A method for monitoring the health status of wind turbine sliding bearings throughout their entire lifecycle based on multi-information fusion
By integrating lubricating film thickness, acoustic emission, oil abrasive and vibration analysis technologies, and combining principal component analysis and hidden Markov models, we have achieved full life-cycle health status monitoring of wind turbine sliding bearings, solved the problems of early fault warning and wear evolution, and improved the service life and stability of bearings.
Patent Information
- Application Number
- CN202310601211.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-25
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2043-05-25
AI Technical Summary
Existing technologies cannot achieve full life-cycle health monitoring of wind turbine sliding bearings, especially lacking effective means in early fault warning and wear evolution, making it difficult to prevent bearing failure.
A multi-information fusion method is adopted, integrating lubricating film thickness detection, acoustic emission collision location detection, oil abrasive analysis and vibration analysis technologies. Multiple information is acquired through ultrasonic sensors, acoustic emission sensors, oil sampling systems and piezoelectric acceleration sensors, and combined with principal component analysis and hidden Markov models to assess the health status.
It enables accurate prediction and assessment of the health status of wind turbine sliding bearings, allowing for early identification of faults, extension of bearing life, and improvement of operational stability.
Smart Images

Figure CN117469098B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of online monitoring of mechanical systems, specifically relating to a method for monitoring the health status of wind turbine sliding bearings throughout their entire life cycle based on multi-information fusion. Background Technology
[0002] Bearings, acting as the "joints" of wind turbines, are crucial for ensuring the stable and reliable operation of these units. Wind turbine bearings experience complex stress and vibration conditions, must withstand enormous impact loads, and must also achieve a lifespan of at least 20 years, the same as the main turbine. The harsh operating conditions and long-life, high-reliability requirements of wind power equipment make wind turbine bearings highly technically complex, widely recognized as one of the most challenging components to manufacture.
[0003] The new trend of wind turbines becoming larger and offshore has led to increasingly larger bearing sizes and a sharp increase in the complexity of operating conditions. For traditional roller bearings, as the bearing size increases, on the one hand, the probability of localized bearing damage caused by bearing materials and external impurities will increase exponentially; on the other hand, the processing cost of large rolling bearings will also increase significantly, hindering the promotion and application of large wind turbines.
[0004] Compared to rolling bearings, well-designed sliding bearings typically offer better stability, greater tolerance to material defects and external impurities, and lower manufacturing costs. The superior operating performance and lower manufacturing cost of sliding bearings have attracted renowned domestic and international wind turbine manufacturer AREVA. Companies such as Vestas, Siemens, Repower, Goldwind Technology, Envision Energy, and Mingyang Smart Energy have been researching and adopting sliding bearings as bearings for large wind turbines, which has become a development trend for bearings in large offshore wind turbines.
[0005] Unlike conventional sliding bearings, wind turbine sliding bearings operate at low speeds (0–20 rpm), bear heavy loads (average specific pressure typically greater than 10 MPa), and experience frequent starts and stops. They also need to withstand time-varying bending moments. This makes them highly susceptible to localized and global contact between the bearing journal and the bearing bush, leading to mixed lubrication, boundary lubrication, or even dry friction. This damages the friction pair surfaces and generates wear particles, which further exacerbate wear and degradation on the bearing surfaces. Consequently, the vibration of the rotor supported by the wind turbine bearing intensifies, ultimately causing bearing failure. To ensure that wind turbine sliding bearings achieve a lifespan of at least 20 years, similar to the main turbine engine, full life-cycle health monitoring and fault prediction are necessary.
[0006] In recent years, wind turbine condition monitoring technologies have mainly focused on vibration, hydraulic fluid, acoustic emission, and temperature. Vibration sensors installed at both ends of the rotor bearing can monitor radial and axial vibration signals. During operation, when a wind turbine bearing experiences non-forced localized damage, periodic impacts occur at the damaged structural parts as the bearing rotates. These periodic impacts generate periodic frequency characteristics, which are reflected in the analyzed vibration signal spectrum. The characteristic frequency of bearing vibration is determined by a combination of factors, including its structural form, geometric dimensions, fit, operating speed, and defect location. Therefore, the location of bearing defects can be indirectly determined based on the fault frequencies analyzed from the monitoring data. By monitoring the vibration signals of the generator bearing during operation, data is collected using sensors and analyzed. A frequency domain response graph is obtained from a portion of the vibration data and compared with the spectrum under normal conditions. This comparison yields abnormal state frequencies in the spectrum that match the characteristic frequencies of various bearing structural parts. Further analysis of the amplitude response and time-domain vibration trends helps determine the location and extent of bearing defects. However, vibration monitoring can only identify the fault state of the bearing and is more sensitive to later-stage bearing failures, making it extremely difficult to provide early warnings of bearing failure.
[0007] Oil is the "blood" of a machine, containing a wealth of tribological information. Therefore, oil analysis methods based on tribological knowledge can also be used to assess bearing condition. Oil monitoring methods mainly include offline and online monitoring methods. Methods that utilize offline oil analysis—such as ferrography, spectroscopy, particle counting, and physicochemical data—to establish neural networks and DS evidence theory for comprehensive analysis have been attempted for various bearing condition monitoring applications. However, offline analysis suffers from long sampling cycles, high costs, and most importantly, significant lag in the results, failing to reflect the current operating status. Online oil monitoring technology primarily monitors the operating status of various components of a wind turbine by analyzing the composition and shape of lubricating oil metal particles, effectively assessing the severity and evolution of bearing wear in real time. Compared to vibration monitoring, oil monitoring can identify the initial wear state of wind turbine bearings.
[0008] Acoustic emission (AE) technology is used to locate acoustic emission sources, identify their types, and assess their severity. AE detection technology involves building an AE acquisition system, using AE sensors to collect fault acoustic emission signals, and then performing parameter or waveform analysis on the acquired signals to achieve non-destructive testing. AE technology determines the location of vibration by analyzing wave arrival time and triangulation. Analysis of vibration signal characteristic parameters can accurately identify the type of rubbing fault during the operation of sliding bearings. Compared to oil monitoring methods, AE technology can be used to monitor the location, quantity, and area of rubbing before bearing wear, and can be used to monitor the evolution of the bearing's condition before wear.
[0009] The above analysis shows that different condition monitoring technologies have varying degrees of sensitivity to the condition of wind turbine bearings at different stages. Acoustic emission technology is more sensitive to the evolution of early wear processes, oil monitoring excels at detecting wear evolution, and vibration monitoring is mainly used to identify later bearing failures. A single monitoring technology cannot achieve full coverage of the entire lifecycle health status and cannot provide comprehensive information on the operating status of wind turbine bearings. Therefore, this invention integrates lubrication film thickness detection technology, acoustic emission wear location detection technology, oil abrasive analysis technology, and vibration analysis technology. It achieves full coverage of key operating parameters of wind turbine sliding bearings from multiple dimensions, including lubrication status detection during normal bearing operation, the number and size of early wear locations, initial wear, and later vibration. Then, through a multi-parameter feature fusion method, it extracts characteristic parameters corresponding to the health status of the wind turbine sliding bearing and uses an expert intelligent diagnostic algorithm to predict the health status of the wind turbine sliding bearing. This method can achieve accurate assessment of the health status of wind turbine sliding bearings. Summary of the Invention
[0010] Based on the above analysis, the purpose of this invention is to provide a method for monitoring the health status of wind turbine sliding bearings throughout their entire lifecycle. From the perspective of bearing tribological evolution, this method selects four sensitive techniques for wind turbine sliding bearings: ultrasonic film thickness detection, acoustic emission rubbing location detection, oil abrasive particle detection, and vibration detection. These techniques are sensitive to the normal operation, early rubbing, initial wear, and later failure states of the wind turbine sliding bearing. The parameters obtained from these four techniques are fused together, and the health status of the wind turbine sliding bearing is assessed using an intelligent expert diagnostic system. To achieve the above objective, the technical solution of this invention is as follows:
[0011] This method for monitoring the health status of sliding bearings includes the following steps:
[0012] Step 1: Based on an in-depth analysis of the health status and tribological characteristics evolution of wind turbine sliding bearings, establish a wind turbine sliding bearing status characterization system based on four types of information: sliding bearing lubricating oil film thickness, contact position, wear particles in lubricating oil, and vibration signal. Establish an integrated information monitoring system, using oil film thickness, friction source, wear particles, and vibration source as measurable features corresponding to the above four types of information.
[0013] Step 2: Collect the lubricating film thickness at each measuring point of the sliding bearing using an ultrasonic sensor, and obtain the global distribution of the lubricating film thickness within the bearing solution domain through interpolation. The specific steps are as follows:
[0014] S1, Measurement point location determination: The theoretical film thickness distribution of the bearing is obtained by numerically solving the basic equation of lubrication of the sliding bearing using the finite difference method. Based on the theoretical distribution of the lubrication film thickness, the location of the distributed measurement points on the bearing back is determined: for areas where the film thickness is less than 1 / 5 of the theoretical maximum film thickness, the sensors are arranged closely at a spacing of about 10 mm; for areas where the film thickness is greater than 1 / 5 of the theoretical maximum film thickness, the sensors are arranged sparsely at a spacing of 20 to 40 mm.
[0015] S2, a circular thin-film ultrasonic sensor is embedded at each measuring point;
[0016] S3, the noise in the ultrasonic sensors at each measuring point is eliminated by a bandpass filter;
[0017] S4, Ultrasonic measurement of film thickness at measuring points: Depending on the range of film thickness, the lubricating film thickness at each measuring point is obtained using an ultrasonic film thickness measuring device by employing the ultrasonic time-of-flight method, ultrasonic thin film resonance method, and ultrasonic spring model method.
[0018] S5. After obtaining the lubricating film thickness at each measuring point, the global distribution of the lubricating film thickness in the bearing solution domain is obtained by using cubic spline interpolation based on the three-dimensional coordinate set composed of the position coordinates of each measuring point and the corresponding film thickness value.
[0019] Step 3: Collect acoustic emission sensor signals and analyze them to determine the location of early rubbing faults in the sliding bearing. The specific steps are as follows:
[0020] S1, Arrange the bearing array sensor unit, including multiple acoustic emission sensors evenly arranged in the circumferential direction of the bearing;
[0021] S2, after being filtered by a bandpass filter to eliminate noise in the acoustic emission sensor, is connected in sequence to a preamplifier, a data acquisition module and a data processing module. The output of the data processing module is connected to the display module. The acoustic emission sensor collects the collision and friction vibration signals of the detected bearing and inputs them to the data processing module.
[0022] S3, the data processing module analyzes the time-domain characteristics of the vibration signals collected by multiple acoustic emission sensors, calculates the position information of the bearing rubbing acoustic emission source, and outputs it through the display module to give the sliding bearing rubbing position information.
[0023] Step 4: Extract features from the acquired abrasive grain images, analyze and identify the type of abrasive grains, and predict the bearing wear condition. The specific steps are as follows:
[0024] S1, through the oil sampling circulation system, the CMOS digital imaging system acquires abrasive video over a certain period of time, and the computer obtains preliminary video images of the flow of lubricating oil containing abrasive particles.
[0025] S2, decompose the video image obtained in each sampling period into a single frame abrasive image and extract the abrasive images at the corresponding intervals and store them in the corresponding folders in order to extract the statistical features and single abrasive features of the abrasive.
[0026] S3, Statistical feature extraction of abrasive particles, specifically extracting the concentration index (IPCA) and color distribution of abrasive particles:
[0027] 1) Open the transmitted light image from the corresponding folder and decompose the image into high-frequency and low-frequency component images using fast binary wavelet transform;
[0028] 2) Extract the total perimeter C of the abrasive grains from the high-frequency component image using the chain code tracing method;
[0029] 3) Low-frequency component images are subjected to grayscale conversion, binarization, filtering, and morphological processing using digital image processing techniques;
[0030] 4) Mark the abrasive particles from the processed low-frequency component image using the grayscale space stacking method to obtain the number of abrasive particles N. At the same time, obtain the area A of the abrasive particles. Divide the area A of the abrasive particles by the total area S of the entire image to obtain the abrasive particle concentration index IPCA = A / S × 100%.
[0031] 5) Divide the deposition abrasive concentration index by the deposition time t to obtain the abrasive generation rate v, i.e.
[0032]
[0033] 6) Perform color segmentation on the reflected light image to remove the background color of the image, and statistically analyze the color of the abrasive particles after removing the background color based on the HSI color space to obtain a color histogram;
[0034] 7) Store the extracted statistical feature parameters in the database.
[0035] S4, extraction of single abrasive grain features, specifically:
[0036] 1) By segmenting the image using the background subtraction method, the moving abrasive particles can be separated, simplifying the target search process and enabling rapid localization of the abrasive particles;
[0037] 2) Use image morphological operation methods to segment abrasive particles and achieve corrected tracking: The image processing methods mainly include target segmentation, binarization, morphological erosion and dilation operations for corrected tracking of adhered abrasive particles.
[0038] 3) Based on the abrasive grain tracking results, feature matching can be used to obtain feature sequences of the same abrasive grain from different perspectives. Combined with pixel scanning methods, the basic contour information of the abrasive grain can be quickly obtained, including its major axis dimension (L), minor axis dimension (W), and area (A).
[0039] 4) Construct typical two-dimensional features of abrasive particles using basic contour information, including:
[0040] a. Aspect ratio (AR), used to identify cutting abrasive grains:
[0041] b. Equivalent circular diameter (ECD), used to characterize the particle size of irregular abrasive grains:
[0042] c. Roundness (R), characterizing the similarity between the shape of the abrasive grains and a circle:
[0043] 5) The minimum minor axis dimension Wmin of the abrasive grain can be obtained from the full-view image. This parameter can be considered as the approximate thickness (T) of the abrasive grain, i.e., T = Wmin. By utilizing the basic spatial features: length, width, and thickness, typical three-dimensional features can be further constructed, including:
[0044] a. The aspect ratio (HAR) can be used as a basis for distinguishing between cutting and layered abrasive grains:
[0045]
[0046] b. Width-to-thickness ratio (HWAR) can be used as a basis for identifying layered abrasive grains:
[0047]
[0048] c. Sphericity (S), characterizing the shape of the abrasive grains and their approximation to a sphere:
[0049] 6) Store the extracted single abrasive grain feature parameters in the database.
[0050] Step 5: By placing a piezoelectric accelerometer on the bearing housing, the sensor signal is collected, converted into digital vibration data by an A / D converter, and transmitted to the processor. The processor determines whether the bearing has failed based on the digital vibration data.
[0051] Step Six: Extract minimum oil film thickness, average oil film thickness, contact point, wear particles in lubricating oil, and vibration signal characteristics from four types of information: lubricating oil film thickness, contact point, maximum single contact area, total contact area, abrasive particle concentration, abrasive particle color, single abrasive particle shape, single abrasive particle size, vibration amplitude, and vibration frequency. Use principal component analysis to extract full life cycle health mode characteristics (including healthy, sub-healthy, faulty, and failed) to achieve the fusion of the four types of information characteristics.
[0052] Step 7: Based on the full life cycle health pattern feature data obtained above, the four unobservable states of health, sub-health, fault, and failure are used as the four hidden states of the Hidden Markov Chain. The parameter estimates of the most suitable Hidden Markov Model are obtained through the forward-backward algorithm. The continuous behavior of the health state is characterized by the probability density function to predict the health state of the wind turbine sliding bearing.
[0053] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:
[0054] This invention is applied to online monitoring of wind turbine sliding bearings during operation. Through coupled analysis, four detection technologies are selected: ultrasonic measurement of lubricating film thickness, acoustic emission detection of rubbing location, oil abrasive particle detection, and vibration analysis, which are sensitive to the normal operation, early rubbing, initial wear, and later faults of wind turbine sliding bearings, respectively. These four types of sensors obtain four types of information about the real-time operating status of the wind turbine sliding bearing. This information contains comprehensive data necessary for identifying different health states of the wind turbine sliding bearing. Principal component analysis is used to fuse the four types of information, and then an intelligent expert algorithm is applied to assess the health status of the wind turbine sliding bearing. This method enables accurate prediction of the health status of wind turbine sliding bearings. Attached Figure Description
[0055] Figure 1 A technical solution diagram of an online monitoring system for wind turbine sliding bearings based on multi-information fusion;
[0056] Figure 2 A diagram illustrating the coupling and evolution of tribological characteristics of wind turbine sliding bearings;
[0057] Figure 3 A schematic diagram of a wind turbine sliding bearing full life cycle health status monitoring system based on multi-information fusion. Detailed Implementation
[0058] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings.
[0059] Reference Figures 1 to 3 A method for online monitoring of the health status of wind turbine sliding bearings throughout their entire lifecycle includes the following steps:
[0060] Step 1: Based on an in-depth analysis of the health status and tribological characteristics evolution of wind turbine sliding bearings, establish a wind turbine sliding bearing status characterization system based on four types of information: sliding bearing lubricating oil film thickness, contact position, wear particles in lubricating oil, and vibration signal. Establish an integrated information monitoring system, using oil film thickness, friction source, wear particles, and vibration source as measurable features corresponding to the above four types of information.
[0061] Step 2: By collecting ultrasonic sensor signals installed on the front and rear sliding bearings 1 and 4 of the spindle, the lubricating film thickness at each measuring point of the sliding bearing is obtained through analysis and processing. The global distribution of the lubricating film thickness in the solution domain of the bearing is obtained by interpolation.
[0062] Step 3: By collecting signals from acoustic emission sensors installed on the front and rear sliding bearings 1 and 4 of the spindle, the location of early rubbing failure in the sliding bearings is analyzed and determined.
[0063] Step 4: Extract features from the abrasive images acquired by the lubricating oil abrasive image system, analyze and identify the abrasive type, and predict the bearing wear condition;
[0064] Step 5: By arranging a piezoelectric accelerometer on the bearing housing 2, the sensor signal is collected, converted into digital vibration data by an A / D converter, and transmitted to the processor. The processor determines whether the bearing has failed based on the digital vibration data.
[0065] Step Six: Extract minimum oil film thickness, average oil film thickness, contact point, wear particles in lubricating oil, and vibration signal characteristics from four types of information: lubricating oil film thickness, contact point, maximum single contact area, total contact area, abrasive particle concentration, abrasive particle color, single abrasive particle shape, single abrasive particle size, vibration amplitude, and vibration frequency. Use principal component analysis to extract full life cycle health mode characteristics (including healthy, sub-healthy, faulty, and failed) to achieve the fusion of the four types of information characteristics.
[0066] Step 7: Based on the full life cycle health pattern feature data obtained above, the four unobservable states of health, sub-health, fault, and failure are used as the four hidden states of the Hidden Markov Chain. The parameter estimates of the most suitable Hidden Markov Model are obtained through the forward-backward algorithm. The continuous behavior of the health state is characterized by the probability density function to predict the health state of the wind turbine sliding bearing.
[0067] This invention collects four types of information: lubricating oil film thickness, contact location, wear particles, and vibration source. These four types of information cover key information at each stage of the evolution of the tribological characteristics of wind turbine sliding bearings. Based on the coupling relationship of these four types of information, measurability analysis is performed, referring to… Figure 2 The specific reasons for selecting the above four types of information are as follows:
[0068] During actual operation, a hydrodynamic oil film forms between the journal and the bearing shell of a sliding bearing, effectively reducing surface wear and ensuring normal bearing operation. However, as the generator set operates, changes in the circumferential load on the sliding bearing may disrupt this oil film, causing it to thin and reducing its load-bearing capacity. This leads to dry friction between the journal and the bearing shell, and wear particles entering the lubricating oil, further reducing the oil film's load-bearing capacity. This reduced capacity directly exacerbates bearing friction and wear, resulting in unstable bearing operation and vibration.
[0069] Therefore, a measurability analysis was performed on the main characteristic components of the hydrodynamic oil film, early rubbing location, wear particles, and vibration. The measurable characteristic components were found to be: oil film thickness, rubbing location, wear particle characteristics, and vibration source. By comprehensively analyzing the characteristic information of the hydrodynamic oil film, early rubbing location, wear particles, and vibration, the real-time operating status of the wind turbine sliding bearing was evaluated, and the trend of subsequent operation was predicted based on the current operating condition. Based on this, pre-maintenance of the wind turbine sliding bearing was provided to improve the performance of the wind turbine and extend its service life.
Claims
1. A method for monitoring the health status of wind turbine sliding bearings throughout their entire lifecycle based on multi-information fusion, characterized in that, Includes the following steps: Step 1: Establish a wind turbine sliding bearing condition characterization system and integrated information monitoring system based on four types of information: lubricating oil film thickness, contact location, wear particles in lubricating oil, and vibration signals; wherein, the four types of information are used to characterize the normal operation state, early contact state, initial wear state, and later failure state of the wind turbine sliding bearing, respectively. Step 2: Obtain the theoretical film thickness distribution and determine the distributed measuring points on the bearing back using the finite difference method. Collect the lubricating film thickness value at each measuring point using an ultrasonic sensor, and obtain the global distribution of the lubricating film thickness through cubic spline interpolation. Collect collision and friction vibration signals using multiple acoustic emission sensors uniformly arranged around the bearing circumference, and calculate the location of the collision and friction acoustic emission source based on its time-domain characteristics. Collect abrasive video using an oil sampling circulation system and a CMOS digital imaging system, and extract abrasive statistical features and single abrasive features from single-frame abrasive images. Collect vibration signals using a piezoelectric accelerometer on the bearing housing and obtain vibration data through A / D analog-to-digital conversion. Step 3: Extract the minimum oil film thickness, average oil film thickness, rubbing location, single maximum rubbing area, total rubbing area, abrasive concentration, abrasive color, single abrasive shape, single abrasive size, vibration amplitude, and vibration frequency from the four types of information. Use principal component analysis to extract full life cycle health mode features including healthy, sub-healthy, faulty, and failed. Step 4: Using healthy, sub-healthy, faulty, and failed as the four hidden states of the Hidden Markov Chain, the parameter estimates of the Hidden Markov Model are obtained through the forward-backward algorithm, and the continuous behavior of the healthy state is characterized by the probability density function to predict the health state of the wind turbine sliding bearing.
2. The method for monitoring the health status of wind turbine sliding bearings throughout their entire life cycle based on multi-information fusion as described in claim 1, characterized in that, The methods for obtaining the four types of information are as follows: S1: For regions where the theoretical film thickness is less than 1 / 5 of the theoretical maximum film thickness, ultrasonic sensors are densely arranged at a spacing of about 10 mm. For regions where the theoretical film thickness is greater than 1 / 5 of the theoretical maximum film thickness, ultrasonic sensors are sparsely arranged at a spacing of 20-40 mm. Circular thin-film ultrasonic sensors are embedded at each measuring point. Bandpass filtering is performed on the sensor signals, and the lubricating film thickness at each measuring point is obtained using the ultrasonic time-of-flight method, ultrasonic thin-film resonance method, or ultrasonic spring model method, depending on the range of the film thickness. S2: Multiple acoustic emission sensors are evenly arranged around the bearing. After the sensor signals are bandpass filtered, preamplified and acquired, the location of the collision and friction acoustic emission source is calculated based on the time-domain characteristics of the collision and friction vibration signals. S3: Decompose the abrasive video of each sampling period into single-frame abrasive images, obtain the abrasive concentration index IPCA through wavelet transform, chain code tracing and image morphology processing, obtain the abrasive color distribution based on the HSI color space, and extract the shape and size features of a single abrasive through background subtraction and morphological processing. S4: Vibration signals are collected by a piezoelectric accelerometer mounted on the bearing housing, and vibration data is obtained through A / D analog-to-digital conversion and analyzed to determine the bearing fault condition.
Citation Information
Patent Citations
Interface friction property monitoring-based rolling-sliding friction life cycle test device
CN106198019A
Equipment health state prediction method and system in centralized monitoring system
CN114429316A
