface damage in the cam follower bearing generates an extremely short pulse when the needle rollers pass over it. From a waveform geometry perspective, a true fault pulse not only has a certain amplitude, but more importantly, it must simultaneously contain a sharp rising edge and a sharp falling edge, forming a sharp peak structure. In contrast, load mutations or sensor zero drift typically only contain unilateral mutations, while the waveform of random noise is chaotic and lacks this compact, bilaterally steep characteristic. To accurately capture this physical characteristic, this invention constructs a pulse singularity that can simultaneously measure amplitude strength and bilateral geometric steepness. This operator only produces a high response to true fault pulses that simultaneously satisfy a large amplitude, forward high steepness, and backward high steepness, thereby achieving targeted geometric enhancement of weak fault characteristics.
[0031] Specifically, for the first Layer detail coefficient sequence The first in Detail coefficients To obtain its pulse singularity The impulse singularity satisfies the expression:
[0032] In the formula, Indicates the first Layer Impulse singularity of detail coefficients; Indicates the first Layer Detail coefficients; Indicates the first Layer Detail coefficients; Indicates the first Layer Detail coefficients; This represents absolute value operations.
[0033] For real damage signals, it in The moment is a peak, on the left and the right side The values are all significantly lower than Therefore, both the backward and forward differences are very large, and their product produces a very large amplification factor, making... Significantly increased; for step signals caused by load changes, the waveform changes abruptly only on one side, and one term in the bilateral difference must be close to 0, so the entire product term will approach 0, thus effectively suppressing such non-damaging interference; for random noise, it lacks a continuous high-intensity bilateral geometric structure, and the statistical mean of the product term is much lower than that of the damaged pulse.
[0034] Furthermore, based on the pulse singularity, the original detail coefficient sequence is adaptively weighted and enhanced to obtain the enhancement coefficients. The enhancement coefficient satisfies the expression:
[0035] In the formula, Indicates the first Layer Enhancement factor for each detail coefficient; Indicates the first Layer Detail coefficients; Indicates the strength enhancement coefficient; Indicates the first Layer Impulse singularity of detail coefficients; Indicates the first The maximum value of all impulse singularities in the layer.
[0036] For a real damage pulse, its pulse singularity The larger the value, the greater the gain term. When the gain term approaches 1, the enhancement factor is significantly amplified, thus highlighting the damage characteristics. For background noise or interference, the impulse singularity is small, the gain term approaches 0, and the amplification factor is about 1, which means that the original amplitude remains basically unchanged. Through this differentiated weighting mechanism, the signal-to-noise ratio is effectively improved, making weak damage impacts easier to identify in subsequent processing.
[0037] The strength coefficient in this invention The preferred value is 1.5; if If the value is too large, it may cause noise fluctuations from non-surface damage to be excessively amplified, introducing false features; if If the value is too small, the enhancement effect on the damage impact will be insignificant, making it difficult to extract weak signals from the background. This operation, based on the geometric realism of the waveform, specifically increases the amplitude of the damage impact, making it more difficult to be accidentally deleted in subsequent processing.
[0038] For example, Figure 2This is a schematic diagram of the detail coefficient enhancement effect based on impulse singularity. The impulse singularity curve can accurately capture the impact component with steep bilateral characteristics, so that the amplitude of the enhancement coefficient at the damage location is significantly amplified relative to the original detail coefficient. This reflects the directional enhancement effect of the present invention on weak damage features, while effectively distinguishing interference noise that lacks bilateral abrupt change features.
[0039] S3. Calculate the threshold for each layer using the Visu Shrink threshold criterion; in response to the enhancement coefficient of the detail coefficient being greater than the threshold, obtain the denoising coefficient of the detail coefficient based on the enhancement coefficient.
[0040] It should be noted that after geometric enhancement, the damage characteristics of the cam follower bearing are relatively prominent, but background noise still exists. Existing soft thresholding functions introduce a constant amplitude deviation during denoising, which is fatal for early damage signals with already weak amplitudes, potentially leading to severe damage to the reconstructed signal energy and an inability to accurately reflect the severity of surface damage. Hard thresholding functions, on the other hand, introduce spurious oscillations, i.e., pseudo-Gibbs phenomena, due to discontinuities, affecting the smoothness of the waveform. This invention constructs a novel nonlinear mapping that eliminates the constant deviation while ensuring function continuity, thereby preserving the energy of weak damage and ensuring the accuracy of subsequent quality assessment.
[0041] Specifically, the Visu Shrink threshold criterion is used to calculate the first... Layer threshold The enhancement coefficients are processed using a threshold to output denoising coefficients, which satisfy the expression:
[0042] In the formula, Indicates the first Layer The noise reduction coefficients for each detail coefficient; Indicates the first Layer Enhancement factor for each detail coefficient; Represents a symbolic function; Indicates the first The threshold of the layer; Indicates the gradual adjustment factor; Represents the natural constant; It represents the absolute value.
[0043] In the formula, when the absolute magnitude of the enhancement coefficient is... Approaching the threshold At that time, the exponent term It equals 1, at which point the contents of the parentheses become The boundary between the zero-value region (below the threshold) and the region below the threshold ensures the mathematical continuity of the function and avoids pseudo-Gibbs oscillations caused by a hard threshold; when the absolute amplitude of the enhancement coefficient... Much larger than the threshold When faced with real damage impact, the exponent in the exponential term becomes a large negative number, causing the entire exponential function value to rapidly approach 0, and the denoising coefficient degenerates to... This means that as the signal strength increases, the processed signal amplitude approaches the true value infinitely, thus restoring the original vibration energy of the weak surface damage of the cam follower bearing to the greatest extent.
[0044] The progressive adjustment factor in this invention The preferred value is 2.0; Used to control the rate of deviation elimination, if If the value is too small, the deviation elimination speed is slow, and large signals will still retain some deviation; if... If the threshold is too large, the nonlinear changes of the function near the threshold will be too drastic, which may affect the smoothness of the signal.
[0045] S4. Reconstruct the signal based on the denoising coefficient and approximation coefficient sequence to obtain the denoised time-domain signal; complete the quality inspection of the cam follower bearing based on the denoised time-domain signal.
[0046] It should be noted that the purpose of this invention is to restore the mathematical coefficients, after geometric enhancement and unbiased denoising, to a physical time series, and then use frequency domain analysis tools for final quality assessment. Surface damage in cam follower bearings is typically modulated onto the high-frequency resonant band of the system. Through envelope demodulation analysis, this modulation effect can be converted into low-frequency damage characteristic frequencies, thereby intuitively identifying whether the bearing has surface damage on the outer ring, inner ring, or needle rollers.
[0047] Specifically, using inverse discrete wavelet transform, the denoised coefficients of each layer are reconstructed with the original approximation coefficient sequence to generate a denoised time-domain signal; Hilbert transform is performed on the time-domain signal to obtain its envelope signal, and the envelope spectrum is calculated; in the envelope spectrum, it is searched for the existence of damage characteristic frequencies related to the geometry of the cam follower bearing, including the outer ring damage frequency or the inner ring damage frequency; if discrete spectral lines with amplitudes exceeding a preset safety threshold are observed at the damage characteristic frequencies and their harmonics, the cam follower bearing is determined to have surface damage; otherwise, it is determined to be qualified.
[0048] For example, Figure 3This is a schematic diagram of the envelope spectrum of the time-domain signal after noise reduction. The spectrum curve shows obvious discrete peaks in the low-frequency band. These peaks correspond to the damage characteristic frequency of the bearing, and their amplitude is significantly higher than the surrounding noise floor. This demonstrates that after using nonlinear threshold denoising and reconstruction, the present invention can accurately restore and identify the fault characteristics of bearing surface damage from a strong noise background.
Claims
1. A method for quality inspection of cam follower bearings based on data analysis, characterized in that, include: Multi-level detail coefficient sequences and approximate coefficient sequences are obtained based on the vibration acceleration signals of the cam follower bearing; For each detail coefficient in each detail coefficient sequence, the impulse singularity of the detail coefficient is obtained based on its absolute magnitude and its difference magnitude with that of its neighboring detail coefficients; the enhancement factor of the detail coefficient is obtained based on the impulse singularity of the detail coefficient. The threshold for each layer is calculated using the Visu Shrink threshold criterion; in response to the enhancement coefficient of the detail coefficient being greater than the threshold, the denoising coefficient of the detail coefficient is obtained based on the enhancement coefficient. Signal reconstruction is performed based on the denoising coefficient and approximation coefficient sequence to obtain the denoised time-domain signal; the quality inspection of the cam follower bearing is completed based on the denoised time-domain signal.
2. The method for quality inspection of cam follower bearings based on data analysis according to claim 1, characterized in that, The acquisition of the multi-level detail coefficient sequence and the approximate coefficient sequence includes: A piezoelectric accelerometer is rigidly mounted on the test station of the cam follower bearing. Vibration acceleration signals are collected during the bearing's rotation. A Symlet wavelet basis is selected, the number of decomposition levels is set, and the vibration acceleration signal is subjected to discrete wavelet decomposition. The output consists of multi-level detail coefficient sequences and the last-level approximation coefficient sequence.
3. A method for quality inspection of cam follower bearings based on data analysis according to claim 1, characterized in that, The impulse singularity satisfies the expression: ； In the formula, Indicates the first Layer Impulse singularity of detail coefficients; Indicates the first Layer Detail coefficients; Indicates the first Layer Detail coefficients; Indicates the first Layer Detail coefficients; This represents absolute value operations.
4. A method for quality inspection of cam follower bearings based on data analysis according to claim 1, characterized in that, The enhancement coefficient satisfies the expression: ； In the formula, Indicates the first Layer Enhancement factor for each detail coefficient; Indicates the first Layer Detail coefficients; Indicates the strength enhancement coefficient; Indicates the first Layer Impulse singularity of detail coefficients; Indicates the first The maximum value of all impulse singularities in the layer.
5. A method for quality inspection of cam follower bearings based on data analysis according to claim 1, characterized in that, The denoising coefficients satisfy the expression: ； In the formula, Indicates the first Layer The noise reduction coefficients for each detail coefficient; Indicates the first Layer Enhancement factor for each detail coefficient; Represents a symbolic function; Indicates the first The threshold of the layer; Indicates the gradual adjustment factor; Represents the natural constant; It represents the absolute value.
6. A method for quality inspection of cam follower bearings based on data analysis according to claim 1, characterized in that, The signal reconstruction based on the denoising coefficient and approximation coefficient sequence includes: By using the inverse discrete wavelet transform, the denoised coefficients of each layer are reconstructed with the original approximation coefficient sequence to generate the denoised time-domain signal.
7. A method for quality inspection of cam follower bearings based on data analysis according to claim 1, characterized in that, The quality inspection of the cam follower bearing based on the noise-reduced time-domain signal includes: The denoised time-domain signal is subjected to Hilbert transform to obtain the envelope signal; the envelope signal is subjected to spectral analysis to obtain the envelope spectrum; and the quality inspection of the cam follower bearing is completed based on the envelope spectrum.
8. A method for quality inspection of cam follower bearings based on data analysis according to claim 7, characterized in that, The quality inspection of the cam follower bearing based on the envelope spectrum includes: The system searches the envelope spectrum for damage characteristic frequencies, which include outer ring damage frequencies or inner ring damage frequencies. In response to the observation of discrete spectral lines with amplitudes exceeding a preset safety threshold at the damage characteristic frequencies and their harmonics, the system determines that the cam follower bearing has surface damage; otherwise, it is deemed qualified.
9. A method for quality inspection of cam follower bearings based on data analysis according to claim 2, characterized in that, The Symlet wavelet basis used is the sym5 wavelet basis.
10. A method for quality inspection of cam follower bearings based on data analysis according to claim 2, characterized in that, The number of decomposition layers is set to 4.