A wind power gear box dynamic health assessment method and device, electronic equipment, storage medium and program product

Abstract

This invention relates to the field of wind power operation and maintenance technology, and discloses a method, device, electronic equipment, storage medium, and program product for dynamic health assessment of wind turbine gearboxes. This invention integrates multi-source heterogeneous operating data, constructs a gearbox proxy model based on a physical information neural network, outputs internal stress time series, and accurately calculates cumulative damage, overcoming the limitation of sensors not being able to directly measure the state of internal contact surfaces. A comprehensive health score is obtained through weighted fusion of multi-dimensional indicators such as vibration, fluid, and stress, significantly advancing the fault warning window compared to traditional vibration monitoring. The model has fast inference speed, adapts to real-time computation at the edge of wind farms, and accurately calculates remaining lifespan and provides graded warnings using a time series network. The assessment results conform to mechanical laws, reducing the probability of misjudgment under extreme operating conditions and enabling early preventative operation and maintenance of gearboxes.

Description

Technical Field

[0001] This invention relates to the field of wind power operation and maintenance technology, specifically to a method, device, electronic equipment, storage medium, and program product for dynamic health assessment of wind turbine gearboxes. Background Technology

[0002] The wind turbine gearbox is a core transmission component of a wind turbine generator set. Operating under complex environments with varying loads and conditions, its gears, bearings, and other core components are prone to fatigue damage, directly impacting the unit's operational stability and service life. To reduce the risk of downtime and lower maintenance costs, the industry needs to rely on monitoring and assessment technologies to accurately predict gearbox health status and implement preventative maintenance. Therefore, high-precision, real-time, and interpretable gearbox condition monitoring technology has become a key research focus in the wind power operation and maintenance field.

[0003] Currently, the industry mainly uses CMS vibration monitoring systems, pure data deep learning models, and high-precision dynamic simulation models for gearbox condition monitoring and assessment. However, existing technologies have many shortcomings: traditional CMS systems suffer from monitoring lag, only capturing impact signals in the later stages of component damage, and cannot reserve sufficient maintenance windows; high-precision dynamic and finite element models have high computational demands, making it difficult to adapt to the real-time computing power requirements at the edge of wind farms; pure data deep learning black-box models lack physical constraints, and prediction results under extreme working conditions and data-deficient scenarios violate mechanical laws, resulting in poor reliability and interpretability. Furthermore, the condition of core contact surfaces such as gear meshing surfaces and bearing raceways cannot be directly detected by sensors, and model input parameters rely on empirical estimations, significantly reducing the accuracy of damage assessment. Summary of the Invention

[0004] This invention provides a method, device, electronic equipment, storage medium, and program product for dynamic health assessment of wind turbine gearboxes, in order to solve the problem that existing technologies are unable to accurately monitor and assess the health status of gearboxes.

[0005] In a first aspect, the present invention provides a method for dynamic health assessment of wind turbine gearboxes, the method comprising: Acquire multi-source heterogeneous data of wind turbine generators and extract feature vectors from the multi-source heterogeneous data, wherein the feature vectors include oil temperature residual, high-frequency vibration characteristics, real-time speed and real-time torque; The oil temperature residual, the high-frequency vibration characteristics, the real-time speed, and the real-time torque are input into a preset gearbox model based on a physical information neural network, and the stress cycle sequence of the gearbox is output. Based on the stress cycle sequence of the gearbox, calculate the total cumulative damage of the gearbox; The oil wear index corresponding to the oil temperature residual, the vibration energy entropy corresponding to the high-frequency vibration characteristics, and the stress damage rate corresponding to the total cumulative damage are weighted and fused to generate a total health score. Based on the gearbox's total health score, the remaining usable lifespan is predicted, and corresponding early warning information is generated according to the remaining usable lifespan.

[0006] This invention integrates multi-source heterogeneous operational data and constructs a gearbox proxy model based on a physical information neural network. It outputs internal stress time series data and accurately calculates cumulative damage, overcoming the limitation of sensors not being able to directly measure the state of internal contact surfaces. By weighted fusion of multi-dimensional indicators such as vibration, fluid, and stress, a comprehensive health score is obtained, significantly advancing the fault warning window compared to traditional vibration monitoring. The model boasts fast inference speed, adapting to real-time computation at the edge of wind farms. Combined with a time series network, it accurately calculates remaining lifespan and provides tiered warnings. The evaluation results conform to mechanical laws, reducing the probability of misjudgment under extreme operating conditions and enabling early preventative maintenance of the gearbox.

[0007] In one optional implementation, the step of acquiring multi-source heterogeneous data of the wind turbine generator and extracting feature vectors from the multi-source heterogeneous data, wherein the feature vectors include oil temperature residual, high-frequency vibration characteristics, real-time speed, and real-time torque, includes: Collect monitoring data of wind turbine units and gearbox data of gearboxes to generate multi-source heterogeneous data; Feature vectors of the wind turbine monitoring data and gearbox data are extracted respectively to obtain the real-time torque and real-time speed of the wind turbine, and the high-frequency vibration acceleration, instantaneous speed, lubricating oil temperature and oil particle size of the input and output bearings of the gearbox. Extract the high-frequency vibration characteristics of the high-frequency vibration acceleration; Calculate the temperature residual between the real-time lubricating oil temperature and the preset oil temperature.

[0008] This invention simultaneously collects multi-dimensional heterogeneous data from unit monitoring and gearboxes, covering various information such as load, vibration, lubrication, and oil, comprehensively reflecting the equipment's operating status. By mining deep high-frequency features from vibration signals and quantifying lubrication anomalies through oil temperature residuals, it can capture early, minute losses that are difficult to identify using traditional single-source vibration monitoring. The complementary use of multi-source features avoids the lag problem of single-data monitoring, providing complete and detailed underlying input for subsequent physical modeling and health assessment, thus improving the ability to identify early faults.

[0009] In one optional implementation, the step of inputting the oil temperature residual, the high-frequency vibration characteristics, the real-time rotational speed, and the real-time torque into a preset gearbox model based on a physical information neural network, and outputting the gearbox stress cycle sequence, includes: The oil temperature residual, the high-frequency vibration characteristics, the real-time speed, and the real-time torque are input into a preset gearbox model based on a physical information neural network, and the contact stress and local slip ratio of the key gear meshing surfaces are output. Based on the contact stress and the real-time torque, a Hertzian contact constraint is constructed. Based on the real-time rotational speed, the oil temperature residual, and the local slip ratio, a lubrication balance constraint is constructed. Based on the contact stress, the true value of the contact stress, the local slip ratio, and the true value of the local slip ratio, a data-driven loss is constructed; The total loss is constructed using the Hertzian contact constraint, the lubrication balance constraint, and the data-driven loss. With the goal of minimizing the total loss, the model parameters of the gearbox model based on the physical information neural network are updated to generate the target gearbox model based on the physical information neural network. The oil temperature residual, the high-frequency vibration characteristics, the real-time speed, and the real-time torque are input into the target gearbox model based on the physical information neural network to generate the stress cycle sequence of the gearbox.

[0010] This invention embeds both Hertzian contact and elastohydrolubrication physical constraints into a neural network, and constructs a total loss optimization model by combining data loss. The constraint prediction results closely align with mechanical laws, solving the problem of physical distortion in pure data models. It can output internal contact stress and slip ratio without complex finite element simulation, and its inference efficiency is adapted to real-time edge computing. By leveraging virtual sensing, it obtains internal stress timing sequences that cannot be directly measured, providing accurate physical basis for fatigue damage calculation.

[0011] In one optional implementation, calculating the total cumulative damage of the gearbox based on the stress cycle sequence of the gearbox includes: Based on the stress cycle sequence of the gearbox, combined with stress and life curves, the percentage of cycle damage at each stress level is calculated. The total cumulative damage of the gearbox is calculated based on the percentage of cyclic damage under each stress level.

[0012] This invention utilizes the continuous stress cycle sequence output by PINN, combined with the material's SN curve, to quantify the proportion of fatigue damage in each stress interval using Miner's theory, and then sums these values ​​to obtain the total cumulative damage. This indicator intuitively characterizes the fatigue loss of the meshing surface inside the gear, overcoming the limitation of traditional monitoring methods that cannot quantify internal damage. It provides reliable underlying physical parameters for the fusion of multi-dimensional health indicators, effectively improving the objectivity and interpretability of health assessments.

[0013] In one optional implementation, the step of weightedly fusing the oil wear index corresponding to the oil temperature residual, the vibration energy entropy corresponding to the high-frequency vibration characteristics, and the stress damage rate corresponding to the total cumulative damage to generate a total health score includes: The oil temperature residual is normalized to obtain the oil wear index; The high-frequency vibration characteristics are normalized to obtain the vibration energy entropy; The total cumulative damage is normalized to obtain the stress damage rate. The oil wear index, the vibration energy entropy, and the stress damage rate are weighted and fused to generate a total health score.

[0014] This invention normalizes three different dimensional indicators—oil, vibration, and stress—to a unified range, eliminating interference from differences in magnitude, and then weights and fuses them to obtain a comprehensive health score. It also considers multi-dimensional information such as lubrication status, abnormal vibration, and internal fatigue damage, overcoming the limitations of a single monitoring dimension in assessment. The health score combines data representation with mechanical damage evidence, resulting in a more comprehensive and reliable evaluation.

[0015] In one optional implementation, the step of predicting the remaining usable lifespan based on the gearbox's total health score and generating corresponding early warning information according to the remaining usable lifespan includes: Based on the total health score of the gearbox and the real-time wind conditions corresponding to the wind turbine, the predicted speed and predicted torque are obtained. The predicted speed and the predicted torque are input into a preset long short-term memory network, and a predicted health score is output. Determine whether the predicted health score is less than a first preset health threshold and greater than a second preset health threshold; If so, it is determined that the first remaining usable lifespan has been reached, and a corresponding warning message is generated; If not, then determine whether the predicted health score is less than the second preset health threshold; If so, it is determined that the second remaining usable lifespan has been reached, and a corresponding warning message is generated.

[0016] This invention combines real-time wind conditions to predict future operating conditions, uses an LSTM time-series network to predict health scores, and divides the remaining lifespan into intervals based on two levels of health thresholds, providing tiered early warnings. It can predict gearbox damage trends in advance, differentiate warnings based on varying degrees of damage, and prevent sudden downtime due to faults; it also enables quantitative lifespan prediction, providing ample time windows for maintenance planning and enhancing the proactiveness of unit maintenance.

[0017] Secondly, the present invention provides a dynamic health assessment device for wind turbine gearboxes, the device comprising: The acquisition module is used to acquire multi-source heterogeneous data of the wind turbine and extract the feature vector of the multi-source heterogeneous data, wherein the feature vector includes oil temperature residual, high-frequency vibration characteristics, real-time speed and real-time torque; The output module is used to input the oil temperature residual, the high-frequency vibration characteristics, the real-time speed and the real-time torque into a preset gearbox model based on a physical information neural network, and output the stress cycle sequence of the gearbox. A calculation module is used to calculate the total cumulative damage of the gearbox based on the stress cycle sequence of the gearbox; The preprocessing module is used to perform weighted fusion processing on the oil wear index corresponding to the oil temperature residual, the vibration energy entropy corresponding to the high frequency vibration characteristics, and the stress damage rate corresponding to the total cumulative damage degree to generate a total health score. The prediction module is used to predict the remaining usable lifespan based on the total health score of the gearbox, and generate corresponding early warning information according to the remaining usable lifespan.

[0018] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the wind turbine gearbox dynamic health assessment method described in the first aspect or any corresponding embodiment thereof.

[0019] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the wind turbine gearbox dynamic health assessment method of the first aspect or any corresponding embodiment thereof.

[0020] Fifthly, the present invention provides a computer program product, including computer instructions for causing a computer to execute the wind turbine gearbox dynamic health assessment method described in the first aspect or any corresponding embodiment thereof. Attached Figure Description

[0021] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the first step of the dynamic health assessment method for wind turbine gearboxes according to an embodiment of the present invention; Figure 2This is a schematic diagram of the second process of the dynamic health assessment method for wind turbine gearboxes according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the process for calculating temperature residuals using an encoder according to an embodiment of the present invention; Figure 4 This is a structural block diagram of a wind turbine gearbox dynamic health assessment device according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0025] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0026] This invention provides a dynamic health assessment method for wind turbine gearboxes. By fusing multi-source heterogeneous operating data and constructing a gearbox proxy model based on a physical information neural network, it outputs internal stress time series and accurately calculates cumulative damage, overcoming the limitation of sensors not being able to directly measure the state of internal contact surfaces. A comprehensive health score is obtained through weighted fusion of multi-dimensional indicators such as vibration, fluid, and stress, significantly advancing the fault warning window compared to traditional vibration monitoring. The model has fast inference speed, adapting to real-time computation at the edge of wind farms. Combined with a time series network, it accurately calculates the remaining lifespan and provides graded warnings. The assessment results conform to mechanical laws, reducing the probability of misjudgment under extreme operating conditions and enabling early preventative maintenance of the gearbox.

[0027] This embodiment provides a method for dynamic health assessment of wind turbine gearboxes. Figure 1 This is a flowchart of a dynamic health assessment method for wind turbine gearboxes according to an embodiment of the present invention, such as... Figure 1As shown, the process includes the following steps: Step S101: Obtain multi-source heterogeneous data of the wind turbine and extract the feature vector of the multi-source heterogeneous data. The feature vector includes oil temperature residual, high-frequency vibration characteristics, real-time speed and real-time torque.

[0028] It should be noted that multi-source heterogeneous data refers to monitoring data from different sources and with inconsistent dimensions and formats, such as wind turbine SCADA operation data, gearbox vibration, lubricating oil, and oil detection data, which are collected synchronously. Feature vector refers to the standardized feature set obtained after preprocessing the original monitoring data, which includes four types of input indicators: oil temperature residual, high-frequency vibration characteristics, real-time speed, and real-time torque, and serves as input for model calculation. Oil temperature residual refers to the difference between the predicted lubricating oil temperature and the real-time measured oil temperature under normal operating conditions of the unit, used to characterize early anomalies in the lubrication system. High-frequency vibration characteristics refer to the frequency domain and time domain characteristics extracted from the high-frequency vibration acceleration signal of the gearbox bearing, reflecting the minor wear and impact of gears and bearings. Real-time speed refers to the shaft rotation rate collected in real time during the operation of the wind turbine gearbox, which is a core load condition parameter of the unit. Real-time torque refers to the real-time output torque of the wind turbine main shaft, which determines the magnitude of the gear meshing contact load and is a core input quantity for mechanical calculation.

[0029] In this embodiment of the invention, the collected SCADA monitoring data of the wind turbine unit, along with gearbox vibration and lubricating oil monitoring data, constitute multi-source heterogeneous data. After preprocessing the raw data, a feature vector of uniform specification is extracted. This vector integrates the oil temperature residual obtained from the difference between the measured temperature of the lubricating oil and the predicted temperature of the model, the high-frequency vibration characteristics obtained from the analysis of the high-frequency vibration signal of the bearing, and the real-time speed and real-time torque collected during the real-time operation of the unit, thus completely preserving multi-dimensional operating information on lubrication, vibration, and load.

[0030] Step S102: Input the oil temperature residual, high-frequency vibration characteristics, real-time speed and real-time torque into the preset gearbox model based on physical information neural network, and output the stress cycle sequence of the gearbox.

[0031] It should be noted that the gearbox model based on physical information neural network (PINN model) refers to a neural network surrogate model that embeds the constraints of Hertzian contact and lubrication physical equations, which can solve for the unmeasurable mechanical parameters inside the gear from external observation data; the stress cycle sequence refers to the time-series gear meshing contact stress output by the PINN model in real time, which characterizes the fatigue load borne by the gear at different times.

[0032] In this embodiment of the invention, oil temperature residual, high-frequency vibration characteristics, real-time speed and real-time torque are input into the pre-trained physical information neural network gearbox model. Based on the Hertzian contact constraint and lubrication balance constraint built into the model, forward deduction is carried out, and a continuous stress cycle sequence of gear meshing position changing with time is output, providing core mechanical time series data for subsequent cumulative damage calculation.

[0033] Step S103: Calculate the total cumulative damage of the gearbox based on the stress cycle sequence of the gearbox.

[0034] It should be noted that the total cumulative damage refers to the total amount of cumulative fatigue damage of the gear calculated from the time-series stress based on Palmgren-Miner theory and the material SN curve, reflecting the degree of component wear.

[0035] In this embodiment of the invention, based on the gearbox time-series stress cycle sequence and combined with the gear material SN stress-life curve, the number of fatigue failure cycles corresponding to different stress amplitudes is obtained. Based on the Palmgren-Miner linear fatigue damage theory, the fatigue damage ratio of each stress level is accumulated and calculated one by one, and finally the total cumulative damage degree that can truly characterize the current fatigue wear state of the gearbox is obtained.

[0036] Step S104: The oil wear index corresponding to the oil temperature residual, the vibration energy entropy corresponding to the high-frequency vibration characteristics, and the stress damage rate corresponding to the total cumulative damage are weighted and fused to generate a total health score.

[0037] It should be noted that the oil wear index is a quantitative indicator obtained by normalizing the oil temperature residual to the 0-1 range, which characterizes the degree of lubricant deterioration and lubrication failure; the vibration energy entropy is an indicator obtained by normalizing the high-frequency vibration characteristics through Hilbert-Huang transform; the stress damage rate is an indicator of the total cumulative damage normalized to the 0-1 range, which directly reflects the level of fatigue wear inside the gear; the total health score is a comprehensive index in the 0-1 range obtained by weighted fusion of the oil wear index, vibration energy entropy, and stress damage rate, and the lower the value, the more serious the equipment damage.

[0038] In this embodiment of the invention, the oil temperature residual is normalized to obtain the oil wear index, the high-frequency vibration characteristics are transformed by energy entropy and normalized to obtain the vibration energy entropy, the total cumulative damage degree is mapped to the range of 0 to 1 to obtain the stress damage rate, and then the three sub-indicators are weighted and summed according to preset weights to generate a total health score with a value range between 0 and 1 that comprehensively reflects the overall operating status of the gearbox.

[0039] Step S105: Based on the total health score of the gearbox, predict the remaining usable lifespan and generate corresponding early warning information according to the remaining usable lifespan.

[0040] It should be noted that the remaining usable lifespan refers to the remaining operating time of the gearbox before it reaches the failure threshold, which is calculated by combining the future health degradation trend predicted by the time series network; the early warning information refers to the graded operation and maintenance prompts generated based on the remaining lifespan level, including control instructions such as power limiting, lubrication inspection, and shutdown maintenance.

[0041] In this embodiment of the invention, the total health score of the gearbox is input into the trained long short-term memory network to deduce the decay trend of the gearbox's health status over time, accurately predict the remaining usable life of the gearbox, and then determine the equipment operation risk level based on the preset multi-level life threshold, and generate corresponding operation and maintenance early warning information to realize early prediction and graded early warning of gearbox failure.

[0042] This embodiment provides a method for dynamic health assessment of wind turbine gearboxes. Figure 2 This is a flowchart of a dynamic health assessment method for wind turbine gearboxes according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps: Step S201: Obtain multi-source heterogeneous data of the wind turbine and extract the feature vector of the multi-source heterogeneous data. The feature vector includes oil temperature residual, high-frequency vibration characteristics, real-time speed and real-time torque.

[0043] Specifically, step S201 includes: Step S2011: Collect wind turbine monitoring data and gearbox data of the wind turbine generator set and generate multi-source heterogeneous data.

[0044] In this embodiment of the invention, SCADA data of the wind turbine (i.e., wind turbine monitoring data) and high-frequency signal data of the gearbox (i.e., gearbox data) are collected to obtain multi-source heterogeneous data.

[0045] Step S2012: Extract feature vectors from the wind turbine monitoring data and gearbox data respectively to obtain the real-time torque and real-time speed of the wind turbine, and the high-frequency vibration acceleration, instantaneous speed, lubricating oil temperature and oil particle size of the input and output bearings of the gearbox.

[0046] In this embodiment of the invention, feature vectors of wind turbine monitoring data are extracted, such as operating condition feature mapping: real-time torque. T Real-time rotation speed n .

[0047] Extract the feature vectors of the gearbox data: high-frequency vibration acceleration, instantaneous speed, lubricating oil temperature, and oil particle size of the gearbox input / output bearings.

[0048] Step S2013: Extract the high-frequency vibration characteristics of the high-frequency vibration acceleration.

[0049] In this embodiment of the invention, the original high-frequency vibration acceleration signal of the gearbox is subjected to noise reduction preprocessing, and high-frequency vibration features that can characterize the gearbox micro-wear and meshing impact abnormalities are extracted through time-domain and frequency-domain analysis. Invalid noise interference is eliminated, and sensitive feature information that can reflect the early failure of the gearbox is retained, providing accurate vibration data support for subsequent model input and health status assessment.

[0050] Step S2014: Calculate the temperature residual between the real-time lubricating oil temperature and the preset oil temperature.

[0051] In this embodiment of the invention, dynamic residual calculation involves using an autoencoder to model the temperature rise-load characteristics under normal operating conditions and calculating the temperature residual between the current real-time oil temperature and the predicted oil temperature. This serves as the first layer of triggering logic for health alerts.

[0052] Construct an autoencoder model to calculate the residuals: Constructing temperature rise residuals using an autoencoder (AE) is an excellent method for identifying "early anomalies." Its core logic is to learn the nonlinear correlation under normal conditions and quantify the deviation under abnormal conditions. The key to the model is that both the input and prediction data are oil temperature; the residuals from oil temperature predictions are used to determine whether the state is healthy or abnormal.

[0053] like Figure 3 As shown, the formula for calculating the residual is:

[0054] In the formula, Real oil temperature ; To predict oil temperature .

[0055] Step S202: Input the oil temperature residual, high-frequency vibration characteristics, real-time speed and real-time torque into the preset gearbox model based on physical information neural network, and output the stress cycle sequence of the gearbox.

[0056] Specifically, step S202 includes: Step S2021: Input the oil temperature residual, high-frequency vibration characteristics, real-time speed and real-time torque into the preset gearbox model based on physical information neural network, and output the contact stress and local slip ratio of the key gear meshing surface.

[0057] In this embodiment of the invention, the network structure design of the gearbox load-fatigue surrogate model based on physical information neural network (i.e., the gearbox model based on physical information neural network) is as follows: The input is... The output is the contact stress of the key gear meshing surface. and local slip ratio .

[0058] Step S2022: Construct Hertzian contact constraints based on contact stress and real-time torque.

[0059] In this embodiment of the invention, Hertz contact constraint: This constraint is used to verify the contact stress output by the neural network. With the real-time torque of the input T The macroscopic mechanical mapping relationship between them.

[0060] According to Hertzian contact theory, two cylinders (simulating gear meshing) experience normal forces... The formula for the maximum contact stress under action is:

[0061] In the formula, (From input torque) Base circle radius and pressure angle (Calculated). Equivalent elastic modulus The equivalent radius of curvature, This is the length of the contact wire.

[0062] Loss function expression:

[0063] Step S2023: Based on real-time rotational speed, oil temperature residual, and local slip ratio, a lubrication balance constraint is constructed.

[0064] In this embodiment of the invention, the lubrication balance constraint is: This constraint incorporates real-time rotational speed into the input features. and oil temperature The physical logic of stress, slip ratio and lubrication state is verified through elastohydrodynamic (EHL) theory.

[0065] Physical equations: The physical values ​​are calculated using the Dowson-Higginson minimum oil film thickness equation, and residuals are calculated by comparing them with the network predictions. The core lies in dynamic viscosity. With input oil temperature Exponential relationship:

[0066]

[0067] In the formula, For material parameters, For speed parameters (and input real-time rotational speed) (positive correlation) For load parameters (and output stress) (Related).

[0068] Loss function expression: stress With slip ratio In combination, a viscoplastic friction coefficient constraint is introduced:

[0069] Step S2024: Based on contact stress, the true value of contact stress, local slip ratio, and the true value of local slip ratio, construct a data-driven loss.

[0070] In this embodiment of the invention, the physical constraint loss is embedded (i.e., data-driven constraints): Physical loss item The aim is to force the output of a neural network to conform to the laws of mechanics and thermodynamics under any operating condition through residual penalty. Its overall expression is:

[0071] In the formula, These are weighting coefficients used to balance physical residuals of different dimensions.

[0072] Step S2025: Construct the total loss using Hertzian contact constraints, lubrication balance constraints, and data-driven losses.

[0073] In this embodiment of the invention, the total loss is: .

[0074] Step S2026: With the goal of minimizing the total loss, update the model parameters of the gearbox model based on the physical information neural network to generate the target gearbox model based on the physical information neural network.

[0075] In this embodiment of the invention, the optimization objective is to minimize the total loss consisting of data-driven loss, Hertzian contact mechanics loss, and elastohydrolubrication loss. The weights and bias parameters of the physical information neural network gearbox model are continuously updated iteratively through the gradient backpropagation algorithm to continuously correct the model prediction bias until the total loss converges and stabilizes. Finally, the model optimization iteration is completed, and a target gearbox model with both accuracy and physical compliance is obtained.

[0076] Step S2027: Input the oil temperature residual, high-frequency vibration characteristics, real-time speed and real-time torque into the target gearbox model based on the physical information neural network to generate the stress cycle sequence of the gearbox.

[0077] In this embodiment of the invention, the target gearbox model of the physical information neural network is optimized by inputting oil temperature residual, high frequency vibration characteristics, real-time speed and real-time torque together. Based on the dual physical constraint rules built into the model, forward reasoning calculation is performed to stably output a continuous stress cycle sequence that fits the actual meshing condition of the gearbox, providing reliable temporal mechanical data support for the accurate calculation of the fatigue cumulative damage of the gearbox.

[0078] Step S203: Calculate the total cumulative damage of the gearbox based on the stress cycle sequence of the gearbox.

[0079] Specifically, step S203 includes: Step S2031: Based on the stress cycle sequence of the gearbox, and combined with the stress and life curves, calculate the proportion of cycle damage at each stress level.

[0080] In this embodiment of the invention, the stress cycle sequence output by the PINN model is substituted into the Palmgren-Miner equation, which is a fundamental method in the engineering field for predicting the fatigue life of materials or structures under variable amplitude loading.

[0081] The core formula of the Palmgren-Miner equation is:

[0082] In the formula, D The total cumulative damage coefficient (when D≥1, the component is theoretically considered to have failed). To achieve a specific stress level Below, the actual number of cycles the component undergoes (obtained in real time from the PINN model output); To maintain a constant stress level The total number of cycles required for a component to fail is determined by the material's SN curve, i.e., the Wöhler curve.

[0083] Step S2032: Calculate the total cumulative damage of the gearbox based on the proportion of cyclic damage under each stress level.

[0084] In this embodiment of the invention, the percentage of cyclic damage corresponding to each stress level obtained from the stress cycle sequence and the material SN curve is summed according to the Palmgren-Miner damage theory. The result is the total cumulative damage degree of the gearbox, which intuitively reflects the overall fatigue loss degree of the gear under long-term variable load, and provides basic parameters for subsequent stress damage rate conversion.

[0085] Step S204: The oil wear index corresponding to the oil temperature residual, the vibration energy entropy corresponding to the high-frequency vibration characteristics, and the stress damage rate corresponding to the total cumulative damage are weighted and fused to generate a total health score.

[0086] Specifically, step S204 includes: Step S2041: Normalize the oil temperature residual to obtain the oil wear index.

[0087] In this embodiment of the invention, the oil wear index is obtained by mapping the oil temperature residual obtained from the autoencoder network to [0,1]. The larger the difference The lower.

[0088] Step S2042: Normalize the high-frequency vibration characteristics to obtain the vibration energy entropy.

[0089] In this embodiment of the invention, the vibration energy entropy is calculated by decomposing the high-frequency vibration characteristics in the frequency domain (Hilbert-Huang transform), calculating the energy distribution entropy, and mapping it to [0,1]. The smaller the entropy, The lower.

[0090] Step S2043: Normalize the total cumulative damage to obtain the stress damage rate.

[0091] In this embodiment of the invention, the PINN stress damage rate is calculated by substituting the contact stress cycle sequence output in real time from the PINN model into the Palmgren-Miner linear cumulative damage theory. Based on the material's SN curve, the cumulative damage degree within the current time period is calculated. Mapped to [0,1], we get The higher the degree of damage, the better. The lower.

[0092] Step S2044: The oil wear index, vibration energy entropy and stress damage rate are weighted and fused to generate a total health score.

[0093] In this embodiment of the invention, health indicators Fusion: By weighted fusion of vibration energy entropy, oil wear index, and PINN stress damage rate, a health score between 0 and 1 is generated. Specifically, the three health scores mentioned above are calculated into a total health score using a weighted average method. Initially, the same weight can be used for each of the three health scores.

[0094] Step S205: Based on the total health score of the gearbox, predict the remaining usable lifespan and generate corresponding early warning information according to the remaining usable lifespan.

[0095] Specifically, step S205 includes: Step S2051: Based on the total health score of the gearbox and the real-time wind conditions corresponding to the wind turbine, the predicted speed and predicted torque are obtained.

[0096] In this embodiment of the invention, RUL prediction is based on the current damage rate and combined with future predicted operating conditions, using a Long Short-Term Memory (LSTM) network to extrapolate the remaining usable life of the unit under different maintenance strategies.

[0097] Data Acquisition: Historical Health Score Wind farms predict wind conditions (meteorological data, etc.) to construct future operating conditions (predicted rotational speed). n Predicted torque T This means that by using the real-time total health score and the real-time wind conditions corresponding to the wind turbine, a predicted future operating condition (predicted rotational speed) can be constructed. n Predicted torque T ).

[0098] Step S2052: Input the predicted speed and predicted torque into a preset long short-term memory network and output the predicted health score.

[0099] In this embodiment of the invention, an LSTM-based network is constructed. The input data is historical operating condition data, and the output data is historical health scores. The predicted future operating conditions are input into the LSTM network to obtain the prediction. Health score.

[0100] Step S2053: Determine whether the predicted health score is less than the first preset health threshold and greater than the second preset health threshold.

[0101] It should be noted that the first preset health threshold is 0.6; the second preset health threshold is 0.2.

[0102] In this embodiment of the invention, the following settings are provided: Thresholds, for example, the first preset health threshold is 0.6; the second preset health threshold is 0.2.

[0103] Predict The health score is compared with two preset health thresholds to determine the prediction. Whether the health score is within the range of the second preset health threshold and the first preset health threshold is used to define the current degree of health degradation of the gearbox, distinguish the operating state of mild damage, and provide a basis for judgment on the remaining life of the corresponding level and the generation of early warning.

[0104] Step S2054: If yes, then it is determined that the first remaining usable lifespan has been reached, and corresponding warning information is generated.

[0105] It should be noted that the first remaining usable lifespan is set according to the actual situation and is not limited here.

[0106] In this embodiment of the invention, a tiered early warning system is implemented: when 0.2 < When the stress is less than 0.6 or the predicted stress of PINN continues to exceed the limit, the "lubrication enhancement" or "power-limited operation" command is triggered.

[0107] Model update: The results of each gearbox opening inspection or endoscopic examination (measured damage) will serve as a feedback signal to automatically fine-tune the physical weight parameters of the PINN model.

[0108] Step S2055: If not, determine whether the predicted health score is less than the second preset health threshold.

[0109] In this embodiment of the invention, if prediction If the health score is not within the range between the two health thresholds, then further prediction will be made. The health score is compared with 0.2 to determine the prediction. Whether the health score is below 0.2 is used to determine whether the gearbox has suffered severe fatigue damage or serious deterioration of its health status, providing a basis for issuing high-level warnings and determining the limit of remaining usable life.

[0110] Step S2056: If yes, then it is determined that the second remaining usable lifespan has been reached, and corresponding warning information is generated.

[0111] It should be noted that the second remaining usable lifespan is set according to the actual situation and is not limited here.

[0112] In this embodiment of the invention, when prediction When the health score falls below 0.2, the predicted lifespan is reached, and a corresponding warning message is generated.

[0113] The present invention has the following advantages: High accuracy and high robustness: Because physical constraints such as the Hertzian contact equation are embedded in the loss function of the neural network, the model can still output results that conform to the laws of mechanics under small samples or variable working conditions, which greatly reduces the false alarm rate and false alarm rate.

[0114] The invention achieves "virtual sensing": it uses algorithms to map externally measurable vibrations, rotational speeds, and torques into internally unmeasurable contact stresses and slip ratios, providing the most core underlying physical indicators for health assessment.

[0115] Significantly improves early warning time: By combining temperature residual (early evolution) and stress damage accumulation (mid-term development), it can increase the early warning time window for gearbox overhaul by more than 20% compared to traditional vibration monitoring.

[0116] Real-time online life assessment: PINN, as a surrogate model, has an inference speed of milliseconds after training, avoiding complex dynamic simulation calculations, and enabling the unit to have the ability to self-test its real-time remaining life (RUL) online.

[0117] Closed-loop self-optimization: By introducing a maintenance feedback mechanism and using measured damage to correct model weights, the problem of fixed physical model parameters that cannot be dynamically updated as components age is solved.

[0118] This embodiment also provides a wind turbine gearbox dynamic health assessment device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0119] This embodiment provides a dynamic health assessment device for wind turbine gearboxes, such as... Figure 4 As shown, it includes: The acquisition module 301 is used to acquire multi-source heterogeneous data of the wind turbine and extract feature vectors from the multi-source heterogeneous data. The feature vectors include oil temperature residual, high-frequency vibration characteristics, real-time speed and real-time torque. Output module 302 is used to input oil temperature residual, high frequency vibration characteristics, real-time speed and real-time torque into a preset gearbox model based on physical information neural network, and output the stress cycle sequence of the gearbox. Calculation module 303 is used to calculate the total cumulative damage of the gearbox based on the stress cycle sequence of the gearbox; The preprocessing module 304 is used to perform weighted fusion processing on the oil wear index corresponding to the oil temperature residual, the vibration energy entropy corresponding to the high frequency vibration characteristics, and the stress damage rate corresponding to the total cumulative damage degree to generate a total health score. The prediction module 305 is used to predict the remaining usable life based on the gearbox's total health score, and generate corresponding early warning information according to the remaining usable life.

[0120] In some optional implementations, the acquisition module 301 includes: The data acquisition unit is used to collect monitoring data of the wind turbine generator and gearbox data of the gearbox, and generate multi-source heterogeneous data. The first extraction unit is used to extract feature vectors from the wind turbine monitoring data and gearbox data respectively, so as to obtain the real-time torque and real-time speed of the wind turbine, and the high-frequency vibration acceleration, instantaneous speed, lubricating oil temperature and oil particle size of the input and output bearings of the gearbox. The second extraction unit is used to extract the high-frequency vibration characteristics of high-frequency vibration acceleration; The calculation unit is used to calculate the temperature residual between the real-time lubricating oil temperature and the preset oil temperature.

[0121] In some alternative implementations, the output module 302 includes: The input unit is used to input oil temperature residual, high-frequency vibration characteristics, real-time speed and real-time torque into a preset gearbox model based on physical information neural network, and output contact stress and local slip ratio. The first building block is used to construct Hertzian contact constraints based on contact stress and real-time torque; The second building unit is used to build lubrication balance constraints based on real-time speed, oil temperature residual and local slip ratio; The third building unit is used to construct data-driven loss based on contact stress, the true value of contact stress, local slip ratio, and the true value of local slip ratio; The fourth building block is used to construct the total loss by employing Hertzian contact constraints, lubrication balance constraints, and data-driven losses. The update unit is used to update the model parameters of the gearbox model based on the physical information neural network with the goal of minimizing the total loss, and to generate the target gearbox model based on the physical information neural network. The circulation unit is used to input oil temperature residual, high-frequency vibration characteristics, real-time speed and real-time torque into the target gearbox model based on physical information neural network to generate the stress circulation sequence of the gearbox.

[0122] In some alternative implementations, the computing module 303 includes: Combined with the unit, it is used for stress cycle sequence based on gearbox, and combined with stress and life curves to calculate the proportion of cycle damage at each stress level; The damage calculation unit is used to calculate the total cumulative damage of the gearbox based on the proportion of cyclic damage at each stress level.

[0123] In some alternative implementations, the preprocessing module 304 includes: The first processing unit is used to normalize the oil temperature residual to obtain the oil wear index. The second processing unit is used to normalize the high-frequency vibration characteristics to obtain the vibration energy entropy. The third processing unit is used to normalize the total cumulative damage to obtain the stress damage rate. The weighted fusion processing unit is used to perform weighted fusion processing on the oil wear index, vibration energy entropy and stress damage rate to generate a total health score.

[0124] In some alternative implementations, the prediction module 305 includes: The prediction unit is used to obtain the predicted speed and predicted torque based on the total health score of the gearbox and the real-time wind conditions corresponding to the wind turbine. The predictive health score unit is used to input the predicted speed and predicted torque into a preset long short-term memory network and output the predicted health score. The first judgment unit is used to determine whether the predicted health score is less than the first preset health threshold and greater than the second preset health threshold. The first early warning unit is used to determine that the first remaining usable lifespan has been reached if the condition is met, and to generate corresponding early warning information. The second judgment unit is used to determine whether the predicted health score is less than the second preset health threshold if no. The second early warning unit is used to determine if the second remaining usable lifespan has been reached and generate corresponding early warning information.

[0125] The wind turbine gearbox dynamic health assessment device provided in this embodiment of the invention can execute the wind turbine gearbox dynamic health assessment method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the above modules and units are the same as in the corresponding embodiments described above, and will not be repeated here.

[0126] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0127] The following is a detailed reference. Figure 5 This diagram illustrates a structural schematic suitable for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 401, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 402 or a program loaded from memory 408 into random access memory (RAM) 403. The RAM 403 also stores various programs and data required for the operation of the electronic device. The processor 401, ROM 402, and RAM 403 are interconnected via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0128] Typically, the following devices can be connected to I / O interface 405: input devices 406 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 407 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 408 including, for example, magnetic tapes, hard disks, etc.; and communication devices 409. Communication device 409 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 5 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0129] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 409, or installed from a memory 408, or installed from a ROM 402. When the computer program is executed by the processor 401, it performs the functions defined in the wind turbine gearbox dynamic health assessment method of the embodiments of the present invention.

[0130] Figure 5 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0131] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the wind turbine gearbox dynamic health assessment method shown in the above embodiments is implemented.

[0132] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0133] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for dynamic health assessment of wind turbine gearboxes, characterized in that, The method includes: Acquire multi-source heterogeneous data of wind turbine generators and extract feature vectors from the multi-source heterogeneous data, wherein the feature vectors include oil temperature residual, high-frequency vibration characteristics, real-time speed and real-time torque; The oil temperature residual, the high-frequency vibration characteristics, the real-time speed and the real-time torque are input into a preset gearbox model based on a physical information neural network, and the stress cycle sequence of the gearbox is output. Based on the stress cycle sequence of the gearbox, calculate the total cumulative damage of the gearbox; The oil wear index corresponding to the oil temperature residual, the vibration energy entropy corresponding to the high-frequency vibration characteristics, and the stress damage rate corresponding to the total cumulative damage are weighted and fused to generate a total health score. Based on the gearbox's total health score, the remaining usable lifespan is predicted, and corresponding early warning information is generated according to the remaining usable lifespan.

2. The method according to claim 1, characterized in that, The process involves acquiring multi-source heterogeneous data from wind turbine generators and extracting feature vectors from this data. These feature vectors include oil temperature residuals, high-frequency vibration characteristics, real-time speed, and real-time torque. Collect monitoring data from wind turbine generators and gearbox data from gearboxes to generate multi-source heterogeneous data; Feature vectors of the wind turbine monitoring data and gearbox data are extracted respectively to obtain the real-time torque and real-time speed of the wind turbine, and the high-frequency vibration acceleration, instantaneous speed, lubricating oil temperature and oil particle size of the input and output bearings of the gearbox. Extract the high-frequency vibration characteristics of the high-frequency vibration acceleration; Calculate the temperature residual between the real-time lubricating oil temperature and the preset oil temperature.

3. The method according to claim 1, characterized in that, The process of inputting the oil temperature residual, the high-frequency vibration characteristics, the real-time speed, and the real-time torque into a preset gearbox model based on a physical information neural network, and outputting the gearbox stress cycle sequence, includes: The oil temperature residual, the high-frequency vibration characteristics, the real-time speed, and the real-time torque are input into a preset gearbox model based on a physical information neural network, and the contact stress and local slip ratio are output. Based on the contact stress and the real-time torque, a Hertzian contact constraint is constructed. Based on the real-time rotational speed, the oil temperature residual, and the local slip ratio, a lubrication balance constraint is constructed. Based on the contact stress, the true value of the contact stress, the local slip ratio, and the true value of the local slip ratio, a data-driven loss is constructed; The total loss is constructed using the Hertzian contact constraint, the lubrication balance constraint, and the data-driven loss. With the goal of minimizing the total loss, the model parameters of the gearbox model based on the physical information neural network are updated to generate the target gearbox model based on the physical information neural network. The oil temperature residual, the high-frequency vibration characteristics, the real-time speed, and the real-time torque are input into the target gearbox model based on the physical information neural network to generate the stress cycle sequence of the gearbox.

4. The method according to claim 1, characterized in that, The calculation of the total cumulative damage of the gearbox based on the stress cycle sequence of the gearbox includes: Based on the stress cycle sequence of the gearbox, combined with stress and life curves, the percentage of cycle damage at each stress level is calculated. The total cumulative damage of the gearbox is calculated based on the percentage of cyclic damage under each stress level.

5. The method according to claim 1, characterized in that, The process of weightedly fusing the oil wear index corresponding to the oil temperature residual, the vibration energy entropy corresponding to the high-frequency vibration characteristics, and the stress damage rate corresponding to the total cumulative damage to generate a total health score includes: The oil temperature residual is normalized to obtain the oil wear index; The high-frequency vibration characteristics are normalized to obtain the vibration energy entropy; The total cumulative damage is normalized to obtain the stress damage rate. The oil wear index, the vibration energy entropy, and the stress damage rate are weighted and fused to generate a total health score.

6. The method according to claim 1, characterized in that, The process of predicting the remaining usable lifespan based on the gearbox's total health score and generating corresponding early warning information according to the remaining usable lifespan includes: Based on the total health score of the gearbox and the real-time wind conditions corresponding to the wind turbine, the predicted speed and predicted torque are obtained. The predicted speed and the predicted torque are input into a preset long short-term memory network, and a predicted health score is output. Determine whether the predicted health score is less than a first preset health threshold and greater than a second preset health threshold; If so, it is determined that the first remaining usable lifespan has been reached, and a corresponding warning message is generated; If not, then determine whether the predicted health score is less than the second preset health threshold; If so, it is determined that the second remaining usable lifespan has been reached, and a corresponding warning message is generated.

7. A dynamic health assessment device for wind turbine gearboxes, characterized in that, The device includes: The acquisition module is used to acquire multi-source heterogeneous data of the wind turbine and extract the feature vector of the multi-source heterogeneous data, wherein the feature vector includes oil temperature residual, high-frequency vibration characteristics, real-time speed and real-time torque; The output module is used to input the oil temperature residual, the high-frequency vibration characteristics, the real-time speed and the real-time torque into a preset gearbox model based on a physical information neural network, and output the stress cycle sequence of the gearbox. A calculation module is used to calculate the total cumulative damage of the gearbox based on the stress cycle sequence of the gearbox; The preprocessing module is used to perform weighted fusion processing on the oil wear index corresponding to the oil temperature residual, the vibration energy entropy corresponding to the high frequency vibration characteristics, and the stress damage rate corresponding to the total cumulative damage degree to generate a total health score. The prediction module is used to predict the remaining usable lifespan based on the total health score of the gearbox, and generate corresponding early warning information according to the remaining usable lifespan.

8. An electronic device, characterized in that, include: A memory and a processor are interconnected, the memory stores computer instructions, and the processor executes the wind turbine gearbox dynamic health assessment method according to any one of claims 1 to 6 by executing the computer instructions.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the wind turbine gearbox dynamic health assessment method according to any one of claims 1 to 6.

10. A computer program product, characterized in that, Includes computer instructions for causing a computer to execute the wind turbine gearbox dynamic health assessment method according to any one of claims 1 to 6.

Images