An integrated method for bearing condition monitoring and fault diagnosis under complex noise conditions

Abstract

This invention relates to an integrated method for bearing condition monitoring and fault diagnosis under complex noise conditions, belonging to the field of mechanical equipment condition monitoring and fault diagnosis. This method extends the classic noise-resistant correlation (NRC) method to resist interference from complex noise. The extended NRC is fused with the frequency domain energy ratio to form a composite correlation coefficient. Integrating the composite correlation coefficient through a sliding window yields a health characteristic index that monotonically changes with bearing degradation. Subsequently, an exponentially weighted moving average (EWMA) control chart is used to track this index to indicate early anomalies. Simultaneously, this method highlights fault-related periodic signals while achieving monitoring, realizing the integration of monitoring and diagnosis.

Description

Technical Field

[0001] This invention relates to mechanical equipment condition monitoring, fault diagnosis and predictive maintenance technology, specifically to an integrated method for bearing condition monitoring and fault diagnosis under complex noise conditions. Background Technology

[0002] Bearings, as core supporting components of rotating machinery, directly affect the reliability and safety of the entire equipment system. Statistics show that approximately 45% of rotating machinery failures originate from bearing defects. Especially in industrial environments such as wind turbine generators and mining machinery, bearings must withstand harsh conditions such as high pressure and high load. Traditional maintenance methods based on manual inspection and periodic overhauls are no longer sufficient to meet the needs of modern industrial equipment health management. Therefore, advancing condition monitoring and fault diagnosis technologies has significant engineering value.

[0003] Current research on bearing condition monitoring mainly focuses on extracting features from vibration signals to construct degradation indicators that reflect performance decline. However, commonly used health indicators often lack sensitivity to early failures.

[0004] To improve the timeliness of early fault detection, researchers have proposed more advanced monitoring strategies. For example, based on the multi-feature characteristics of the envelope spectrum, two health monitoring indicators—Root Mean Square Cumulative Sum (RMS-CUSUM) and Real-Time Mahalanobis Distance Cumulative Sum Growth Rate (GRRMD-CUSUM)—can accurately and efficiently monitor bearing operating status. Simultaneously, monitoring indicators based on the amplitude-direction gradient of the feature matrix can also detect early faults in real time and identify multi-stage changes in bearing performance degradation. The Attention Lempel-Ziv Complexity (ATLZ) method improves upon the traditional Lempel-Ziv complexity by using attention entropy, significantly enhancing computational efficiency while achieving monitoring functionality. As powerful tools for statistical process control, control charts such as Shewhart charts and Exponentially Weighted Moving Average (EWMA) charts can visualize operating status and improve monitoring sensitivity by establishing dynamic control limits. However, such control charts typically only provide binary results ("healthy / unhealthy") and cannot provide diagnostic analysis based on fault type.

[0005] Therefore, once an early anomaly is detected, additional diagnostic steps are required. Mainstream diagnostic methods aim to separate fault components from vibration signals. For example, adaptive variational mode decomposition (VMD) optimized by particle swarm optimization (PSO) can extract defect-related components. Simultaneously, an interpretable Transformer network with an adaptive frequency attention mechanism extends the standard Transformer model from a frequency interpretability perspective, thereby improving model interpretability while achieving accurate fault diagnosis. When operating conditions change, the shift in data features can lead to a decrease in diagnostic performance. To mitigate this domain shift problem, the Adaptive Gaussian Guided Feature Alignment Network (AGFAN) achieves rolling bearing fault diagnosis by promoting indirect alignment between source and target features. Complementing these learning-based strategies, the Noise Relevance Responsive (NRC) method provides another path, achieving fault diagnosis by segmenting the signal and superimposing enhanced Gaussian white noise to mask periodic fault components. This method exhibits excellent noise resistance performance.

[0006] Nevertheless, regardless of whether learning-based methods or signal processing techniques are used, most existing methods focus only on one aspect of monitoring or diagnosis, resulting in a significant disconnect between condition monitoring and fault diagnosis.

[0007] To address this issue, a comprehensive solution has emerged. The Refined Composite Multiscale Amplitude-Aware Permutation Entropy (RCMAAPE) method simultaneously achieves feature extraction, deterioration feature health indicator construction, and fault diagnosis. The Enhanced Noise-Resistant Correlation Method (RE-NRC) constructs a novel correlation function, from which health indicators can be established for early fault detection. Simultaneously, it outputs the current correlation function curve for further diagnosis, achieving a unification of condition monitoring and fault diagnosis. This method performs excellently in environments with strong background noise. However, most of these methods retain the white noise assumption, and their performance will significantly degrade when the background noise exhibits non-Gaussian or Gaussian-Laplace noise characteristics. Summary of the Invention

[0008] To address the aforementioned problems, this application proposes a distributed adaptive composite correlation framework. The technical method of this invention is: an integrated method for bearing condition monitoring and fault diagnosis under complex noise conditions. It includes the following steps:

[0009] Generalized derivation of the NRC method: Derivation of the analytical form of NRC under Laplace noise and Gaussian-Laplace mixed noise;

[0010] Construction of composite correlation function: The composite correlation function is constructed by weighting and fusing the generalized noise immunity correlation with the frequency domain energy ratio of the signal.

[0011] Health indicator construction: The expected value of the composite correlation function is calculated to construct health characterization indicators;

[0012] Integrated monitoring and diagnosis: The constructed health characterization indicators are input into the EWMA control chart for real-time tracking and anomaly detection; when the control chart issues an anomaly alarm, the curve of the composite correlation function corresponding to the first change point is extracted for fault diagnosis, thereby realizing the integration of bearing condition monitoring and fault diagnosis.

[0013] Furthermore, the derivation of the NRC method under Laplace noise and Gaussian-Lappland mixed noise backgrounds in the generalization derivation of the NRC method further includes:

[0014] A new signal is obtained by segmenting and superimposing the original vibration signal;

[0015] A new correlation function is constructed and analyzed based on the new signal and the original signal.

[0016] Furthermore, the construction of the composite correlation function further includes:

[0017] The frequency domain energy ratio is calculated based on the original signal.

[0018] By using weighting factors to fuse the frequency domain energy ratio and the correlation function, a composite correlation function is constructed. Its expression is:

[0019] ;

[0020] Solving for weighting factors to simplify the composite correlation function The expression.

[0021] Furthermore, the construction of the health indicators further includes:

[0022] Based on composite correlation function Calculate its expected value as a health monitoring indicator:

[0023] ;

[0024] Furthermore, the integrated monitoring and diagnosis system, which incorporates control charts for real-time monitoring, further includes:

[0025] Control chart statistics are constructed based on the calculated expected value indicators.

[0026] The control limits for the control chart are calculated based on the mean and variance of the calculated set of expected values ​​for normal data.

[0027] The parameters of the control chart are plotted for monitoring and early warning.

[0028] Furthermore, the integrated monitoring and diagnosis system, which uses a composite correlation function for fault diagnosis, further includes:

[0029] Based on the first change point in the control chart, output the current composite correlation function curve.

[0030] Fault cycle is calculated and fault matching is performed based on the peak point of the composite correlation function curve.

[0031] Furthermore, the method is applied to the integrated condition monitoring and fault diagnosis of rolling bearings under complex noise conditions.

[0032] The aforementioned integrated method for condition monitoring and fault diagnosis of rolling bearings under complex noise conditions generalizes the NRC method, enabling it to resist interference from non-Gaussian noise and even mixed noise. By weightedly fusing the generalization result with the frequency domain energy ratio to obtain a distributed adaptive composite correlation function, the periodicity extraction capability of the signal is enhanced, making it easier to separate periodic components even under mixed noise conditions. The expected value calculated based on this function can be introduced into the control chart for operational status monitoring, making the bearing's running trajectory clearer and more stable. Simultaneously, the function curve can display the periodic components in the current data for fault diagnosis, achieving integrated condition monitoring and fault diagnosis. This significantly saves operation and maintenance time costs, enabling more timely and accurate fault detection. Attached Figure Description

[0033] Figure 1 This is a bearing life data acquisition test bench in one embodiment;

[0034] Figure 2 This is a fitting graph of a hybrid noise probability model based on bearing health data in one embodiment.

[0035] Figure 3 This is a graph showing the results of monitoring bearing life data using the method of the present invention in one embodiment;

[0036] Figure 4 This is a graph showing the results of monitoring bearing life data using the weighted squared envelope dispersion entropy (WSEDisE) method in one embodiment;

[0037] Figure 5 This is a graph showing the results of monitoring bearing life data using the Attention Lempel-Ziv Complexity (ATLZ) method in one embodiment;

[0038] Figure 6 This is a graph showing the results of monitoring bearing life data using an integrated method of direct fast iterative filtering decomposition and effective weighted sparse kurtosis in one embodiment.

[0039] Figure 7This is a diagram showing the result of fault diagnosis of early fault points using the method of the present invention in one embodiment. Detailed Implementation

[0040] To make the purpose, technical solution, and advantages of this application clearer, the following description is provided in conjunction with the appendix. Figures 1-7 The present application will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the application.

[0041] The present invention provides an integrated method for bearing condition monitoring and fault diagnosis under complex noise conditions, which performs modeling and analysis on vibration signals under different noise backgrounds to achieve generalization and derivation of the NRC method.

[0042] Vibration signals against a pure Laplace noise background In cycles A new function is obtained by segmenting and superimposing. :

[0043] ;

[0044] in It is a periodic signal. For the newly constructed function, , , ( Indicates less than The largest integer), For signal length, based on Compared with the original signal A new related function is constructed as follows:

[0045] ;

[0046] For the newly constructed related functions, let In the NRC method, It is a periodic function, and it will exhibit local maxima at integer multiples of the period.

[0047] To simplify ,make:

[0048] , , ;

[0049] in for The maximum value of the period, It is a constant. It follows a Laplace distribution, and , Let be the location parameter of the Laplace distribution, then we have:

[0050] ;

[0051] According to the strong number theorem:

[0052] ;

[0053] Similarly, we can conclude that:

[0054] ;

[0055] Finally obtained The expression is:

[0056] ;

[0057] in, To determine the number of segments, consider the vibration signal against a Gaussian-Laplace mixed noise background. In cycles A new function is obtained by segmenting and superimposing.

[0058] ;

[0059] in , It follows a Gaussian distribution, and , Let be the standard deviation of the Gaussian distribution, and the two are independent of each other.

[0060] Should Functions and The correlation coefficient between them is:

[0061] ;

[0062] because According to the powerful number theorem, Finally obtained The expression is

[0063] ;

[0064] Furthermore, we fuse the time-domain NRC with the frequency-domain energy ratio to construct a distributed adaptive composite correlation function.

[0065] Frequency domain energy ratio measures the degree of energy concentration of a signal at a specific frequency and its harmonics. For periodic signals... Frequency domain energy ratio The calculation formula is:

[0066] ;

[0067] in Vibration signal Frequency domain transform, The energy of a periodic signal, For signal sampling frequency, For the number of harmonics, For bandwidth.

[0068] Through weighting factors Frequency domain energy ratio Related functions By fusing the data, a distributed adaptive composite correlation function is constructed. Its expression is

[0069] ;

[0070] Weighting factor The solution is obtained by making the coefficient of the bias term caused by noise in the fused function zero.

[0071] ;

[0072] Finally obtained The approximate expression is as follows Substitute back From

[0073] ;

[0074] Based on distribution-adaptive composite correlation function Calculate its expected value as a health monitoring indicator.

[0075] ;

[0076] The value will increase as the severity of the fault increases.

[0077] To track health indicator sequences We embed this sequence into an EWMA control chart, and the recursive statistic is:

[0078] ;

[0079] In the formula, This is the smoothing factor for the control chart.

[0080] In EWMA control charts, the center line typically represents the average value of the observed process. Control limits are used to determine if the process is showing signs of being out of control. Control limits are set as follows:

[0081] ;

[0082] ;

[0083] in To control the limit coefficient, This represents the average of health data. The standard deviation of health data.

[0084] Two actions are triggered when the signal goes out of control: 1. Status alarm: When the exponentially weighted moving average statistic exceeds the control limit, the bearing is judged to be in a deteriorated state. 2. Fault identification: Output the composite correlation function curve of the out-of-control time. It reveals fault components through significant peak values, thereby achieving integrated condition monitoring and diagnosis.

[0085] Example

[0086] This embodiment uses a bearing fatigue life test bench as the application scenario to further verify the effectiveness and practicality of this application.

[0087] In practical measurement scenarios, the acquired bearing vibration signals are easily affected by non-Gaussian noise and complex noise, which can obscure early fault characteristics. To verify the effectiveness of the proposed method, experimental data was collected on a test bench, such as... Figure 1 As shown, a deep groove ball bearing, model 6007ZZCM, was installed on the test bench, operating at a speed of 7200 rpm. Vibration signals were sampled at a rate of 25.6 kHz per minute, resulting in 427 continuous records, each containing 38,400 data points (approximately 1.5 seconds). By the end of the experiment, the bearing had experienced rolling element failure.

[0088] The Gaussian-Laplace mixture model was fitted using the signals from sample groups 1-10 (healthy baseline). Figure 2 The results confirm that the hybrid model successfully captures the characteristics of the heavy-tailed empirical distribution, thus validating the effectiveness of the hybrid noise hypothesis.

[0089] Four schemes were applied to the complete lifetime data, including this embodiment, the weighted square envelope dispersion entropy method (WSEDisE), the Lempel-Ziv complexity method, and the integrated method of direct fast iterative filtering decomposition combined with effective weighted sparse kurtosis (dFIF+EWSK). Figure 3-6 The monitoring results of each method on the control charts are shown.

[0090] exist Figure 3In the healthy phase, the EWMA statistic remained close to the baseline, then rose steadily and broke through the UCL at the 182nd sample. Although there were brief fluctuations in the data, the overall trend was upward. This curve demonstrates the sensitivity of the indicator constructed in this invention to early, minor fault changes.

[0091] Figure 4 In (WSEDisE), the EWMA curve fluctuates within the control limits and fails to fully characterize the bearing's operating state. This result indicates that the method may be limited in complex noise environments.

[0092] Figure 5 In (dFIF+EWSK), the EWMA curve intersects the lower control limit (LCL) at the 193rd sample point, indicating a delay in alarm triggering. The curve remains within the control limit for an extended period after crossing it, suggesting a deficiency in the monitoring system's ability to identify fault conditions.

[0093] exist Figure 6 In (ATLZ), the EWMA curve intersects with LCL at the 189th sampling point, also indicating an alarm trigger delay. The continued fluctuations after the intersection reveal the instability of this method under complex noise conditions.

[0094] By combining correlation functions Analyze the signal at the 182nd sample location, such as... Figure 7 As shown, the local peak interval is 2.8 milliseconds, and according to the bearing geometry and rotational speed, the theoretical rolling frequency corresponding to the rolling element defect is also 2.8 milliseconds. This confirms the existence of early rolling element failures and verifies the accuracy of the method proposed in this invention.

[0095] The integrated bearing condition monitoring and fault diagnosis method proposed in this embodiment can be used for product quality control on production lines in the field of industrial automation. It can be deployed on industrial computers or embedded platforms to realize online, real-time health management of bearings and ensure high efficiency and quality in production.

[0096] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0097] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. An integrated method for bearing condition monitoring and fault diagnosis under complex noise conditions, characterized in that, Includes the following steps: Generalized derivation of the NRC method: Derivation of the analytical form of NRC under Laplace noise and Gaussian-Laplace mixed noise; Construction of composite correlation function: The composite correlation function is constructed by weighting and fusing the generalized noise immunity correlation with the frequency domain energy ratio of the signal. Health indicator construction: The expected value of the composite correlation function is calculated to construct health characterization indicators; Integrated monitoring and diagnosis: The constructed health characterization indicators are input into the EWMA control chart for real-time tracking and anomaly detection; when the control chart issues an anomaly alarm, the curve of the composite correlation function corresponding to the first change point is extracted for fault diagnosis, thereby realizing the integration of bearing condition monitoring and fault diagnosis.

2. The method according to claim 1, characterized in that, The generalization derivation of the NRC method further includes: Vibration signals against a pure Laplace noise background Vibration signals against a background of mixed Gaussian-Laplace noise In cycles By segmenting and superimposing, new functions are obtained respectively. and ; Based on respectively Compared with the original signal , Compared with the original signal Construct new related functions and .

3. The method according to claim 1, characterized in that, The construction of the composite correlation function further includes: calculating the frequency domain energy ratio based on the original signal; Using weighting factors to correlate frequency domain energy ratio with correlation function By fusion, a composite correlation function is constructed. Its expression is: ； Where α is the weighting factor, D(T) is the frequency domain energy ratio, S(T) is the periodic correlation component in the signal, and E is a periodic function. x (T0) is the energy of the periodic signal, σ is the variance of the overall noise, m is the number of segments, Γ is the total length of the signal, and T0 is the segment length; Solve for the weighting factors to simplify the expression of the composite correlation function F(T).

4. The method according to claim 3, characterized in that, The construction of the health indicators further includes: based on a composite correlation function. Calculate its expected value as a health monitoring indicator: ； Where X represents a health monitoring indicator. for The expected value.

5. The method according to claim 1, characterized in that, The integrated monitoring and diagnostic process includes real-time monitoring of operational status, which further includes: Control chart statistics are constructed based on the calculated expected value indicators; The control limits for the control chart are calculated based on the mean and variance of the calculated set of expected values ​​for normal data. The parameters of the control chart are plotted for monitoring and early warning.

6. The method according to claim 1, characterized in that, The fault diagnosis described in the integrated monitoring and diagnosis process further includes: Based on the first change point in the control chart, output the current composite correlation function curve; Fault cycle is calculated and fault matching is performed based on the peak point of the composite correlation function curve.

7. The method according to claim 1, characterized in that, The method is applied to the integrated condition monitoring and fault diagnosis of rolling bearings under complex noise conditions.

Images