Lubricating system fault early warning method based on data analysis

Abstract

The invention provides a lubricating system fault early warning method based on data analysis, which relates to the technical field of data analysis, and is used for comprehensively analyzing vibration data and pressure data in operation data of a lubricating system, judging whether an oil way blockage phenomenon occurs in the lubricating system or not, and judging the specific blockage position and degree. Therefore, the problem that in a high-speed bearing lubricating system, due to the fact that metal abrasive dust, oil sludge and external impurities in lubricating oil accumulate and deposit in an oil way, oil way blockage is caused, poor lubrication is caused, and the bearing is damaged due to overheating can be solved.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and more specifically, to a lubrication system fault warning method based on data analysis. Background Art

[0002] As an important part of mechanical equipment, the lubrication system plays key roles such as reducing friction, decreasing wear, cooling and heat dissipation, and preventing rust during equipment operation. With the development of industrial automation and intelligence, the complexity and importance of the lubrication system are increasing day by day, and the reliability of its operating state is crucial for the stability of the entire equipment. However, during long-term operation, problems such as oil circuit blockage, deterioration of lubricating oil quality, and abnormal bearing vibration often occur in the lubrication system. If these problems are not discovered and handled in a timely manner, it may lead to a decline in equipment efficiency, and even cause serious equipment damage or shutdown accidents. Therefore, how to use modern data analysis technology to monitor the lubrication system in real time and give fault warnings has become an important research direction in the fields of mechanical engineering and intelligent maintenance.

[0003] Currently, research on lubrication system fault warning mainly focuses on aspects such as vibration signal analysis, pressure change monitoring, and lubricating oil characteristic detection. For example, existing research has shown that abnormal patterns during equipment operation can be effectively identified by analyzing the vibration signals of bearings; while the monitoring method based on pressure sensors can promptly detect pressure abnormalities caused by oil circuit blockage. However, these traditional methods usually only rely on a single data source, do not fully consider the coupling relationship between different parameters in the lubrication system, and are difficult to comprehensively reflect the system operating state. In addition, due to the presence of noise interference in vibration signals and pressure signals, existing technologies still have deficiencies in data processing accuracy and fault location accuracy, and are prone to false negatives or false positives. Especially in a complex industrial environment, the occurrence mechanism of lubrication system faults is diverse and hidden, and existing technologies are difficult to achieve early warning of faults and quantitative evaluation of blockage degree. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention is proposed. The present invention provides a lubrication system fault warning method based on data analysis, which can, to a certain extent, reduce the problem of lubrication failure caused by the accumulation and deposition of metal chips, sludge, and external impurities in the lubricating oil in the oil circuit of a high-speed bearing lubrication system, resulting in oil circuit blockage, and further causing bearing overheating and damage.

[0005] According to one aspect of the present invention, there is provided a lubrication system fault warning method based on data analysis, which includes:

[0006] Collect the operation data of the lubrication system, and perform denoising processing on the operation data based on wavelet transform;

[0007] Construct a coupling feature matrix of bearing vibration data and lubricating oil pressure data in the preprocessed operating data. When the coincidence degree of the energy mutation point of the bearing vibration data and the abnormal point of the lubricating oil pressure data in the time dimension exceeds a preset threshold, trigger in-depth analysis to obtain a fault feature vector and determine the blockage location;

[0008] Conduct a comparative analysis on the fault feature vector to evaluate and give an early warning of the blockage degree of the located oil circuit blockage position.

[0009] Furthermore, use the db4 wavelet basis to process the operating data of the lubrication system;

[0010] Among them, the vibration data in the operating data of the lubrication system is processed by an improved adaptive threshold method;

[0011] The improved adaptive threshold method includes the following steps:

[0012] Calculate the standard deviation of the wavelet decomposition coefficients of each layer, and denote this standard deviation as σ j ; then calculate the initial threshold λ according to the length N of the wavelet coefficients of this layer j ; j ;

[0013] Calculate the local variance of the wavelet coefficients of this layer;

[0014] When the local variance is greater than the preset variance threshold, it indicates that this area contains the mutation characteristics of the vibration signal, then divide the threshold λ of this area j by the adaptive factor α;

[0015] When the local variance is less than or equal to the preset variance threshold, then maintain the original threshold λ j unchanged.

[0016] Furthermore, the value of the adaptive factor α is determined based on the energy distribution characteristics and mutation degree of the vibration signal, and the value of the adaptive factor α is set according to the non-Gaussian characteristics of the vibration signal.

[0017] Furthermore, the determination of the energy mutation point is as follows:

[0018] Calculate the vibration energy value of the preprocessed bearing vibration data according to a fixed time window to obtain a vibration energy sequence, calculate the mean and standard deviation of this vibration energy sequence. When the vibration energy value at a certain moment exceeds the range of plus or minus three times the standard deviation of the mean, mark this moment as the vibration energy mutation point. If multiple consecutive data points exceed this range, take the first moment exceeding the range as the mutation point;

[0019] The identification of the pressure abnormal point is as follows:

[0020] Calculate the moving average of the lubricating oil pressure data as the pressure reference value;

[0021] When the deviation between the actual pressure value at a certain moment and the pressure reference value exceeds twice the standard deviation, mark this moment as a candidate point for abnormal pressure;

[0022] If the duration of this abnormal state exceeds the preset minimum duration threshold, confirm this moment as an abnormal pressure point.

[0023] Furthermore, the in-depth analysis extracts the fault feature vector by calculating the singular value decomposition of the coupled feature matrix; and locates the specific pipeline position of the oil pipeline blockage by comparing the projection amplitude differences of the fault feature vectors between adjacent measuring points.

[0024] Furthermore, the comparative analysis includes:

[0025] Separate the vibration component and the pressure component from the fault feature vector;

[0026] Slide-connect the fault feature vectors of adjacent sub-matrices to form a vibration feature change sequence and a pressure feature change sequence. When the feature vector in a certain time window mutates, mark this window as a feature mutation window;

[0027] Calculate the covariance matrix of the vibration component and the pressure component within each time window, and extract the main eigenvalue of the covariance matrix;

[0028] Construct a feature vector transfer matrix between adjacent time windows, analyze the migration trajectory of the feature vector in the state space. When the migration trajectory shows an abnormal deviation, determine that the system performance begins to deteriorate;

[0029] Calculate the divergence index of the vibration component and the attenuation index of the pressure component. When the sum of the two indices exceeds the preset stability threshold, confirm that the system enters the fault development stage.

[0030] Furthermore, the determination condition for the mutation of the feature vector includes three determination conditions. When all three determination conditions are satisfied, it indicates that the feature vector in a certain time window mutates.

[0031] Furthermore, judge the degree of oil pipeline blockage based on the results of the comparative analysis;

[0032] The degree of oil pipeline blockage includes mild blockage, moderate blockage, and severe blockage;

[0033] When all of the following conditions are met, it is determined to be mild blockage:

[0034] The sum of the divergence index of the vibration component and the attenuation index of the pressure component first exceeds the preset stability threshold;

[0035] The eigenvector transfer matrix shows that the trajectory begins to deviate abnormally;

[0036] The growth rate of the ratio of the dominant eigenvalues of the vibration component and the pressure component exceeds the preset threshold for the first time;

[0037] It is determined as moderate blockage when all of the following conditions are met:

[0038] Feature vector mutations are detected within 3 consecutive time windows;

[0039] The change rate of the main eigenvalue of the covariance matrix continuously exceeds the preset threshold for 5 time windows;

[0040] The growth rate of the eigenvalue ratio continuously exceeds the preset threshold for 4 time windows;

[0041] It is determined as severe blockage when any of the following conditions is met:

[0042] Feature vector mutations are detected within 10 consecutive time windows;

[0043] The slope of the fault development trend graph exceeds the safe operation threshold;

[0044] The migration trajectory of the eigenvector in the state space shows more than 5 abnormal deviations continuously.

[0045] According to another aspect of the present invention, there is provided a lubrication system fault warning system based on data analysis, which includes:

[0046] An acquisition module for acquiring the operation data of the lubrication system and performing denoising processing on the operation data based on wavelet transform;

[0047] A blockage position determination module for constructing a coupling feature matrix of the bearing vibration data and the lubricating oil pressure data in the preprocessed operation data, and triggering in-depth analysis when the coincidence degree of the energy mutation point of the bearing vibration data and the abnormal point of the lubricating oil pressure data in the time dimension exceeds the preset threshold, obtaining a fault feature vector and determining the blockage position;

[0048] A blockage degree determination module for performing comparative analysis on the fault feature vector and evaluating the blockage degree of the located oil circuit blockage position;

[0049] A warning module for giving a warning according to the result of the blockage degree determination module.

[0050] Compared with the prior art, the lubrication system fault warning method based on data analysis provided by the present invention comprehensively analyzes the vibration data and pressure data in the operation data of the lubrication system to determine whether there is an oil circuit blockage in the lubrication system, and to determine the specific blockage location and degree. In this way, it is possible to reduce the problem of poor lubrication caused by the accumulation and deposition of metal debris, sludge and external impurities in the lubricating oil in the oil circuit of the high-speed bearing lubrication system, resulting in oil circuit blockage, and further causing overheating and damage of the bearing. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings. In the drawings:

[0052] Figure 1 FIG. is a flowchart of a lubrication system fault warning method based on data analysis according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0053] Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments of the present invention. It should be understood that the present invention is not limited by the exemplary embodiments described herein.

[0054] Figure 1 FIG. is a flowchart of a lubrication system fault warning method based on data analysis according to an embodiment of the present invention. As Figure 1 shown, in the lubrication system fault warning method based on data analysis, it includes:

[0055] S1: Collect the operation data of the lubrication system, and perform denoising processing on the operation data based on wavelet transform;

[0056] Collect the system operation data through a sensor array arranged at each key monitoring point of the high-speed bearing lubrication system, specifically including: installing vibration sensors on the surface of the bearing housing along the radial and axial directions to collect the vibration data, and the sampling frequency of the vibration sensors is 1000 Hz; setting pressure sensors at the oil inlet and oil return ports of the lubricating oil supply pipeline to collect the lubricating oil pressure data, and the sampling frequency of the pressure sensors is 200 Hz.

[0057] Perform preprocessing on the collected vibration data and the lubricating oil pressure data, specifically including:

[0058] The vibration data and the lubricating oil pressure data are uniformly subjected to three-layer wavelet decomposition using the db4 wavelet basis to obtain wavelet coefficients in different frequency bands. Among them, the wavelet coefficients of the vibration data are processed using an improved adaptive threshold method to retain the mutation characteristics of the signal; the wavelet coefficients of the lubricating oil pressure data are processed using a soft threshold method to eliminate measurement noise. Then, the processed wavelet coefficients are reconstructed to obtain the denoised vibration data and the lubricating oil pressure data.

[0059] More specifically, when processing the wavelet coefficients of the vibration data using the improved adaptive threshold method, first calculate the standard deviation of the wavelet decomposition coefficients of each layer, and denote this standard deviation as σ j ; then calculate the initial threshold λ j according to the length N j of the wavelet coefficients of this layer, where λ j is equal to σ j multiplied by the square root of the natural logarithm of the length N j of the wavelet coefficients of this layer; then calculate the local variance of the wavelet coefficients of this layer, specifically calculate the variance of the wavelet coefficients within a sliding window of length L. When the local variance is greater than the preset variance threshold, it indicates that this region contains the mutation characteristics of the vibration signal, and then divide the threshold λ j of this region by the adaptive factor α, where the value range of α is 1.5 to 2.0; when the local variance is less than or equal to the preset variance threshold, then maintain the original threshold λ j unchanged; finally, perform soft threshold processing on the wavelet coefficients. When the absolute value of the wavelet coefficient is greater than the adjusted threshold, subtract the threshold from the absolute value of the coefficient and keep the sign unchanged. When the absolute value of the wavelet coefficient is less than or equal to the adjusted threshold, then set the coefficient to zero. Through this improved adaptive threshold method, while filtering out noise, the mutation characteristics in the vibration signal are effectively retained, improving the accuracy of subsequent fault feature extraction.

[0060] The value of the adaptive factor α is determined based on the energy distribution characteristics and mutation degree of the vibration signal, including:

[0061] First, calculate the kurtosis value \(K\) and skewness value \(S\) of the wavelet coefficients within the sliding window. When \(K\) is greater than the preset kurtosis threshold and the absolute value of \(S\) is greater than the preset skewness threshold, it indicates that the vibration signal in this region has significant non-Gaussian characteristics. At this time, set the adaptive factor \(\alpha\) to the first preset adaptive coefficient; when \(K\) is greater than the preset skewness threshold and the absolute value of \(S\) is less than or equal to the preset skewness threshold, it indicates that the vibration signal in this region has medium non-Gaussian characteristics. At this time, set the adaptive factor \(\alpha\) to the second preset adaptive coefficient; when \(K\) is less than or equal to the preset skewness threshold and the absolute value of \(S\) is greater than the preset skewness threshold, it indicates that the vibration signal in this region has weak non-Gaussian characteristics. At this time, set the adaptive factor \(\alpha\) to the third preset adaptive coefficient; when \(K\) is less than or equal to the preset skewness threshold and the absolute value of \(S\) is less than or equal to the preset skewness threshold, it indicates that the vibration signal in this region is close to the Gaussian distribution. At this time, set the adaptive factor \(\alpha\) to the fourth preset adaptive coefficient. Among them, the kurtosis value \(K\) reflects the sharpness of the signal distribution. The larger the \(K\) value, the more significant the mutation characteristics of the signal; the skewness value \(S\) reflects the asymmetry of the signal distribution. The larger the absolute value of \(S\), the stronger the asymmetry of the signal. By associating the value of the adaptive factor \(\alpha\) with the statistical characteristics of the vibration signal, the adaptive adjustment of the threshold processing intensity is realized, and good noise reduction effects can be obtained under different working conditions.

[0062] The improved method is particularly suitable for processing vibration signals when an oil circuit blockage occurs in a bearing system. When a local blockage occurs in the oil circuit, poor lubrication will cause instantaneous dry friction between the bearing moving pairs, which is manifested as non-stationary impact characteristics in the vibration signal. By associating the adaptive factor \(\alpha\) with the kurtosis value and skewness value of the signal, when the signal shows obvious non-Gaussian characteristics, the adaptive factor takes a larger value, reducing the threshold processing intensity, so as to better retain these key fault characteristics; when the signal is close to the Gaussian distribution of the normal working condition, the adaptive factor takes a smaller value, increasing the threshold processing intensity, and effectively suppressing background noise.

[0063] Align the denoised vibration data and the lubricating oil pressure data according to a unified time tag to form preprocessed lubricating system operation data. Among them, the lubricating oil pressure data is aligned with the vibration data in time by using the linear interpolation method to ensure the synchronization and integrity of the data. Through the above preprocessing steps, while ensuring the denoising effect, the computational complexity is significantly reduced.

[0064] It should be noted that although the above description outlines the general process of denoising vibration data based on wavelet transform, the specific implementation details may vary according to different industrial scenarios and their equipment characteristics. For example, in the monitoring of a rolling mill bearing in a large steel enterprise, due to the high environmental noise, the technical staff set the sliding window length L to 512 sampling points and adjusted the preset variance threshold to 1.5 times the mean variance of this batch of data to better identify the mutation characteristics in the bearing vibration signal. In the application of monitoring the fan bearing in a wind farm, considering the periodic change characteristics of the fan operating conditions, the technical staff adopted a sliding window of 128 sampling points and set the preset variance threshold to 2 times the mean variance of this batch of data to reduce the interference of environmental factors.

[0065] In practical applications, the selection of the adaptive factor α also needs to be optimized according to the operating characteristics of specific equipment. For example, in the bearing monitoring system of a centrifugal pump in a petrochemical enterprise, when the equipment is in the startup or shutdown process, due to the strong non-stationarity of the vibration signal, the technical staff increased the judgment threshold of the kurtosis value K to more accurately capture the vibration characteristics during these transition processes. In the monitoring of the mill bearing in a cement plant, considering that the equipment operates under stable conditions for a long time, the technical staff reduced the judgment threshold of the skewness value S to improve the detection sensitivity to minor faults.

[0066] S2: Construct a coupling feature matrix of the preprocessed bearing vibration data and the lubricating oil pressure. When the coincidence degree of the detected energy mutation point and the pressure anomaly point in the time dimension exceeds the preset threshold, trigger in-depth analysis;

[0067] Segment the preprocessed bearing vibration data according to time windows, calculate the root mean square value, kurtosis value and waveform factor within each time window to form a vibration feature vector; at the same time, extract the pressure mean value, pressure volatility and pressure drop rate from the lubricating oil pressure data within the corresponding time window to form a pressure feature vector; combine these two feature vectors in the time dimension to construct an M×N-dimensional coupling feature matrix, where M represents the feature dimension and N represents the number of time windows.

[0068] Then perform coincidence analysis: perform time series analysis on the bearing vibration data and the lubricating oil pressure data respectively, and identify the vibration energy mutation point and the pressure anomaly point by setting the standard deviation multiple method:

[0069] Calculate the vibration energy value of the preprocessed bearing vibration data according to a fixed time window to obtain a vibration energy sequence. Calculate the mean and standard deviation of this vibration energy sequence. When the vibration energy value at a certain moment exceeds the range of plus or minus three times the standard deviation of the mean, mark this moment as a vibration energy mutation point. If multiple consecutive data points exceed this range, take the first moment that exceeds the range as the mutation point.

[0070] Identification of abnormal pressure points: First, calculate the moving average of the lubricating oil pressure data as the pressure reference value. When the deviation between the actual pressure value at a certain moment and the pressure reference value exceeds twice the standard deviation, mark this moment as a candidate abnormal pressure point. If the duration of this abnormal state exceeds the preset minimum duration threshold, confirm this moment as an abnormal pressure point.

[0071] Calculate the coincidence degree of these mutation points and abnormal points on the time axis. The coincidence degree calculation formula is: Coincidence degree = Number of coincidence points / Total number of abnormal points. When this coincidence degree exceeds the preset threshold (e.g., 0.8), it indicates that there is a significant correlation between vibration abnormality and pressure abnormality, triggering in-depth analysis.

[0072] The in-depth analysis extracts the fault feature vector by calculating the singular value decomposition of the coupling feature matrix to locate the specific pipeline position of the oil circuit blockage. Specifically, through the process of calculating the singular value decomposition of the coupling feature matrix, the fault feature vector is obtained as follows:

[0073] Divide the coupling feature matrix into several sub-matrices along the main diagonal direction. Each sub-matrix represents the feature change law within a local time window; calculate the eigenvalues and eigenvectors of each sub-matrix respectively and sort them in descending order according to the eigenvalue size; according to the eigenvalue sequence of each sub-matrix, calculate the contribution rate of each eigenvalue. When the cumulative contribution rate reaches the preset 85%, combine the corresponding eigenvectors to construct the fault feature vector.

[0074] Furthermore, project the fault feature vector F into the pre-established oil circuit topology space, which is composed of information such as pipeline connection relationships, pipeline geometric parameters, and measuring point positions; then, calculate the projection amplitude of the fault feature vector F at each pipeline node. The larger the projection amplitude, the higher the possibility of oil circuit blockage at this node; then, according to the spatial distribution characteristics of the projection amplitude and combined with the pressure propagation law, determine the most likely pipeline section where the oil circuit blockage occurs; finally, accurately locate the blockage position by comparing the differences in the projection amplitudes of the fault feature vectors between adjacent measuring points: When the difference in the projection amplitudes between adjacent measuring points exceeds the preset determination threshold, it is determined that the blockage position is in the pipeline section between these two measuring points.

[0075] More specifically, according to the spatial distribution characteristics of the projection amplitude and combined with the pressure propagation law, determine the most likely pipeline section where the oil circuit blockage occurs as follows:

[0076] First, in the oil circuit topological space, number the measuring points in sequence along the oil flow direction of the oil circuit, establish a pressure transmission link diagram, and record the projection amplitude of the fault feature vector at each measuring point. Then, starting from the oil circuit inlet, calculate the pressure gradient value between adjacent measuring points along the oil flow direction. When it is detected that the pressure gradient value at a certain measuring point is significantly higher than the historical normal value of this point, mark this measuring point as an abnormal point. Next, analyze the change trend of the projection amplitude of the measuring points before and after the abnormal point. If the projection amplitude of the measuring point before the abnormal point shows a stable or slowly increasing trend, while the projection amplitude of the measuring point after the abnormal point shows a sudden drop, then determine the location where the abnormal point is located as the blockage point. Subsequently, by comparing the phase difference of the pressure fluctuations before and after the abnormal point, when the phase difference is greater than the preset determination threshold, further verify the accuracy of the blockage position. Finally, according to the ratio of the projection amplitudes of the measuring points before and after the abnormal point, calculate the blockage degree coefficient. When this coefficient exceeds the preset severity threshold, mark this pipe section as the key monitoring area.

[0077] Among them, the calculation of the blockage degree coefficient can be expressed by the following formula:

[0078]

[0079] Among them, C is the blockage degree coefficient, and its value range is [0, 1]. 0 indicates no blockage, and 1 indicates complete blockage. When C ≥ 0.75, mark this pipe section as the key monitoring area; β is the normalization coefficient, which is used to normalize the calculation result to the [0, 1] interval; P i is the pressure value at the abnormal point; P i-1 is the pressure value of the measuring point before the abnormal point; P n is the standard pressure value when the system is operating normally; V i is the projection amplitude of the fault feature vector at the abnormal point; V i-1 is the projection amplitude of the measuring point before the abnormal point; Δφ is the phase difference of the pressure fluctuations before and after the abnormal point; φ0 is the phase difference reference threshold.

[0080] The technical characteristics of the above in-depth analysis are as follows: The fault feature vector extracted by singular value decomposition contains the coupled feature information of vibration and pressure data, and can effectively reflect the characteristic mode of the oil circuit blockage fault; Projecting the fault feature vector into the oil circuit topological space for positioning analysis makes full use of the physical structure information of the system and improves the accuracy of fault location; Through the spatial distribution characteristics of the projection amplitude and the amplitude difference analysis between adjacent measuring points, the precise positioning of the oil circuit blockage position is realized, providing reliable position information support for subsequent fault early warning.

[0081] It should be noted that although the above description outlines the general process of detecting the oil pipeline blockage location, the specific implementation details may vary according to different industrial sites and their equipment configurations. For example, in the rolling mill bearing lubrication system of a large steel enterprise, the projection of the fault feature vector calculated in the oil pipeline topological space shows that the projection amplitude at measuring point No. 5 suddenly rises to 2.5 times the normal value, while the projection amplitude at measuring point No. 6 drops to 0.4 times the normal value. At the same time, the pressure gradient value exceeds 180% of the historical normal value. Based on this, the system determines that the blockage location is in the pipeline section between measuring points No. 5 and No. 6. Subsequent maintenance verifies that there is indeed a blockage formed by the accumulation of metal chips in this section of the pipeline. In addition, in the centrifugal pump bearing lubrication system of a chemical plant, by analyzing the phase difference of the pressure fluctuations between adjacent measuring points, it is found that the phase difference between measuring point No. 3 and measuring point No. 4 reaches 85 degrees, far exceeding the 15-degree phase difference under normal operating conditions. Combining with the mutation characteristics of the projection amplitude, the blockage location is accurately located. Maintenance finds that a large amount of sludge deposits adhere to the inner wall of the pipeline at this place, fully verifying the effectiveness of this method.

[0082] S3: Conduct a comparative analysis on the fault feature vector, and evaluate and give early warnings on the blockage degree of the located oil pipeline blockage location.

[0083] First, separate the vibration component and the pressure component from the fault feature vectors obtained from each sub-matrix. The vibration component reflects the change of vibration characteristics within a local time window, and the pressure component reflects the change of pressure characteristics within the corresponding time window. Then, slide-connect the fault feature vectors of adjacent sub-matrices to form a vibration characteristic change sequence and a pressure characteristic change sequence. When the feature vector in a certain time window mutates, mark this window as a feature mutation window.

[0084] Among them, the determination conditions for the mutation of the feature vector are as follows:

[0085] When the system detects that the following three conditions are simultaneously met, it is determined that the feature vector mutates:

[0086] Condition 1: Calculate the Euclidean distance D between the feature vector of the current time window and the feature vector of the previous time window. When D exceeds 2 times the average Euclidean distance in the normal operating state of the system, it is recorded as meeting Condition 1;

[0087] Condition 2: Calculate the change amount Δθ of the direction angle θ of the current feature vector in the state space relative to the previous time window. When Δθ exceeds 45 degrees and this state lasts for more than 2 sampling periods, it is recorded as meeting Condition 2;

[0088] Condition 3: Calculate the correlation coefficient ρ between the vibration component and the pressure component within the current time window. When the absolute value of ρ is less than 0.3 (indicating a significant reduction in the correlation between the two components) and this low-correlation state lasts for more than 3 sampling periods, it is recorded as meeting Condition 3.

[0089] When the system detects that the above three conditions are simultaneously satisfied within the same time window, immediately mark this time window as a characteristic mutation window and trigger the subsequent blockage degree evaluation process.

[0090] Among them, the calculation of the correlation coefficient between the vibration component and the pressure component within the current time window can be expressed by the following formula:

[0091]

[0092] Among them, ρ is the improved correlation coefficient, v i is the vibration component sequence, p i is the pressure component sequence, μ v is the local mean of the vibration component, μ ρ is the local mean of the pressure component, α ρ is the time decay coefficient, Δt i is the time interval relative to the start time of the window, β ρ is the standard deviation difference penalty factor, σ v is the standard deviation of the vibration component, σ p is the standard deviation of the pressure component, n is the number of sampling points within the time window, and i is the sampling sequence number.

[0093] Furthermore, calculate the covariance matrix of the vibration component and the pressure component within each time window, extract the main eigenvalue of the covariance matrix. When the change rate of the main eigenvalue exceeds the preset threshold, it indicates that the coupling relationship between the vibration and pressure characteristics has changed significantly; then, construct the eigenvector transfer matrix between adjacent time windows, analyze the migration trajectory of the eigenvector in the state space. When the migration trajectory shows an abnormal deviation, it is determined that the system performance begins to deteriorate; then, based on the time-series evolution law of the eigenvector, calculate the divergence index of the vibration component and the attenuation index of the pressure component. When the sum of the two indices exceeds the preset stability threshold, it is confirmed that the system enters the fault development stage.

[0094] Finally, for each characteristic mutation window, extract the ratio of the dominant eigenvalues of the vibration component and the pressure component, and establish the time series of the eigenvalue ratio; when this ratio continues to increase and the growth rate exceeds the preset threshold, the system outputs a warning signal; at the same time, by analyzing the projection distribution of the eigenvector in different time windows, construct a fault development trend graph. When the slope of the trend graph exceeds the safe operation threshold, trigger an emergency warning.

[0095] Among them, the determination condition for abnormal deviation is:

[0096] Calculate the Euclidean distance sequence between adjacent time window feature vectors in the normal operating state, establish a reference migration distance threshold. When the Euclidean distance between the new adjacent time window feature vectors exceeds 1.5 times of this reference migration distance threshold, it is recorded as a distance anomaly point. Then, calculate the migration direction angle sequence of the feature vectors in the state space, obtain the range of direction angle changes in the normal operating state. When the new migration direction angle exceeds plus or minus 20 degrees of this range, it is recorded as a direction anomaly point. Next, within 5 consecutive time windows, if 3 or more distance anomaly points and direction anomaly points appear simultaneously, it is determined that the migration trajectory has an abnormal deviation.

[0097] More specifically, judge the degree of oil circuit blockage according to the results of comparative analysis:

[0098] It is determined to be a mild blockage when all of the following conditions are met:

[0099] The sum of the divergence index of the vibration component and the attenuation index of the pressure component first exceeds the preset stability threshold;

[0100] The feature vector transfer matrix shows that the trajectory begins to have an abnormal deviation (meeting the given abnormal deviation determination conditions);

[0101] The growth rate of the ratio of the dominant eigenvalues of the vibration component and the pressure component first exceeds the preset threshold;

[0102] It is determined to be a moderate blockage when all of the following conditions are met:

[0103] Within 3 consecutive time windows, a feature vector mutation is detected (meeting the three given mutation determination conditions);

[0104] The change rate of the main eigenvalue of the covariance matrix continuously exceeds the preset threshold for 5 time windows;

[0105] The growth rate of the eigenvalue ratio continuously exceeds the preset threshold for 4 time windows.

[0106] It is determined to be a severe blockage when any of the following conditions is met:

[0107] Feature vector mutations are detected within 10 consecutive time windows;

[0108] The slope of the fault development trend graph exceeds the safe operation threshold;

[0109] The migration trajectory of the feature vectors in the state space continuously shows more than 5 abnormal deviations.

[0110] Early warning mechanism, and at the same time output the degree of blockage and the blockage location obtained by the above determination:

[0111] When it is determined that there is a mild blockage, the system issues a warning prompt;

[0112] When it is determined that there is a moderate blockage, the system issues a warning signal;

[0113] When it is determined that there is a severe blockage, the system triggers an emergency warning;

[0114] In summary, the lubrication system fault warning method based on data analysis according to the embodiments of the present invention is clarified. It comprehensively analyzes the vibration data and pressure data in the operation data of the lubrication system to determine whether there is an oil passage blockage in the lubrication system, and determines the specific blockage location and degree. In this way, it is possible to reduce the problem of poor lubrication caused by the accumulation and deposition of metal chips, sludge, and external impurities in the lubricating oil in the oil passage of the high-speed bearing lubrication system, resulting in oil passage blockage, and further causing overheating and damage of the bearing.

[0115] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned lubrication system fault warning method based on data analysis have been described in detail in the description of the lubrication system fault warning method based on data analysis above with reference to Figure 1 and therefore, the repeated description thereof will be omitted.

[0116] In summary, the lubrication system fault warning system based on data analysis according to the embodiments of the present invention is clarified. It comprehensively analyzes the vibration data and pressure data in the operation data of the lubrication system to determine whether there is an oil passage blockage in the lubrication system, and determines the specific blockage location and degree. In this way, it is possible to reduce the problem of poor lubrication caused by the accumulation and deposition of metal chips, sludge, and external impurities in the lubricating oil in the oil passage of the high-speed bearing lubrication system, resulting in oil passage blockage, and further causing overheating and damage of the bearing.

Claims

1. A lubrication system fault early warning method based on data analysis, characterized in that: include: Collecting operation data of the lubrication system, and performing denoising on the operation data based on wavelet transform; Constructing a coupling feature matrix of the bearing vibration data and the lubricating oil pressure data in the preprocessed operating data, and triggering a deep analysis when it is detected that the energy mutation point of the bearing vibration data and the abnormal point of the lubricating oil pressure data in the time dimension exceed a preset threshold, obtaining a fault feature vector and determining the blockage location; The fault feature vectors are compared and analyzed, and the blockage degree of the located oil circuit blockage position is evaluated and warned.

2. The lubrication system fault early warning method based on data analysis according to claim 1 is characterized in that: Using db4 wavelet basis to process the operation data of the lubrication system; Wherein, the vibration data in the lubrication system operation data is processed by using an improved adaptive threshold method; The improved adaptive threshold method comprises the following steps: Calculate the standard deviation of the wavelet decomposition coefficients of each layer and record the standard deviation as σ j ; Then according to the length N of the wavelet coefficients of this layer j Calculate the initial threshold λ j ; Calculate the local variance of the wavelet coefficients of this layer; When the local variance is greater than the preset variance threshold, it indicates that the region contains the mutation feature of the vibration signal, and the threshold λ of the region is increased. j Divided by the adaptation factor α; When the local variance is less than or equal to the preset variance threshold, the original threshold λ is maintained. j constant.

3. The lubrication system fault early warning method based on data analysis according to claim 2 is characterized in that: The value of the adaptive factor α is determined based on the energy distribution characteristics and mutation degree of the vibration signal, and is set according to the non-Gaussian characteristics of the vibration signal.

4. The lubrication system fault early warning method based on data analysis according to claim 3 is characterized in that: The energy mutation point is determined as follows: The vibration energy value of the preprocessed bearing vibration data is calculated according to a fixed time window to obtain a vibration energy sequence. The mean and standard deviation of the vibration energy sequence are calculated. When the vibration energy value at a certain moment exceeds the range of plus or minus three times the standard deviation of the mean, the moment is marked as a vibration energy mutation point. If multiple consecutive data points exceed the range, the first moment that exceeds the range is taken as the mutation point. The identification of the abnormal pressure point is as follows: Calculate the sliding mean of lubricating oil pressure data as the pressure reference value; When the deviation between the actual pressure value at a certain moment and the pressure reference value exceeds two standard deviations, the moment is marked as a candidate point of pressure anomaly; If the duration of the abnormal state exceeds a preset minimum duration threshold, the moment is confirmed as a pressure abnormality point.

5. The lubrication system fault early warning method based on data analysis according to claim 4 is characterized in that: The in-depth analysis extracts the fault feature vector by calculating the singular value decomposition of the coupling feature matrix; and locates the specific pipeline position of the oil circuit blockage by comparing the difference in the projection amplitude of the fault feature vector between adjacent measuring points.

6. The lubrication system fault early warning method based on data analysis according to claim 1 is characterized in that: The comparative analysis includes: Separating a vibration component and a pressure component from the fault characteristic vector; The fault feature vectors of adjacent sub-matrices are connected by sliding connection to form a vibration feature change sequence and a pressure feature change sequence. When the feature vector in a certain time window mutates, the window is marked as a feature mutation window. Calculate the covariance matrix of the vibration component and the pressure component in each time window, and extract the main eigenvalue of the covariance matrix; Construct the eigenvector transfer matrix between adjacent time windows and analyze the migration trajectory of the eigenvector in the state space. When the migration trajectory deviates abnormally, it is determined that the system performance begins to deteriorate. The divergence index of the vibration component and the attenuation index of the pressure component are calculated. When the sum of the two indexes exceeds the preset stability threshold, it is confirmed that the system has entered the fault development stage.

7. The lubrication system fault early warning method based on data analysis according to claim 6 is characterized in that: The determination conditions for the mutation of the feature vector include three determination conditions. When all three determination conditions are satisfied, it indicates that the feature vector in a certain time window has mutation.

8. The lubrication system fault warning system based on data analysis according to claim 5, characterized in that: Determining the degree of oil circuit blockage based on the results of the comparative analysis; The oil circuit blockage degree includes mild blockage, moderate blockage and severe blockage; Light congestion is determined when all of the following conditions are met: The sum of the divergence index of the vibration component and the attenuation index of the pressure component exceeds the preset stability threshold for the first time; The eigenvector transfer matrix shows that the trajectory begins to deviate abnormally; The growth rate of the ratio of the dominant eigenvalues ​​of the vibration component to the pressure component exceeds a preset threshold for the first time; Moderate congestion is determined when all of the following conditions are met: In three consecutive time windows, feature vector mutations were detected; The change rate of the main eigenvalue of the covariance matrix exceeds the preset threshold continuously for 5 time windows; The growth rate of the eigenvalue ratio continues to exceed the preset threshold for 4 time windows; Severe blockage is determined when any of the following conditions are met: The feature vector mutations were detected in 10 consecutive time windows; The slope of the fault development trend graph exceeds the safe operation threshold; The migration trajectory of the eigenvector in the state space shows abnormal deviations for more than 5 times in a row.

9. A lubrication system fault warning system based on data analysis, characterized in that: include: An acquisition module, used for acquiring operation data of the lubrication system and performing denoising processing on the operation data based on wavelet transform; A blockage position determination module is used to construct a coupling feature matrix of the bearing vibration data and the lubricating oil pressure data in the preprocessed operation data, and when it is detected that the overlap between the energy mutation point of the bearing vibration data and the abnormal point of the lubricating oil pressure data in the time dimension exceeds a preset threshold, a deep analysis is triggered to obtain a fault feature vector and determine the blockage position; A blockage degree determination module is used to compare and analyze the fault feature vectors and evaluate the blockage degree of the located oil circuit blockage position; The early warning module is used to issue an early warning according to the result of the congestion degree determination module.

Images