Degradation degree detection method and system for fan lubricating oil

Abstract

The invention relates to the related technical field of degradation degree detection, in particular to a fan lubricating oil degradation degree detection method and system, and the method comprises the steps: obtaining lubricating oil characteristic parameters, determining a first degradation sensitive characteristic set and a second degradation sensitive characteristic set, carrying out the correlation verification through combining historical fault data with a lubricating oil replacement record, correcting the deviation of a degradation degree evaluation function, and obtaining a degradation degree evaluation result. And obtaining a deterioration degree evaluation result, and generating a deterioration degree detection report. The technical problems that a degradation degree evaluation function depends on a single parameter, degradation inducement recognition is fuzzy, an evaluation result deviates from reality, and an accurate basis cannot be provided for lubricating oil replacement are solved, a degradation sensitive feature set dominated by pollutants and mechanical wear is distinguished, degradation inducement is accurately recognized, and the lubricating oil replacement efficiency is improved. The historical fault data and the replacement record are combined to carry out relevance verification and deviation correction on the deterioration degree evaluation function, the accuracy of the deterioration degree evaluation result is improved, and the generated deterioration degree detection report can provide a scientific basis for lubricating oil replacement and gearbox maintenance.

Description

Technical Field

[0001] This invention relates to the technical field of degradation detection, specifically to a method and system for detecting the degradation of fan lubricating oil. Background Technology

[0002] As a core piece of equipment in fields such as new energy power generation and industrial ventilation, the stable operation of the gearbox of a wind turbine is directly related to the power generation efficiency, service life and operational safety of the equipment. As a key medium for reducing friction, cooling and cleaning in the gearbox, the degree of deterioration of the lubricating oil will significantly affect the lubrication effect of the gearbox. If it deteriorates excessively and is not replaced in time, it may lead to accelerated gear wear, equipment failure and shutdown, or even safety accidents.

[0003] Current methods for detecting the deterioration of wind turbine lubricating oil rely on a single parameter (such as viscosity or acid value) to assess the degree of deterioration. They fail to systematically distinguish the different effects of contaminants and mechanical wear on deterioration, making it difficult to accurately capture the key causes of deterioration. The deterioration assessment function is prone to bias, resulting in inaccurate judgment of the degree of deterioration and failing to accurately reflect the complex process of lubricating oil deterioration.

[0004] In summary, existing technologies suffer from the problem that the degradation assessment function relies on a single parameter, leading to ambiguity in identifying degradation causes and deviations from reality in assessment results. This makes it impossible to provide accurate information for lubricant replacement, thereby increasing the risk of abnormal operation of the wind turbine gearbox. Summary of the Invention

[0005] This application provides a method and system for detecting the deterioration degree of wind turbine lubricating oil, aiming to solve the technical problem that the existing deterioration degree assessment function relies on a single parameter, leading to ambiguity in the identification of deterioration causes and deviation of assessment results from reality, thus failing to provide accurate basis for lubricating oil replacement and increasing the risk of abnormal operation of wind turbine gearboxes.

[0006] In view of the above problems, the technical solution to achieve the present application is as follows: In a first aspect, this application provides a method for detecting the degradation degree of wind turbine lubricating oil, wherein the method includes: obtaining lubricating oil characteristic parameters based on a lubricating oil sample from a wind turbine gearbox, the lubricating oil characteristic parameters including solid particle concentration, spectral data, and pressure data; determining a first degradation sensitive feature set under contaminant-dominated conditions and a second degradation sensitive feature set under mechanical wear-dominated conditions based on the lubricating oil characteristic parameters; performing correlation verification based on the first degradation sensitive feature set and the second degradation sensitive feature set, combined with historical fault data of the wind turbine gearbox and lubricating oil replacement records, correcting the deviation of the degradation degree evaluation function, and obtaining a degradation degree evaluation result; and generating a degradation degree detection report for the lubricating oil sample based on the degradation degree evaluation result.

[0007] Preferably, by combining the solid particle concentration and spectral data in the lubricating oil characteristic parameters with the equipment operating status of the fan gearbox, a coupled analysis is performed to determine the first set of sensitive deterioration features under the dominance of pollutants; at the same time, in the internal oil flow velocity field corresponding to the fan gearbox, the differential relationship of pollutant migration and diffusion is analyzed to determine the dynamic response coefficient as the pollutant diffusion kinetic index.

[0008] Preferably, by combining the pressure data and spectral data in the lubricating oil characteristic parameters with the equipment operating status of the fan gearbox, a coupled analysis is performed to determine the second set of sensitive features under mechanical wear dominance; at the same time, the ratio of the fluctuation amplitude of the pressure data during the meshing cycle to the peak area of ​​the metal-organic compound in the spectral data is obtained, and the ratio is normalized and used as the oil film degradation stability factor.

[0009] Preferably, a degradation feature matrix is ​​constructed based on the first degradation sensitive feature and the second degradation sensitive feature; the row vectors of the degradation feature matrix correspond to lubricating oil samples under multiple time windows, and the column vectors include the particle size distribution entropy and spectral contaminant characteristic peak intensity in the first degradation sensitive feature set, and the pressure fluctuation coefficient and spectral wear metal characteristic peak area in the second degradation sensitive feature set.

[0010] Preferably, the pollution diffusion kinetics index is used as the query vector of the attention network, the oil film degradation stability factor is used as the key vector of the attention network, and the degradation feature matrix is ​​used as the value vector of the attention network; the key degradation features under the weight distribution are dynamically focused through the attention mechanism.

[0011] Preferably, the attention mechanism uses a similarity metric based on kernel density estimation instead of inner product operation; the similarity index is obtained by obtaining the kernel density overlap area of ​​the query vector and the key vector: kernel density estimation is performed on the query vector and the key vector respectively, using a Gaussian kernel function, and the bandwidth is adaptively determined by the Silverman rule; the kernel density overlap area of ​​the two kernel density curves corresponding to the query vector and the key vector is calculated.

[0012] Preferably, based on the similarity index and the first similarity threshold and the second similarity threshold; if the first similarity threshold is met, the attention mechanism tilts the weight distribution towards the low-frequency temporal variation components of particle size distribution entropy and pressure fluctuation coefficient; if the second similarity threshold is met, the attention mechanism tilts the weight distribution towards the high-frequency transient change components of the intensity of spectral pollutant characteristic peaks and the area of ​​spectral wear metal characteristic peaks.

[0013] Preferably, the low-frequency time-series variation component and the high-frequency transient change component are obtained by performing signal decomposition on each column of the degradation feature matrix through empirical mode decomposition, which respectively correspond to the low-order and high-order components in multiple intrinsic mode functions, and the instantaneous amplitude and frequency features are extracted by combining Hilbert transform, and differential weighted fusion is performed.

[0014] Preferably, the key degradation features after differential weighting are input into the improved gated loop unit; the noise features are suppressed by the improved gated loop unit, and a degradation evaluation function is constructed based on the output features of the GRU.

[0015] In a second aspect, this application provides a system for detecting the deterioration degree of wind turbine lubricating oil. The system includes: a feature parameter acquisition module for acquiring lubricating oil feature parameters from a lubricating oil sample from a wind turbine gearbox, the lubricating oil feature parameters including solid particle concentration, spectral data, and pressure data; a deterioration sensitivity feature set determination module for determining a first deterioration sensitivity feature set dominated by contaminants and a second deterioration sensitivity feature set dominated by mechanical wear based on the lubricating oil feature parameters; a deviation correction module for verifying the correlation between the first and second deterioration sensitivity feature sets and historical fault data of the wind turbine gearbox and lubricating oil replacement records, correcting the deviation of the deterioration degree evaluation function, and obtaining a deterioration degree evaluation result; and a deterioration degree detection report generation module for generating a deterioration degree detection report for the lubricating oil sample based on the deterioration degree evaluation result.

[0016] In summary, one or more technical solutions provided in this application achieve the following technical effects: distinguishing between contaminants and mechanical wear-dominated degradation sensitive feature sets; accurately identifying degradation causes; verifying the correlation and correcting deviations of the degradation degree assessment function by combining historical fault data and replacement records; improving the accuracy of degradation degree assessment results; and generating degradation degree detection reports that provide a scientific basis for lubricant replacement and gearbox maintenance. Attached Figure Description

[0017] Figure 1 This application provides a flowchart illustrating a method for detecting the deterioration of fan lubricating oil.

[0018] Figure 2 This application provides a schematic diagram of the structure of a fan lubricating oil degradation detection system.

[0019] Figure labeling: Feature parameter acquisition module M100, degradation sensitive feature set determination module M200, deviation correction module M300, degradation degree detection report generation module M400. Detailed Implementation

[0020] Example 1: The present application will be described in detail below with reference to the accompanying drawings, as follows... Figure 1 As shown, this application provides a method for detecting the deterioration degree of fan lubricating oil, wherein the method includes: S1: Obtain lubricating oil characteristic parameters based on the lubricating oil sample of the wind turbine gearbox. The lubricating oil characteristic parameters include solid particle concentration, spectral data, and pressure data. S2: Based on the lubricating oil characteristic parameters, determine the first set of sensitive features for degradation under pollutant-dominated conditions and the second set of sensitive features for degradation under mechanical wear-dominated conditions.

[0021] Specifically, lubricating oil samples refer to data collected from different key parts of the wind turbine gearbox, including the oil outlet and gear meshing area, for subsequent characteristic parameter detection; solid particle concentration is the quantity or mass of lubricating oil per unit volume, a key indicator reflecting the degree of contamination, with common solid particles including dust, impurities, and wear debris; spectral data is spectral characteristic information obtained from lubricating oil samples through spectral analysis equipment, which can reveal the composition of contaminants in the oil, the metallic elements involved in gear wear, etc.; pressure data refers to the pressure value and fluctuation of the lubricating oil during circulation within the gearbox, collected by pressure sensors installed in the oil circuit, reflecting the operating status of the lubrication system, such as... Oil circuit blockage can lead to abnormally high pressure. Obtaining the characteristic parameters of lubricating oil involves testing lubricating oil samples with detection equipment to obtain the specific values ​​of the aforementioned parameters. The first set of sensitive features for deterioration under the dominance of contaminants is a combination of features that are more sensitive to the deterioration of lubricating oil caused by contaminants, such as the intensity of the spectral peaks of contaminant elements. The second set of sensitive features for deterioration under the dominance of mechanical wear is a combination of features that are more sensitive to the deterioration caused by wear debris generated by mechanical wear of components such as gears, such as the spectral peak area of ​​wear metal elements and the frequency of pressure fluctuations. Determining the first and second sets of sensitive features for deterioration refers to extracting features that are strongly correlated with the two types of dominant deterioration factors through the screening and analysis of characteristic parameters.

[0022] Execution steps: When obtaining the characteristic parameters of the lubricating oil, multiple parallel samples are collected from the fan gearbox to reduce errors. A laser particle counter is used to detect the concentration of solid particles. Spectral data is obtained through an atomic emission spectrometer, which can identify the characteristic peaks of elements such as silicon, iron, and copper. The peak intensity of Si corresponds to external dust contamination, while the peak intensity of Fe and Cu corresponds to gear wear. Pressure data is collected using a pressure sensor, and pressure fluctuations during the meshing cycle are recorded. The normal range is 2MPa-3MPa, and abnormal conditions exceed 4MPa. These steps provide raw data for subsequent analysis and are the basis for distinguishing the causes of deterioration.

[0023] When determining the first and second degradation-sensitive feature sets, a correlation analysis was performed on the proportion of 1μm-5μm particles in the solid particle concentration and the intensity of the Si element spectral peak. If the correlation coefficient was >0.85, the proportion of 1μm-5μm particles and the Si element peak intensity were selected to form the first degradation-sensitive feature set, reflecting degradation dominated by pollutants. The fluctuation amplitude of pressure data and the area of ​​the Fe element spectral peak were analyzed, and the Fe element peak area and pressure fluctuation coefficient were extracted to form the second degradation-sensitive feature set, reflecting degradation dominated by mechanical wear. The pressure fluctuation coefficient is the fluctuation amplitude / average pressure. Through the above steps, the degradation causes were accurately classified, providing targeted features for the subsequent construction of the evaluation function.

[0024] S3: Based on the first degradation sensitive feature set and the second degradation sensitive feature set, and combined with the historical fault data of the wind turbine gearbox and the lubricating oil replacement record, perform correlation verification, correct the deviation of the degradation degree evaluation function, and obtain the degradation degree evaluation result; S4: Based on the degradation degree evaluation result, generate the degradation degree detection report of the lubricating oil sample.

[0025] Specifically, correlation verification involves matching and analyzing the first and second sensitive feature sets of deterioration with historical fault data and lubricating oil replacement records of the wind turbine gearbox. Historical fault data includes gearbox fault types, occurrence times, and lubricating oil conditions at the time of fault over a past period. Lubricating oil replacement records include replacement time, lubricating oil characteristic parameters at the time of replacement, and running time. The verification process verifies the correlation between the feature sets and the actual deterioration results and fault causes. The deterioration degree assessment function is a mathematical model built based on the first and second sensitive feature sets, used to quantitatively calculate the degree of lubricating oil deterioration. Deviation correction involves adjusting the parameters of the deterioration degree assessment function based on the correlation verification results to reduce the deviation between the function's output value and the actual deterioration state. The deterioration degree assessment result is the quantitative value or level output by the corrected function, reflecting the current degree of lubricating oil deterioration. The deterioration degree detection report is a document generated from the comprehensive assessment results, containing the degree of deterioration, main causes, and recommended maintenance measures.

[0026] Execution steps: Extract the first and second degradation sensitive feature sets corresponding to gearbox bearing wear failures from historical fault data, as well as the feature data from normal oil replacement records; compare the current first and second degradation sensitive feature sets with the aforementioned historical fault data and calculate the correlation degree; if the degradation degree initially output by the degradation degree evaluation function deviates from the actual degradation level under the same conditions in the historical fault data, then correct it by adjusting the weight of the Fe element peak area in the degradation degree evaluation function to make the corrected evaluation result closer to the actual degradation level.

[0027] Through the above steps, the error between the corrected assessment results and the actual deterioration level is reduced, significantly improving the accuracy of the assessment. The deterioration detection report should include: deterioration assessment results, main deterioration causes, and related historical fault references. The main deterioration causes can be contaminant-driven or mechanical wear-driven. Based on the deterioration detection report, the complex process of lubricant deterioration is accurately reflected, and lubricant can be replaced in a timely manner to avoid an increase in the risk of abnormal equipment operation of the fan gearbox due to lubricant deterioration.

[0028] Furthermore, based on the lubricating oil characteristic parameters, a first set of degradation-sensitive features dominated by contaminants is determined. The method of this application includes: By combining the solid particle concentration and spectral data in the lubricating oil characteristic parameters with the equipment operating status of the fan gearbox, a coupled analysis is performed to determine the first set of sensitive features under the dominance of pollutants; at the same time, in the internal oil flow velocity field corresponding to the fan gearbox, the differential relationship of pollutant migration and diffusion is analyzed to determine the dynamic response coefficient as the pollutant diffusion kinetic index.

[0029] Specifically, coupling analysis refers to the correlation analysis between the characteristic parameters of lubricating oil and the equipment operating status parameters of the fan gearbox, such as speed, load, and running time. By establishing mathematical relationships between multiple parameters, the variation law of characteristic parameters under different equipment operating states is revealed. The equipment operating status includes parameters that reflect the working condition of the fan gearbox, such as the real-time speed, output torque, and cumulative running time of the gearbox.

[0030] The internal oil flow velocity field refers to the flow velocity distribution area formed by lubricating oil in the wind turbine gearbox under gear transmission and oil pump drive. Its characteristics can be simulated by fluid simulation. The differential relationship of pollutant migration and diffusion is a partial differential equation describing the migration and diffusion law of pollutants in the internal oil flow velocity field with time and space. The dynamic response coefficient is a coefficient that quantifies the degree of response of pollutant diffusion to changes in oil flow velocity and fluctuations in equipment operating status. The pollution diffusion kinetic index, i.e., the dynamic response coefficient, is used to characterize the diffusion rate and distribution characteristics of pollutants in the wind turbine gearbox and reflects the dynamic process of pollutant-dominated degradation.

[0031] Execution steps: Collect equipment operating status parameters and simultaneously obtain solid particle concentration and spectral data for the corresponding time period; establish a correlation model through coupling analysis. For example, when the rotation speed increases by 10%, if the proportion of 10-20μm particles in the solid particle concentration increases by 15% and the Si element peak intensity increases by 12% simultaneously, then the proportion of 10-20μm particles and the Si element peak intensity are included in the first degradation sensitive feature set. Simultaneously, in the internal oil flow velocity field analysis, the oil flow velocity distribution is obtained based on fluid simulation. The rate of change of pollutant concentration at different locations is calculated using the pollutant migration and diffusion differential equation, thereby determining the dynamic response coefficient. Specifically, the pollutant migration and diffusion differential equation is as follows: Where c is the pollutant concentration, t is time, v is the oil flow velocity, and D is the diffusion coefficient; the dynamic response coefficient is used as the pollution diffusion kinetic index. Coupled analysis strengthens the correlation between characteristic parameters and pollutant degradation, making the first degradation-sensitive feature set more targeted, while the pollution diffusion kinetic index quantifies the dynamic characteristics of pollutant diffusion, providing a time-dimensional reference for subsequent degradation assessment.

[0032] Furthermore, based on the lubricating oil characteristic parameters, a second set of degradation-sensitive features under mechanical wear-dominated conditions is determined. The method of this application includes: By combining the pressure data and spectral data in the lubricating oil characteristic parameters with the equipment operating status of the fan gearbox, a coupled analysis is performed to determine the second set of sensitive features under the dominance of mechanical wear. At the same time, the ratio of the fluctuation amplitude of the pressure data during the meshing cycle to the peak area of ​​the metal-organic compound in the spectral data is obtained, and the ratio is normalized and used as the oil film degradation stability factor.

[0033] Specifically, the meshing cycle refers to the complete time from the initial contact to the disengagement of a pair of gear teeth in a wind turbine gearbox, and its length is related to the gear speed; the fluctuation amplitude of pressure data within the meshing cycle refers to the difference between the maximum and minimum values ​​of lubricating oil pressure within a meshing cycle, reflecting the drastic pressure change during gear meshing; the metal-organic compound peak area refers to the integral area of ​​the characteristic peaks corresponding to compounds formed by the combination of metal and organic components generated by mechanical wear in the lubricating oil, detected by spectral analysis, and its size is directly related to the amount of wear products; normalization processing converts the ratio of pressure fluctuation amplitude to metal-organic compound peak area to a numerical range of 0-1, eliminating dimensional differences under different equipment or testing conditions, and making the results comparable; the oil film degradation stability factor is the above ratio after normalization processing, used to quantify the stability of the lubricating oil film under mechanical wear; coupling analysis specifically refers to correlating pressure data, spectral data, and the equipment operating status of the gearbox to establish mathematical relationships between multiple parameters in order to screen out features strongly correlated with mechanical wear.

[0034] Execution steps: Collect equipment operating status parameters, and simultaneously acquire pressure and spectral data for the corresponding time period. For example, the pressure fluctuation amplitude during the meshing cycle is 0.4 MPa, and the peak area of ​​the iron-based organometallic compound is 3200 area units. The area unit is a value obtained by integrating the spectral curve within the wavelength range corresponding to the characteristic peak, reflecting the total amount of the substance corresponding to the characteristic peak. The horizontal axis of the spectral curve represents wavelength / wavenumber, and the vertical axis represents absorbance / intensity. Establish a corresponding correlation model through coupling analysis. For example, if the pressure fluctuation amplitude increases by 15% and the peak area of ​​the iron-based organometallic compound increases by 20% when the load increases by 10%, then the pressure fluctuation amplitude and the peak area of ​​the iron-based organometallic compound are included in the second degradation sensitive feature set under mechanical wear dominance. Simultaneously, calculate the ratio of the pressure fluctuation amplitude to the organometallic compound peak area, i.e., 0.4 MPa / 3200 = 1.25 × 10⁻⁶. -4 Then, normalization is performed to obtain the oil film degradation stability factor, which is the historical maximum ratio of the wind turbine gearbox, 2 × 10⁻⁶. -4 Based on this, the normalized result is 0.625. In the above steps, the coupling analysis ensures that the second degradation-sensitive feature set can accurately reflect the impact of mechanical wear on lubricant degradation, effectively supporting the accurate differentiation of degradation causes.

[0035] Furthermore, the method of this application also includes: Based on the first and second degradation sensitive features, a degradation feature matrix is ​​constructed. The row vectors of the degradation feature matrix correspond to lubricating oil samples under multiple time windows, and the column vectors include the particle size distribution entropy and spectral contaminant characteristic peak intensity in the first degradation sensitive feature set, and the pressure fluctuation coefficient and spectral wear metal characteristic peak area in the second degradation sensitive feature set.

[0036] Specifically, the degradation feature matrix is ​​a two-dimensional data table formed by organizing the key features of the first and second degradation sensitive feature sets according to time series, used to systematically integrate degradation feature information at different time points; the time window refers to the sampling period divided by fixed time intervals, with each time window corresponding to the collection and testing of a lubricating oil sample; the particle size distribution entropy is an indicator describing the uniformity of solid particle size distribution in lubricating oil, with a higher entropy value indicating a more dispersed particle size distribution, reflecting the complexity of contaminant sources; the intensity of spectral contaminant characteristic peaks is the peak height of the corresponding contaminant characteristic peaks in the spectral data, directly characterizing the contaminant content; the pressure fluctuation coefficient is the ratio of the fluctuation amplitude of the pressure data to the average pressure, used to quantify the relative severity of pressure changes, reflecting the lubrication stability during gear meshing.

[0037] The main material of gears usually contains iron. The area of ​​the wear metal characteristic peak in the spectrum is the integral area of ​​the wear metal characteristic peak corresponding to the iron element in the spectral data. It can reflect the total amount of wear metal more comprehensively than the peak intensity. The size of the integral area is positively correlated with the content of iron filings generated by gear wear in the lubricating oil. That is, the larger the integral area, the more severe the gear wear and the more iron-based wear filings are generated. Constructing the degradation feature matrix means filling the matrix with the feature parameter values ​​under each time window according to the rule that the row vector corresponds to the time window and the column vector corresponds to the specific feature, forming a structured data set.

[0038] Execution steps: Determine the time window, and for the lubricating oil sample in each time window, detect and calculate the feature parameters corresponding to the column vector. For example, in the first time window, obtain the particle size distribution entropy, the intensity of the spectral contaminant characteristic peak, the pressure fluctuation coefficient, and the area of ​​the spectral wear metal characteristic peak corresponding to iron. Similarly, based on multiple time windows, form a degradation feature matrix. Systematically integrate the scattered multi-dimensional and multi-time point feature data to provide structured input for focusing on key degradation features through attention mechanisms. Matrix processing improves the utilization rate of feature data and can intuitively reflect the changing trend of each feature over time, providing a data foundation for accurately identifying the dynamic changes of degradation causes.

[0039] Furthermore, the method of this application also includes: The pollution diffusion kinetics index is used as the query vector of the attention network, the oil film degradation stability factor is used as the key vector of the attention network, and the degradation feature matrix is ​​used as the value vector of the attention network; the key degradation features under the weight distribution are dynamically focused through the attention mechanism.

[0040] Specifically, the attention network is an algorithmic model that simulates the human attention mechanism, automatically focusing on key content from a large amount of information. The query vector is used to retrieve information, using the pollution diffusion kinetics index as the query basis, to locate key features related to pollutant degradation. The key vector is used to match the query vector, using the oil film degradation stability factor as the matching basis, to filter key features related to mechanical wear degradation. The value vector is the original data vector to be processed, i.e., the degradation feature matrix, containing the specific values ​​of various degradation features. The attention mechanism assigns weights to different features in the value vector by calculating the similarity between the query vector and the key vector; the higher the weight, the more critical the feature. The weight distribution refers to the importance percentage assigned to each column of features in the degradation feature matrix. The dynamic focusing of key degradation features means that as the pollution diffusion kinetics index, oil film degradation stability factor, and degradation feature matrix change, the weights of each feature are adjusted in real time, so that the model prioritizes the features that have the most significant impact on degradation.

[0041] Execution steps: Determine the input vector for the attention mechanism: Use the pollution diffusion kinetic index with timestamps as the query vector, the oil film degradation stability factor under the same timestamp as the key vector, and the degradation feature matrix as the value vector; calculate the similarity between the query vector and the key vector, and assign weights to each feature in the value vector based on the similarity. Multiply the weights by the values ​​of the corresponding features in the value vector to obtain a weighted feature combination, thereby achieving dynamic focusing on the current key degradation features; In the above steps, the attention mechanism automatically distinguishes the importance of features, solving the problem of key information being overwhelmed by treating all features equally, improving the identification response efficiency of pollutant-dominated degradation and the identification accuracy of mechanical wear-dominated degradation, and providing more targeted feature inputs for the subsequent construction of degradation degree assessment functions.

[0042] Furthermore, the method of this application includes: The attention mechanism replaces the inner product operation with a similarity metric based on kernel density estimation. The similarity index is obtained by obtaining the kernel density overlap area between the query vector and the key vector: kernel density estimation is performed on the query vector and the key vector respectively, using a Gaussian kernel function, and the bandwidth is adaptively determined by the Silverman rule; the kernel density overlap area of ​​the two kernel density curves corresponding to the query vector and the key vector is calculated.

[0043] Specifically, kernel density estimation is used to estimate the probability density function of a random variable. It fits a smooth density curve by assigning kernel function weights around data points, describing the distribution characteristics of query and key vectors. The Gaussian kernel function is a commonly used kernel function in kernel density estimation: K( )= , where K( ) represents the weight values ​​of the kernel function. Represents the standardized distance. = , The smaller the absolute value, the closer it is to the target point, and the greater the weight assigned by the kernel function. Standardization eliminates the influence of the original dimensions of the data and effectively captures the local distribution characteristics of the data. The Silverman rule is an adaptive method for determining the bandwidth in kernel density estimation, with the formula h=0.9. min( (IQR / 1.34) Where h is the bandwidth. is the standard deviation, IQR is the interquartile range, and n is the sample size.

[0044] The bandwidth is automatically adjusted based on data characteristics to ensure the accuracy of density estimation. The bandwidth is used to control the smoothness of the kernel function. The kernel density overlap area refers to the area of ​​the overlapping region between the kernel density curves of the query vector and the key vector. The larger the area, the more similar the distributions of the two vectors are, and vice versa. It serves as an indicator to measure the similarity between the two. The inner product operation is a method in the attention mechanism to calculate the similarity between the query and the key. The inner product operation refers to calculating the sum of the products of corresponding elements of the two vectors. However, the similarity metric based on kernel density estimation replaces the inner product operation by using the kernel density overlap area to characterize the similarity, which better reflects the similarity of the overall distribution of the vectors rather than simple numerical matching.

[0045] Execution steps: First, perform kernel density estimation on the query vector and key vector. If the pollution diffusion kinetic index sequence of a randomly selected time period for the query vector is [0.62, 0.65, 0.68, 0.70], and the oil film degradation stability factor sequence for the corresponding time period for the key vector is [0.50, 0.53, 0.55, 0.58], then use a Gaussian kernel function and calculate the bandwidth using the Silverman rule to fit two smooth kernel density curves. Next, calculate the area of ​​the overlapping region of the two curves by integration, and calculate the kernel density overlap area of ​​the two kernel density curves corresponding to the query vector and key vector. Preferably, when the query and key vectors are non-linearly correlated, the kernel density overlap area more accurately measures the similarity between the query and key. Compared to the inner product operation, which only focuses on the sum of numerical products, it can capture the overall distribution characteristics of the vectors, reduce the interference of outliers on similarity judgment, and provide a more reliable similarity basis for the weight distribution of the subsequent attention mechanism.

[0046] Furthermore, the method of this application includes: Based on the similarity index and the first similarity threshold and the second similarity threshold; if the first similarity threshold is met, the attention mechanism will distribute the weights towards the low-frequency temporal variation components of particle size distribution entropy and pressure fluctuation coefficient; if the second similarity threshold is met, the attention mechanism will distribute the weights towards the high-frequency transient change components of the intensity of spectral pollutant characteristic peaks and the area of ​​spectral wear metal characteristic peaks.

[0047] Specifically, the similarity index is the quantitative similarity value between the query vector and the key vector obtained through the kernel density overlap area, as mentioned above. The first and second similarity thresholds are preset thresholds used to judge the degree of similarity and to divide different similarity intervals. The low-frequency temporal variation component refers to the part of the degradation feature that changes slowly and periodically over time, reflecting the long-term trend of degradation and is related to particle size distribution entropy and pressure fluctuation coefficient. The high-frequency transient mutation component refers to the part of the degradation feature that changes suddenly and significantly in a short period of time, reflecting the sudden anomaly of degradation and is related to the intensity of spectral pollutant characteristic peaks and the area of ​​spectral wear metal characteristic peaks. The weight distribution tilts towards the low-frequency temporal variation components of particle size distribution entropy and pressure fluctuation coefficient, and towards the high-frequency transient mutation components of spectral pollutant characteristic peak intensity and spectral wear metal characteristic peak area. This means that the attention mechanism increases the weight ratio of the corresponding feature components, so that the model focuses more on the corresponding part of the information during calculation, ensuring that the weight distribution reflects both similarity and feature difference, thereby avoiding the dominant influence of vector angle on weight allocation.

[0048] Execution steps: First and second similarity thresholds are preset. The calculated similarity index is compared with both. If the similarity index is 0.7 and the first similarity threshold is 0.6, then the first similarity threshold is met, indicating a high degree of similarity between the distribution trends of the pollution diffusion kinetics index and the oil film degradation stability factor. In this case, the attention mechanism increases the weight of the low-frequency component of the particle size distribution entropy, while correspondingly decreasing the weight of the high-frequency component, focusing on the long-term degradation trend. If the similarity index is 0.2 and the second similarity threshold is 0.3, it indicates a significant difference in the distribution trends of the two, with a sudden increase in the pollution diffusion index and a decrease in the oil film stability factor. If the qualitative factor drops sharply, the weight of the high-frequency components of the intensity of the characteristic peaks of spectral pollutants and the area of ​​the characteristic peaks of spectral wear metals will be increased to focus on capturing sudden deterioration anomalies. In the above steps, the weights are selectively tilted by dividing the weights by threshold values, so that the model can adaptively focus on key components according to the dynamic relationship of deterioration features. When the first similarity threshold is met, focusing on low-frequency components can improve the accuracy of long-term deterioration trend prediction. When the second similarity threshold is met, focusing on high-frequency components can shorten the response time for identifying sudden deterioration events, providing more accurate feature inputs for timely warning of potential gearbox failures.

[0049] Furthermore, the method of this application includes: The low-frequency time-series variation component and the high-frequency transient change component are obtained by performing signal decomposition on each column of the degradation feature matrix through empirical mode decomposition. They correspond to the low-order and high-order components in multiple intrinsic mode functions, respectively. The instantaneous amplitude and frequency features are extracted by combining Hilbert transform and then differentially weighted fusion is performed.

[0050] Specifically, Empirical Mode Decomposition (EMD) is an adaptive signal decomposition method that does not require pre-defined basis functions. It decomposes nonlinear and non-stationary time-series signals into multiple physically meaningful Eigenmode Functions (IMFs), which are used to decompose each column of the degradation feature matrix to extract different frequency components. IMFs are components obtained from EMD decomposition and must satisfy the condition that the number of extrema and zero-crossings are equal or differ by at most one, and the mean of the upper and lower envelopes is 0. Lower-order IMFs correspond to slowly changing low-frequency components in the signal, while higher-order IMFs correspond to rapidly changing high-frequency components. Hilbert Transform (HFT) is a mathematical transformation performed on each IMF, extracting instantaneous amplitude and frequency by calculating the analytic signal to quantify the dynamic characteristics of the feature components. Differential weighted fusion assigns different weights to low-frequency time-series components and high-frequency transient abrupt components based on their importance at different degradation stages, and then merges them. For example, the weight of low-frequency components is increased during stable degradation, and the weight of high-frequency components is increased during sudden degradation to highlight key information.

[0051] Execution steps: Perform Empirical Mode Decomposition (EMD) on each column of the degraded feature matrix: Iteratively filter components that meet the IMF conditions using the EMD algorithm. If three IMFs are obtained, the low-order IMF1 has a frequency range of 0.02Hz-0.1Hz, corresponding to low-frequency time-series variation components, reflecting the slow accumulation of features over time; the high-order IMF2 has a frequency range of 0.5Hz-2Hz, corresponding to high-frequency transient change components, reflecting sudden changes in a short period of time. Perform Hilbert transform on each IMF to extract instantaneous amplitude and instantaneous frequency. Perform differentiated weighted fusion based on the degraded stage: If the similarity index meets the first similarity threshold, it is considered a stable operating period. If the current period is stable, the weight corresponding to the low-frequency time-series variation components is set to exceed 0.5, and the weight corresponding to the high-frequency transient change components is set to 0.3. If the similarity index meets the second similarity threshold, it is considered an abnormal fluctuation period. If the current period is abnormal, the weight corresponding to the high-frequency transient change components is set to exceed 0.5, and the weight corresponding to the low-frequency time-series variation components is set to 0.4. After fusion, a more comprehensive feature vector is formed.

[0052] In the above steps, the temporal information of the original features is transformed into quantifiable dynamic features through signal decomposition and transformation, which solves the problem that it is difficult to distinguish between fast and slow changes when traditional features are used directly. Preferably, after EMD decomposition, the low-frequency components are associated with long-term degradation trends, and the high-frequency components are associated with sudden failures. Combined with the instantaneous features of Hilbert transform, the time sensitivity of degradation assessment is improved when the differentiated weighted fusion features are input into the subsequent model, providing more refined feature support for accurately assessing the degradation state of lubricating oil.

[0053] Furthermore, the method of this application includes: The key degradation features, after differential weighting, are input into the improved gated loop unit; the improved gated loop unit suppresses noise features, and a degradation evaluation function is constructed based on the output features of the GRU.

[0054] Specifically, the improved gated loop unit is an optimization of the conventional gated loop unit, retaining the core capabilities of the GRU in processing time-series data. The improvements lie in enhancing the noise filtering ability of the gating mechanism or improving the memory effect of long-term time-series features. It is suitable for processing time-series data on lubricating oil degradation characteristics. The core capability of the GRU in processing time-series data is to capture time-series dependencies through update and reset gates. The differentiated weighted key degradation features refer to the feature vectors obtained after empirical mode decomposition, Hilbert transform, and weight fusion. These feature vectors contain the weighted results of low-frequency time-series variation components and high-frequency transient change components, reflecting the differences between... Key information in the same degradation stage; noise characteristics refer to interference information in the feature data that is unrelated to lubricant degradation, such as random errors of detection equipment, accidental data fluctuations caused by ambient temperature fluctuations, etc.; noise suppression features are the process by which the improved GRU filters interference data through a gating mechanism to highlight effective degradation features. Specifically, it updates the retention ratio of historical information by the gating control and resets the current information filtering degree of the gating control; constructing a degradation degree evaluation function based on the output features of the GRU refers to mapping the denoised features output by the GRU after processing to a quantified degradation degree value. This mapping from features to degradation degree is usually achieved through a regression model or neural network.

[0055] Execution steps: The differentially weighted key degradation features are input into an improved gated recurrent unit (GRU), containing the weighted results of low-frequency temporal variation components and high-frequency transient change components. The improved GRU processes the temporal features through optimized update and reset gates: the update gate dynamically adjusts the retention weights of historical features, while the reset gate strengthens the memory of effective historical features, thereby suppressing noise. After GRU processing, a denoised feature sequence is output, smoothing out noise-induced jumps. A degradation evaluation function is constructed based on the GRU's output features. A neural network is used to fit the relationship between the features and the historically labeled degradation degree to evaluate the degradation. The degradation degree evaluation function is biased to control the deviation between the output value of the degradation degree evaluation function and the actual degradation degree within a preset threshold, thereby improving the accuracy of the degradation degree evaluation results. In the above steps, the improved GRU is used to process the time-series features and suppress noise, providing high-quality input features for the degradation degree evaluation function. This solves the problem of evaluation bias caused by noise interference in the original features. Preferably, the improved goodness of fit of the evaluation function lays an accurate feature foundation for subsequent correlation verification and deviation correction by combining historical data, ensuring that the final degradation degree evaluation result can truly reflect the degradation state of the lubricating oil.

[0056] In summary, the beneficial effects of the embodiments of this application are: This application provides a method and system for detecting the deterioration of wind turbine lubricating oil. It achieves the technical effect of distinguishing between the deterioration-sensitive feature sets dominated by contaminants and mechanical wear, accurately identifying deterioration causes, verifying the correlation and correcting the bias of the deterioration assessment function by combining historical fault data and replacement records, and generating a deterioration detection report for the lubricating oil sample. This method and system improves the accuracy of the deterioration assessment results by combining historical fault data and replacement records to obtain a scientific basis for lubricating oil replacement and gearbox maintenance.

[0057] Example 2, based on the same inventive concept as the method for detecting the deterioration of fan lubricating oil in the foregoing examples, such as... Figure 2 As shown in the figure, this application provides a system for detecting the deterioration degree of fan lubricating oil, wherein the system includes: Feature parameter acquisition module M100: Based on the lubricating oil sample of the wind turbine gearbox, acquire the lubricating oil feature parameters, including solid particle concentration, spectral data, and pressure data.

[0058] Deterioration Sensitive Feature Set Determination Module M200: Based on the lubricating oil characteristic parameters, determine the first deterioration sensitive feature set dominated by contaminants and the second deterioration sensitive feature set dominated by mechanical wear.

[0059] Deviation correction module M300: Based on the first degradation sensitive feature set and the second degradation sensitive feature set, and combined with the historical fault data of the wind turbine gearbox and the lubricating oil replacement record, the correlation verification is performed to correct the deviation of the degradation degree evaluation function and obtain the degradation degree evaluation result.

[0060] Deterioration Detection Report Generation Module M400: Generates a deterioration detection report for the lubricating oil sample based on the deterioration assessment results.

[0061] Furthermore, the degradation-sensitive feature set determination module M200 is used to perform the following method: By combining the solid particle concentration and spectral data in the lubricating oil characteristic parameters with the equipment operating status of the fan gearbox, a coupled analysis is performed to determine the first set of sensitive features under the dominance of pollutants; at the same time, in the internal oil flow velocity field corresponding to the fan gearbox, the differential relationship of pollutant migration and diffusion is analyzed to determine the dynamic response coefficient as the pollutant diffusion kinetic index.

[0062] Furthermore, the degradation-sensitive feature set determination module M200 is used to perform the following method: By combining the pressure data and spectral data in the lubricating oil characteristic parameters with the equipment operating status of the fan gearbox, a coupled analysis is performed to determine the second set of sensitive features under the dominance of mechanical wear. At the same time, the ratio of the fluctuation amplitude of the pressure data during the meshing cycle to the peak area of ​​the metal-organic compound in the spectral data is obtained, and the ratio is normalized and used as the oil film degradation stability factor.

[0063] Furthermore, the degradation-sensitive feature set determination module M200 is also used to perform the following method: Based on the first and second degradation sensitive features, a degradation feature matrix is ​​constructed. The row vectors of the degradation feature matrix correspond to lubricating oil samples under multiple time windows, and the column vectors include the particle size distribution entropy and spectral contaminant characteristic peak intensity in the first degradation sensitive feature set, and the pressure fluctuation coefficient and spectral wear metal characteristic peak area in the second degradation sensitive feature set.

[0064] Furthermore, the degradation-sensitive feature set determination module M200 is also used to perform the following method: The pollution diffusion kinetics index is used as the query vector of the attention network, the oil film degradation stability factor is used as the key vector of the attention network, and the degradation feature matrix is ​​used as the value vector of the attention network; the key degradation features under the weight distribution are dynamically focused through the attention mechanism.

[0065] Furthermore, the degradation-sensitive feature set determination module M200 is also used to perform the following method: The attention mechanism replaces the inner product operation with a similarity metric based on kernel density estimation. The similarity index is obtained by obtaining the kernel density overlap area between the query vector and the key vector: kernel density estimation is performed on the query vector and the key vector respectively, using a Gaussian kernel function, and the bandwidth is adaptively determined by the Silverman rule; the kernel density overlap area of ​​the two kernel density curves corresponding to the query vector and the key vector is calculated.

[0066] Furthermore, the degradation-sensitive feature set determination module M200 is also used to perform the following method: Based on the similarity index and the first similarity threshold and the second similarity threshold; if the first similarity threshold is met, the attention mechanism will distribute the weights towards the low-frequency temporal variation components of particle size distribution entropy and pressure fluctuation coefficient; if the second similarity threshold is met, the attention mechanism will distribute the weights towards the high-frequency transient change components of the intensity of spectral pollutant characteristic peaks and the area of ​​spectral wear metal characteristic peaks.

[0067] Furthermore, the degradation-sensitive feature set determination module M200 is also used to perform the following method: The low-frequency time-series variation component and the high-frequency transient change component are obtained by performing signal decomposition on each column of the degradation feature matrix through empirical mode decomposition. They correspond to the low-order and high-order components in multiple intrinsic mode functions, respectively. The instantaneous amplitude and frequency features are extracted by combining Hilbert transform and then differentially weighted fusion is performed.

[0068] Furthermore, the deviation correction module M300 is also used to perform the following method: The key degradation features, after differential weighting, are input into the improved gated loop unit; the improved gated loop unit suppresses noise features, and a degradation evaluation function is constructed based on the output features of the GRU.

[0069] In summary, any step can be stored as a computer instruction or program in an unrestricted computer memory and can be called and recognized by an unrestricted computer processor; no further restrictions are imposed here.

[0070] Furthermore, the above technical solutions only embody the preferred technical solutions of the embodiments of this application. Any changes that those skilled in the art may make to certain parts of these solutions embody the novel principles of the embodiments of this application. Obviously, those skilled in the art can make various modifications and variations to this application without departing from the scope of this application.

Claims

1. A method for detecting the deterioration degree of fan lubricating oil, characterized in that, The method includes: Based on the lubricating oil sample of the wind turbine gearbox, the characteristic parameters of the lubricating oil are obtained, including solid particle concentration, spectral data, and pressure data; Based on the lubricating oil characteristic parameters, a first set of sensitive features for degradation under the dominance of contaminants and a second set of sensitive features for degradation under the dominance of mechanical wear are determined. Based on the first set of degradation sensitive features and the second set of degradation sensitive features, the correlation is verified by combining the historical fault data of the wind turbine gearbox with the lubricating oil replacement records, the deviation of the degradation degree evaluation function is corrected, and the degradation degree evaluation result is obtained. Based on the degradation assessment results, a degradation test report for the lubricating oil sample is generated.

2. The method for detecting the deterioration degree of fan lubricating oil as described in claim 1, characterized in that, Based on the lubricating oil characteristic parameters, a first set of degradation-sensitive features under contaminant-dominated conditions is determined, the method comprising: By combining the solid particle concentration and spectral data in the lubricating oil characteristic parameters with the equipment operating status of the fan gearbox, a coupled analysis is performed to determine the first set of sensitive features for degradation under the dominance of pollutants. Meanwhile, in the internal oil flow velocity field corresponding to the wind turbine gearbox, the differential relationship of pollutant migration and diffusion is analyzed, and the dynamic response coefficient is determined as the pollution diffusion dynamic index.

3. The method for detecting the deterioration degree of fan lubricating oil as described in claim 2, characterized in that, Based on the lubricating oil characteristic parameters, a second set of degradation-sensitive features under mechanical wear-dominated conditions is determined, the method comprising: By combining the pressure data and spectral data in the lubricating oil characteristic parameters with the equipment operating status of the fan gearbox, a coupled analysis is performed to determine the second set of sensitive features under mechanical wear. Simultaneously, the ratio of the fluctuation amplitude of the pressure data during the meshing cycle to the peak area of ​​the organometallic compound in the spectral data is obtained, and the ratio is normalized and used as the oil film degradation stability factor.

4. The method for detecting the deterioration degree of fan lubricating oil as described in claim 3, characterized in that, The method further includes: Based on the first degradation-sensitive feature and the second degradation-sensitive feature, a degradation feature matrix is ​​constructed; The row vectors of the degradation feature matrix correspond to lubricating oil samples under multiple time windows, and the column vectors include the particle size distribution entropy and the intensity of spectral contaminant characteristic peaks in the first degradation sensitive feature set, and the pressure fluctuation coefficient and the area of ​​spectral wear metal characteristic peaks in the second degradation sensitive feature set.

5. The method for detecting the deterioration degree of fan lubricating oil as described in claim 4, characterized in that, The method further includes: The pollution diffusion dynamics index is used as the query vector of the attention network, the oil film degradation stability factor is used as the key vector of the attention network, and the degradation feature matrix is ​​used as the value vector of the attention network. Dynamically focus on key degradation features under weighted distribution using an attention mechanism.

6. The method for detecting the deterioration degree of fan lubricating oil as described in claim 5, characterized in that, The method includes: The attention mechanism replaces the inner product operation with a similarity metric based on kernel density estimation; The similarity index is obtained by using the kernel density overlap area of ​​the query vector and the key vector: kernel density is estimated for the query vector and the key vector respectively, using a Gaussian kernel function, and the bandwidth is adaptively determined by the Silverman rule; the kernel density overlap area of ​​the two kernel density curves corresponding to the query vector and the key vector is calculated.

7. The method for detecting the deterioration degree of fan lubricating oil as described in claim 6, characterized in that, The method includes: Based on the aforementioned similarity index and the first similarity threshold and the second similarity threshold; If the first similarity threshold is met, the attention mechanism will shift the weight distribution towards the low-frequency temporal variation components of particle size distribution entropy and pressure fluctuation coefficient. If the second similarity threshold is met, the attention mechanism will distribute the weights towards the high-frequency transient change components of the intensity of the spectral pollutant characteristic peaks and the area of ​​the spectral wear metal characteristic peaks.

8. The method for detecting the deterioration degree of fan lubricating oil as described in claim 7, characterized in that, The method includes: The low-frequency time-series variation component and the high-frequency transient change component are obtained by performing signal decomposition on each column of the degradation feature matrix through empirical mode decomposition. They correspond to the low-order and high-order components in multiple intrinsic mode functions, respectively. The instantaneous amplitude and frequency features are extracted by combining Hilbert transform and then differentially weighted fusion is performed.

9. The method for detecting the deterioration degree of fan lubricating oil as described in claim 8, characterized in that, The method includes: The key degradation features, after differential weighting, are input into the improved gated loop unit; The improved gated loop unit suppresses noise characteristics, and a degradation evaluation function is constructed based on the output characteristics of the GRU.

10. A system for detecting the deterioration degree of fan lubricating oil, characterized in that, The system is used to implement the method for detecting the deterioration degree of a fan lubricating oil according to any one of claims 1-9, wherein the system comprises: Feature parameter acquisition module: Based on the lubricating oil sample of the wind turbine gearbox, acquire the lubricating oil feature parameters, including solid particle concentration, spectral data, and pressure data; Deterioration-sensitive feature set determination module: Based on the lubricating oil characteristic parameters, determine the first deterioration-sensitive feature set under the dominance of contaminants and the second deterioration-sensitive feature set under the dominance of mechanical wear; Deviation correction module: Based on the first degradation sensitive feature set and the second degradation sensitive feature set, and combined with the historical fault data of the wind turbine gearbox and the lubricating oil replacement record, the module performs correlation verification, corrects the deviation of the degradation degree evaluation function, and obtains the degradation degree evaluation result. Deterioration test report generation module: Based on the deterioration assessment results, generate a deterioration test report for the lubricating oil sample.

Images