A life prediction system for offshore wind turbine gearbox based on digital twinning

Abstract

This application discloses a digital twin-based offshore wind turbine gearbox life prediction system, relating to the field of gearbox life prediction. The system includes: a digital twin database storing the three-dimensional geometric model of the offshore wind turbine gearbox, an initial multiphysics simulation model, and historical data of the digital twin throughout its entire lifecycle; a real-time data acquisition module collecting multi-source operating data of the offshore wind turbine gearbox in real time; a multiphysics simulation module constructing the final multiphysics simulation model to form a digital twin; a life prediction module integrating the multiphysics simulation results output by the digital twin, combined with measured vibration signals and measured oil data, using an LSTM network with a fusion attention mechanism to learn the life decay law and predict the remaining life; and a visualization and interaction module displaying the digital twin and showing the remaining life prediction curve through a three-dimensional visualization platform. This application achieves intelligent management of the entire process from data integration to life prediction and operation and maintenance decision-making.

Description

Technical Field

[0001] This application relates to the field of gearbox life prediction, and in particular to a digital twin-based offshore wind turbine gearbox life prediction system. Background Technology

[0002] In the operation and maintenance (O&M) scenarios of offshore wind power equipment, with the continuous growth of wind power installed capacity and the advancement of deep-sea wind power development, offshore wind turbine gearboxes, as the core components of energy transmission, face severe challenges in terms of operational reliability and lifespan management. Traditional O&M models are no longer suitable for the equipment management needs of the complex offshore environment, and there are many problems that urgently need to be solved. Currently, 3D models and simulation data from the design phase, processing parameters from the manufacturing phase, real-time monitoring data from the operation phase, and maintenance records from the maintenance phase are scattered and stored in different media, lacking systematic integration, resulting in a broken data chain throughout the entire lifecycle and low data utilization. Moreover, O&M personnel need to manually summarize and analyze multi-source data, which not only consumes a lot of time but is also prone to delays in fault diagnosis and deviations in lifespan assessment due to data matching errors and missing historical records, seriously affecting the timeliness and accuracy of O&M decisions.

[0003] Meanwhile, in terms of lifespan prediction, existing methods mostly rely on single-physics field simulations, making it difficult to integrate the coupling effects of multiple fields such as mechanics, thermodynamics, and fluid mechanics. Furthermore, they cannot dynamically correct prediction models by combining real-time operational data, resulting in large errors in remaining lifespan prediction and often leading to situations of "over-maintenance" or "under-maintenance." In the face of the complex marine environment, the damage mechanism of gearboxes exhibits multi-factor coupling characteristics. Traditional fault warning methods based on human experience are unable to accurately capture early weak fault signals, resulting in significant warning lag. In addition, the equipment's three-dimensional structural information, operating status data, and lifespan prediction results lack intuitive visual correlation and display. Maintenance personnel cannot fully grasp the internal stress distribution of gearboxes, the wear status of key components, and the remaining lifespan trend through traditional interfaces. The decision-making process relies excessively on personal experience rather than data support, which seriously restricts the level of intelligent operation and maintenance of offshore wind power equipment.

[0004] Therefore, there is an urgent need for a system that can integrate gearbox lifecycle data, achieve real-time acquisition and unified processing of multi-source operating data, construct accurate multi-physics digital twins, dynamically predict remaining lifespan, provide intelligent early warning and visual interaction, and realize intelligent management of the entire process from data integration to lifespan prediction to operation and maintenance decision-making, so as to provide strong support for the safe and stable operation of offshore wind turbine gearboxes. Summary of the Invention

[0005] The purpose of this application is to provide a digital twin-based offshore wind turbine gearbox life prediction system to achieve intelligent management of the entire process from data integration to life prediction and operation and maintenance decision-making.

[0006] To achieve the above objectives, this application provides the following solution.

[0007] This application provides a digital twin-based offshore wind turbine gearbox life prediction system, which includes the following modules.

[0008] A digital twin database is used to store the three-dimensional geometric model of the offshore wind turbine gearbox, the initial multiphysics simulation model, and the historical data of the digital twin throughout its entire life cycle; the historical data of the digital twin includes multi-source operational data.

[0009] The real-time data acquisition module is used to collect multi-source operating data of offshore wind turbine gearboxes in real time, providing real-time input for the digital twin.

[0010] The multiphysics simulation module is used to construct the final multiphysics simulation model based on the three-dimensional geometric model and the initial multiphysics simulation model of the offshore wind turbine gearbox, forming a digital twin.

[0011] The life prediction module is used to integrate the multiphysics simulation results output by the digital twin, combine them with measured vibration signals and measured oil data, and use an LSTM network with a fusion attention mechanism to learn the life decay law and predict the remaining life; the multiphysics simulation results are the simulation results of the multi-source operating data.

[0012] The visualization and interaction module is used to display the digital twin through a 3D visualization platform and show the remaining life prediction curve.

[0013] According to the specific embodiments provided in this application, this application has the following technical effects.

[0014] This application utilizes a digital twin and an LSTM network architecture with a fusion attention mechanism to integrate multiphysics simulation results with real-time vibration and oil data. It identifies the damage evolution trends of gearbox tooth surface fatigue, bearing wear, etc., effectively avoiding the prediction bias caused by traditional life assessment relying on single data. It significantly reduces the risk of excessive maintenance costs or sudden failures caused by misjudging the remaining life, and provides reliable data support for the full life cycle operation and maintenance of offshore wind turbine gearboxes.

[0015] This application not only achieves unified storage and management of the three-dimensional geometric model, multiphysics simulation model, and historical data of the digital twin of the offshore wind turbine gearbox through a digital twin database, but also continuously optimizes the multiphysics simulation model by combining real-time monitoring data. This enables it to accurately adapt to the multi-field coupling damage mechanism in complex environments such as high salt spray and strong vibration at sea, significantly improving its adaptability to different operating stages and different failure modes of the gearbox, and ensuring that the accuracy of remaining life prediction is continuously optimized as the equipment operates.

[0016] This application achieves accurate life prediction while intuitively presenting the gearbox's three-dimensional structure, multi-physics field distribution, remaining life prediction curve, and early warning information through a three-dimensional visualization platform. In addition, maintenance personnel can view the status of the digital twin in real time, provide feedback on maintenance data to optimize the simulation model, and obtain multi-level risk warnings in advance through the threshold early warning module, which greatly improves the efficiency of maintenance decision-making and equipment reliability. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of a digital twin-based offshore wind turbine gearbox life prediction system provided in one embodiment of this application. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] To make the objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] like Figure 1 As shown in the figure, this application provides a digital twin-based offshore wind turbine gearbox life prediction system, which includes the following modules.

[0022] A digital twin database is used to store the three-dimensional geometric model of the offshore wind turbine gearbox, the initial multiphysics simulation model, and the historical data of the digital twin throughout its entire life cycle; the historical data of the digital twin includes multi-source operational data.

[0023] The real-time data acquisition module is used to collect multi-source operating data of offshore wind turbine gearboxes in real time, providing real-time input for the digital twin.

[0024] The multiphysics simulation module is used to construct the final multiphysics simulation model based on the three-dimensional geometric model and the initial multiphysics simulation model of the offshore wind turbine gearbox, forming a digital twin.

[0025] The life prediction module is used to integrate the multiphysics simulation results output by the digital twin, combine them with measured vibration signals and measured oil data, and use an LSTM network with a fusion attention mechanism to learn the life decay law and predict the remaining life; the multiphysics simulation results are the simulation results of the multi-source operating data.

[0026] The visualization and interaction module is used to display the digital twin through a 3D visualization platform and show the remaining life prediction curve.

[0027] In an exemplary embodiment, the three-dimensional geometric model of the offshore wind turbine gearbox includes: gear tooth profile parameters, shaft system structural parameters, bearing assembly parameters, housing parameters, and assembly constraint information such as the relative positional relationships and connection methods of each component.

[0028] The initial multiphysics simulation model includes: a gear meshing dynamics model, a bearing contact mechanics model, a lubricating oil dynamics model, a structural thermodynamics model, and the basic parameters of each model; the gear meshing dynamics model includes the meshing force calculation equation; the bearing contact mechanics model includes the contact stress calculation formula; the lubricating oil dynamics model includes the oil film thickness calculation model; the structural thermodynamics model includes the temperature field distribution equation; the basic parameters include material properties and initial boundary conditions; material properties include elastic modulus, Poisson's ratio, fatigue limit, and density; initial boundary conditions include rated speed, rated load, ambient temperature, simulation step size, and convergence criteria, etc.

[0029] The historical data of the digital twin throughout its entire lifecycle includes multi-source operational data, which includes: scheme iteration data and simulation verification reports during the design phase; real-time sensing historical data, digital twin dynamic simulation data, and maintenance records during the operation phase; and final inspection reports and life assessment summary data during the decommissioning phase.

[0030] The method for obtaining the three-dimensional geometric model of the offshore wind turbine gearbox is as follows: point cloud data is obtained by laser scanning of the gearbox entity, and then processed by Geomagic DesignX to generate a three-dimensional model.

[0031] The initial multiphysics simulation model was obtained as follows: based on the three-dimensional geometric model, it was constructed in the multiphysics simulation software ANSYS and Abaqus. The initial modeling was completed by defining the control equations, material property parameters, and boundary conditions of each physical field. The core parameters of the model were verified by combining the theoretical formulas in the gearbox design manual. Then, the model was initially calibrated by the factory test data to form the basic version.

[0032] The acquisition of historical data for the digital twin throughout its entire lifecycle is achieved through the following methods: Design phase data is automatically recorded and synchronized to the digital twin database by design software and manufacturing execution system; Operation phase data is collected in real time by sensors and uploaded via industrial Ethernet, while simulation data of the digital twin is output and stored in real time by simulation software; Maintenance records are manually entered by maintenance personnel, including maintenance work orders and component replacement records; Decommissioning phase data is generated through on-site inspection reports and manually uploaded to the digital twin database, ultimately forming a complete lifecycle data chain.

[0033] In one exemplary embodiment, the multi-source operating data includes: vibration signals, oil data, operating speed, and load torque; the vibration signals include spectral characteristics of each frequency band; the oil data includes abrasive particle concentration, viscosity, and component percentage.

[0034] In practical applications, the vibration signal acquisition method is as follows: piezoelectric accelerometers are deployed and installed at key locations in the gearbox housing. The mechanical vibration is converted into an electrical signal by the sensor, amplified by a signal conditioning amplifier, then converted from analog to digital, and finally extracted by Fourier transform to obtain spectral features, including gear meshing frequency, bearing fault characteristic frequency and its sideband energy, so as to realize real-time acquisition and feature extraction of vibration signals.

[0035] The oil data acquisition method is as follows: an online oil monitoring system is used to continuously extract a small amount of oil from the gearbox lubrication circuit through the built-in oil circulation pipeline; a laser particle counter is used to detect the concentration and size distribution of abrasive particles in the oil sample in real time; a viscosity sensor is used to measure the kinematic viscosity at 40℃ / 100℃; and an infrared spectrometer is used to identify the amount of additive loss and the proportion of contaminant components in the oil. All oil parameters are fused to form oil data.

[0036] The operating speed is collected by a magnetoelectric speed sensor.

[0037] The load torque is acquired by a strain gauge torque sensor.

[0038] In one exemplary embodiment, the multiphysics simulation module specifically includes the following units.

[0039] An initialization unit is used to initialize the initial multiphysics simulation model.

[0040] The multiphysics coupling simulation model determination unit is used to establish the coupling relationship between the multiphysics fields of the gear on the initialized multiphysics simulation model and output the multiphysics coupling simulation model.

[0041] The simulation unit is used to collect multi-source operating data of the offshore wind turbine gearbox in real time based on the three-dimensional geometric model of the offshore wind turbine gearbox, and input the real-time collected multi-source operating data into the multi-physics coupled simulation model for simulation calculation. The simulation output results are compared, and the key parameters of the model are dynamically adjusted through optimization algorithms to drive calibration and output the calibrated simulation model.

[0042] The digital twin generation unit is used to verify the simulation accuracy of the calibrated simulation model, construct the final multiphysics simulation model, and form a digital twin.

[0043] In practical applications, the construction of the final multiphysics simulation model specifically includes the following steps, wherein the physical field coupling setting process of the multiphysics coupled simulation model specifically includes: time-varying meshing stiffness excitation force calculation sub-unit, friction heat source power density calculation sub-unit, oil viscosity update unit, and oil film pressure distribution calculation unit.

[0044] A1. Model initialization.

[0045] It needs to be explained that when constructing the final multiphysics simulation model, the gearbox 3D geometric model and the initial multiphysics simulation model from the database are first called. The 3D model is imported into the multiphysics simulation software. Based on the geometric constraints in the base version, the mesh is discretized using the governing equations. A fine mesh of 0.1-1mm is selected for critical regions, and a coarse mesh of 5-10mm is used for non-critical regions. Stress and temperature gradients are captured by local mesh refinement. At the same time, the material properties in the base version are loaded, and linear elastic constitutive relations are used. , It is a stress vector. This is an elasticity matrix containing material properties, including elastic modulus, Poisson's ratio, etc. It is the strain vector. Then, set the initial boundary conditions such as rated speed and no-load torque to lay the foundation for subsequent calculations.

[0046] A2. Physical field coupling settings.

[0047] It needs to be explained that, based on the initial model, a multi-field coupling relationship is established between gear meshing dynamics, bearing contact mechanics, lubricating oil dynamics, and structural thermodynamics. The dynamic model is linked to the structural mechanics model through gear meshing force.

[0048] The time-varying meshing stiffness excitation force calculation sub-unit is used to calculate the time-varying meshing stiffness excitation force. , It is time-varying meshing stiffness, determined by gear parameters such as the number of teeth and the contact ratio, reflecting the dynamic changes in stiffness during meshing. Gear transmission errors, including manufacturing errors and equipment errors, are key factors affecting meshing force fluctuations.c It is the meshing damping coefficient, used to describe the energy dissipation during the meshing process.

[0049] The frictional heat source power density calculation subunit is used to calculate... As the load input to the structural mechanics model, the stress distribution σ(t) of the gears and shaft system is calculated, and the power density of the frictional heat source is also calculated. , u The coefficient of friction; To contact pressure, It is calculated using Hertz contact theory to reflect the pressure distribution in the gear / bearing contact area; The sliding velocity is derived from the gear meshing motion relationship.

[0050] Oil viscosity update unit, used to update oil viscosity When the heat source is input into the thermodynamic model to calculate the temperature field T(t), the oil viscosity is then updated using the Vogel equation based on the temperature field T(t): A, B, and C are oil characteristic parameters, determined by the type of lubricating oil. The temperature field values ​​calculated for the thermodynamic model.

[0051] The oil film pressure distribution calculation unit is used to input the updated oil viscosity into the simplified one-dimensional Reynolds equation and calculate the oil film pressure distribution. , h The oil film thickness is a dynamic variable affected by gear / bearing motion and load. x These are the one-dimensional coordinates for oil film analysis. p Let be the oil film pressure, and be the oil film force distribution to be solved. U The rolling speed is derived from the gearbox's operating speed. The dynamic viscosity of the oil is determined by the updated value of the Vogel equation.

[0052] The oil film pressure distribution p(x) is used to correct the bearing contact force and fed back into the gear meshing dynamics model for recalculation, according to time step t. The process iterates from t+Δt until the temperature field converges, enabling real-time interaction of multi-physics parameters.

[0053] A3. Real-time data-driven calibration.

[0054] The vibration spectrum characteristics, oil viscosity, and operating speed output by the real-time data acquisition module are used as dynamic boundary conditions and input into the above coupling for simulation calculation.

[0055] The simulation unit specifically includes: a comparison subunit, used to construct a parameter optimization objective function by comparing simulation results with measured data. ， For parameters to be optimized, It is the meshing stiffness scaling factor, and T is the transpose matrix. This is the viscosity correction factor. For simulation output vectors, containing 、 ; This is a vector of measured data, containing... and .

[0056] The parameter adjustment subunit is used to optimize the objective function based on the parameters and dynamically adjust the key parameters of the model using a sample group optimization algorithm. , Let be the particle velocity in the i-th stage and k-th iteration; Let be the particle velocity in the (k+1)th iteration of the i-th stage; Let be the particle position in the i-th stage and k-th iteration; Let this be the particle position in the (k+1)th iteration of the i-th stage. w Inertial weight; c 1. c 2 represents the learning factor; r 1. r 2 is a random number; This represents the optimal position in the particle's history. This is the globally optimal position.

[0057] The key parameters of the model were dynamically adjusted until the error between the simulation results and the measured data was less than 5%, to ensure that the multiphysics simulation results could accurately reflect the real-time state of the gearbox; among these simulation results were: tooth surface contact stress. Bearing vibration acceleration The measured data includes: vibration inversion stress values. Accelerometer data .

[0058] The simulation model calibration subunit is used to output the calibrated simulation model based on the adjusted key parameters of the model.

[0059] A4. Simulation accuracy verification and version fixation.

[0060] Historical typical operating conditions such as rated load, startup process, shutdown process, and overload operation under extreme wind speed are selected. The simulation results of the calibrated model are compared and verified with the measured data of the corresponding operating conditions. The mean absolute error and root mean square error are calculated. When the error index meets the preset threshold, the model is determined as the final multiphysics simulation model, which serves as the core carrier of the digital twin and supports subsequent functions.

[0061] In the multiphysics coupling setting, the calculation order of different physical fields adopts an "implicit coupling" strategy: first, the gear meshing dynamics calculation is performed to obtain the instantaneous meshing force, which is then used as a load input to the structural mechanics model to calculate the stress distribution. Simultaneously, the meshing force is converted into a frictional heat source and input into the thermodynamic model to calculate the temperature field. Then, the oil viscosity change output by the temperature field is fed back to the hydrodynamic model to update the oil film thickness calculation, forming a closed-loop iteration to ensure that the real-time interaction influence of each physical field parameter is accurately captured.

[0062] The final digital twin is a digital model that can be synchronized with the offshore wind turbine gearbox in real time. It is mainly presented through a 3D visualization platform. Operation and maintenance personnel can zoom and rotate the model to view internal details, and can also simulate the impact of faults by adjusting parameters. They can grasp the real-time status and potential problems of the gearbox without disassembling the equipment.

[0063] In one exemplary embodiment, the lifetime prediction module specifically includes the following units.

[0064] Define the unit to define damage quantification indicators and combine multi-field simulations with real-time data to model key failure modes.

[0065] The cumulative damage output unit is used to calculate the cumulative damage by combining the simulated physical parameters of the damage quantification index with real-time monitoring data, and output the cumulative damage.

[0066] The remaining lifetime prediction output unit is used to predict the remaining lifetime based on the accumulated damage using an LSTM network with a fusion attention mechanism, output the current remaining lifetime prediction value of the gearbox, and perform uncertainty quantization on the remaining lifetime prediction value using a Weibull distribution.

[0067] The specific process for predicting remaining useful life is as follows.

[0068] B1. Damage Quantification: Modeling key failure modes.

[0069] It needs to be explained that the cumulative fatigue damage on the tooth surface is extracted from the multiphysics simulation results of the digital twin. Bearing roller wear depth Physical parameters such as oil film thickness change Δh(t) are collected; simultaneously, the gear meshing frequency sideband energy of the vibration signal is extracted from the real-time monitoring data. Bearing fault characteristic frequency amplitude and the rate of increase in abrasive particle concentration in the oil. viscosity change rate Features such as...

[0070] The cumulative amount of tooth surface fatigue damage Substituting the stress spectrum into Miner's linear cumulative damage rule, the calculated value is the value at the current moment. .

[0071] The bearing roller wear depth The contact pressure and relative sliding speed between the bearing rollers and raceways are simulated and then integrated into physical models such as the Archard wear model. The result is the cumulative wear depth.

[0072] The change in oil film thickness Δh(t) is the difference between the current thickness and the initial design thickness or the thickness under healthy conditions.

[0073] The vibration signal's gear meshing frequency sideband energy On the spectrum diagram, locate the gear meshing frequency and its sidebands, i.e., the meshing frequency ± shaft rotation frequency, and calculate the total energy within a certain bandwidth around these sideband frequencies.

[0074] The bearing fault characteristic frequency amplitude The theoretical fault characteristic frequencies of the bearing, such as the fault frequencies of the inner ring, outer ring, and rolling elements, are calculated based on the bearing's geometric parameters, and the amplitudes of these frequency points are directly read from the spectrum.

[0075] Oil abrasive concentration growth rate Calculate the average increase in abrasive particle concentration per unit time.

[0076] viscosity change rate You can directly read the rate of change of the current viscosity value relative to the initial viscosity of the new oil, or calculate its gradient over time.

[0077] Combining multi-field simulations and real-time data, key failure modes are modeled: fatigue modeling of gear surfaces is performed; allowable contact stress is determined by sub-elements used in the modified ISO 6336 standard contact stress life model. ， Allowable contact stress reflects the gear material's ability to resist contact fatigue. For lifespan factor, ， ， This refers to the correction factor for lubricant, speed, and surface roughness. The better the lubrication, the higher the speed, and the smoother the surface, the larger the correction factor and the higher the allowable stress. The fatigue limit of a material is obtained through material experiments and is an inherent property of materials such as gear steel.

[0078] The stress spectrum conversion subunit is used to extract the time-varying tooth surface contact stress based on the allowable contact stress. The stress spectrum was converted using the rainflow counting method. , Let i be the stress amplitude in stage i. To correspond to the number of cycles, combined with real-time abrasive particle concentration Correct the cumulative rate of gear fatigue damage.

[0079] B2. Perform cumulative damage calculation.

[0080] The alignment subunit is used to align the simulated physical parameters of the damage quantification index output by the digital twin with the characteristic parameters of the vibration signal in the real-time monitoring data according to the time series, and construct the input feature vector, which is the simulated physical parameters output by the digital twin, i.e., the cumulative amount of tooth surface fatigue damage. Bearing roller wear depth Physical parameters such as oil film thickness change Δh(t); and the gear meshing frequency sideband energy of the vibration signal output from real-time monitoring data. Bearing fault characteristic frequency amplitude and the rate of increase in abrasive particle concentration in the oil. viscosity change rate Features are aligned according to time series to ensure that simulated data corresponds to measured data at the same time point. Through normalization, the data scale is unified, and the input feature vector is constructed as follows: .

[0081] Cumulative damage calculation subunit, used for calculating damage based on input feature vectors Calculate the cumulative damage to gears and bearings.

[0082] The cumulative damage calculation subunit specifically includes: a linear damage accumulation subunit, used to calculate linear damage accumulation using the Miner rule.

[0083] Assuming damage accumulates linearly with stress cycles, the linear cumulative damage of gears and bearings is determined using the following formula: , For linear cumulative damage, for The actual number of loops, Nj for The corresponding failure cycle count; Let D be the stress amplitude in stage j. When D ≥ 1, fatigue failure of the gear / bearing is determined.

[0084] The nonlinear damage correction sub-element is used to calculate nonlinear damage correction using the Corten-Dolan model: taking into account the nonlinear effect of load sequence on damage, the formula is: , Non-cumulative damage degree This represents the stress amplitude in the first stage. dThe material constant is typically taken as 5 to 6, reflecting the material's sensitivity to the load sequence. Utilizing the real-time high stress amplitude characteristics, its contribution to damage is calculated first.

[0085] B3. Perform remaining life prediction.

[0086] The normalized multi-source feature vector X is input into an LSTM network with an attention mechanism, allowing the network to learn the mapping between feature changes and lifetime decay. Specifically, the LSTM network first automatically assigns weights to different features, focusing on key features such as the abrasive particle concentration growth rate of the oil. Cumulative fatigue damage on tooth surface Give higher attention weight : , It is the first Each input feature H is the hidden state of the LSTM network, and Score is the attention score function used to measure the correlation between features and lifetime decay.

[0087] Next, the LSTM network is trained and used for prediction, utilizing multi-source features from historical data. - Remaining lifespan To train the LSTM network on training samples, the network's output layer maps the remaining lifetime through a fully connected (FC) layer. , The damage growth coefficient is used as a reference, and after correction based on the physical model, the damage growth coefficient is dynamically adjusted to ultimately output the predicted remaining life of the gearbox. , This is the current cumulative damage, calculated by fusing multi-source features from the network. It is the historical average damage rate, dynamically updated through real-time data, reflecting the long-term trend of damage growth.

[0088] B4. Quantification of uncertainty.

[0089] Taking into account the uncertainties of model assumptions and data noise, the Weibull distribution is used to quantify the remaining lifetime probabilistically: , F ( t )yes t The probability of failure at time 1. It is the characteristic lifetime, determined by the median of the remaining lifetime prediction, representing the lifetime at 50% reliability. β It is a shape parameter, when β When the value is greater than 1, it indicates that wear and tear has taken effect.

[0090] The purpose of uncertainty quantization is to fit the parameters of the Weibull distribution using multiple sets of predictions output by the LSTM network. andβ It outputs the remaining lifespan range under different confidence levels, providing a more comprehensive risk reference for operation and maintenance decisions.

[0091] The role of the fusion attention mechanism is to allow the model to focus on features that have a more significant impact on lifespan during the learning process. For example, when the abrasive particle concentration rises rapidly, the model will automatically increase the weight of this feature in the prediction calculation, avoid interference from secondary features, and thus improve the prediction accuracy.

[0092] During model training, data from different stages such as normal operation, minor damage, and severe wear in the historical data of the entire life cycle are used as samples to ensure accurate learning of various lifespan decay scenarios. During the prediction process, the model input is updated every hour with the latest simulation results and measured data to dynamically correct the predicted value, so that the remaining life prediction is always synchronized with the actual state of the gearbox.

[0093] In one exemplary embodiment, this application also includes the following modules.

[0094] The error calibration module is used to compare the multiphysics simulation results of the digital twin, the measured vibration signal, and the measured oil data. By adjusting the simulation parameters, the deviation between the digital twin and the physical equipment is dynamically calibrated.

[0095] The threshold warning module is used to set multi-level warning thresholds and generate warning information based on the remaining life prediction results.

[0096] The error calibration module first obtains simulation output from the digital twin, including multi-physics field data such as tooth surface contact stress, bearing vibration acceleration, and oil film thickness. At the same time, it collects real-time vibration and oil data of the physical equipment. By constructing error evaluation indicators, it quantifies the deviation between simulation and measured data, such as comparing the peak frequency deviation and energy distribution difference between the simulated vibration spectrum and the measured spectrum.

[0097] Error evaluation metrics are divided into absolute square error and root mean square error. Absolute square error measures the squared absolute deviation between simulated and measured values, and the formula is: N This refers to the number of data samples, such as the number of vibration signal sampling points or the number of oil parameter measurements within a certain time period. For the first Simulation data for individual samples, such as vibration acceleration and oil film thickness output by the digital twin, For the first Measured data for each sample, such as vibration acceleration collected by sensors and viscosity values ​​measured by oil analyzers.

[0098] The root mean square error (RMSE) measures the square root of the average of the squared deviations between simulated and measured values. It is more sensitive to large deviations. The formula is: The physical meaning of RMSE is: by squaring the weight of larger deviations, it can better reflect the degree of deviation between outliers and measured values ​​in simulation data. The unit is the same as that of the original data. The smaller the value, the more stable the overall deviation.

[0099] The threshold warning module first uses the remaining life (RUL) predicted value output by the life prediction module, combined with key parameters such as the gearbox's design life and maintenance cycle, to divide the warning threshold range into multiple levels: 30% of the remaining life as the design life is set as "Level 1 Warning", 20% as "Level 2 Warning", and 10% as "Level 3 Warning". At the same time, the threshold division will be dynamically adjusted in combination with real-time operating conditions. For example, when the load increases suddenly under extreme wind speeds, the trigger ratio of the warning threshold will be appropriately increased. For example, the Level 1 warning threshold will be temporarily raised to 35% to avoid warning delays caused by sudden damage acceleration.

[0100] When the predicted remaining life falls within the corresponding threshold range, the module automatically generates structured early warning information, including the warning level, the current remaining life estimate, key damage indicators such as the excessive value of tooth surface stress and the growth rate of abrasive particle concentration, and recommended measures. The early warning information is output through a 3D visualization platform and is simultaneously linked to the 3D model of the digital twin, highlighting the corresponding warning area in the model to help maintenance personnel quickly locate potential risk points and reduce the risk of unplanned equipment downtime.

[0101] In one exemplary embodiment, the visualization interaction module displays the remaining life prediction curve of the gearbox digital twin through a 3D visualization platform, allowing maintenance personnel to view the status of the digital twin and provide feedback on on-site maintenance data to optimize the simulation model.

[0102] The 3D visualization platform (hereinafter referred to as the platform) presents the internal structure and operating status of the gearbox in a dynamic 3D model. Gears, bearings and other components operate dynamically according to the actual speed. The distribution of physical quantities such as stress and temperature in real time is displayed using color cloud maps at key locations such as tooth surfaces and bearings. Different colors represent the high and low values ​​of the parameters. The left panel of the platform synchronously displays the remaining life prediction curve. The horizontal axis is time and the vertical axis is the estimated remaining life. The curve is marked with multi-level warning thresholds using dashed lines. When the predicted value is close to the threshold, the corresponding interval will be automatically highlighted in red. Maintenance personnel can zoom and rotate the model with the mouse and click on any component to view its full life cycle data in a pop-up window.

[0103] The platform allows maintenance personnel to submit on-site maintenance data through a built-in data entry interface, such as the model of replaced components, measured damage dimensions, and post-repair performance test values. This data is automatically synchronized to the digital twin database as a basis for optimizing the simulation model. For example, if maintenance reveals that the actual wear of a bearing is greater than the simulation prediction, the system will feed this deviation data back to the error calibration module to adjust the key parameters of the bearing wear model, making the simulation results of the digital twin closer to the actual state. At the same time, the platform provides interactive functions, allowing maintenance personnel to manually input parameters. For example, to simulate an increase in the meshing clearance of a gear, the platform generates the remaining life change curve and stress distribution simulation results for that scenario within 3 minutes, providing a digital tool for verifying maintenance solutions.

[0104] This application achieves accurate life prediction while intuitively presenting the gearbox's three-dimensional structure, multi-physics field distribution, remaining life prediction curve, and early warning information through a visual interactive platform. In addition, maintenance personnel can view the status of the digital twin in real time, provide feedback on maintenance data to optimize the simulation model, and obtain multi-level risk warnings in advance through the threshold early warning module, which greatly improves the efficiency of maintenance decision-making and equipment reliability.

[0105] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0106] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A digital-twin-based offshore wind turbine gearbox life prediction system, characterized in that, include: A digital twin database is used to store the three-dimensional geometric model of offshore wind turbine gearboxes, the initial multiphysics simulation model, and historical data of the digital twin throughout its entire life cycle. The historical data of the digital twin includes multi-source operational data; The real-time data acquisition module is used to collect multi-source operating data of offshore wind turbine gearboxes in real time, providing real-time input for the digital twin; The multiphysics simulation module is used to construct the final multiphysics simulation model based on the three-dimensional geometric model and the initial multiphysics simulation model of the offshore wind turbine gearbox, forming a digital twin. The multiphysics simulation module specifically includes: An initialization unit is used to initialize the initial multiphysics simulation model. The multiphysics coupling simulation model determination unit is used to establish the coupling relationship between the gear's multiphysics fields on the initialized multiphysics simulation model and output the multiphysics coupling simulation model; the physical field coupling setting process of the multiphysics coupling simulation model specifically includes: a time-varying mesh stiffness excitation force calculation subunit configured to calculate a time-varying mesh stiffness excitation force ; wherein, ; is a time-varying mesh stiffness at the time instant t ; is a gear transmission error at the time instant t ; c is a mesh damping coefficient; is a time-varying mesh stiffness at the time instant t is a rate of change of the gear transmission error at the time instant The frictional heat source power density calculation subunit is used to calculate... As a load, the structural mechanics model is input to calculate the stress distribution σ(t) of the gears and shafts, as well as the power density of the frictional heat source. ;in, , u The coefficient of friction; For contact pressure; The sliding speed; Oil viscosity update unit, used to update oil viscosity As a heat source, the temperature field T(t) is calculated using a thermodynamic model, and the oil viscosity is updated based on the temperature field T(t) using the Vogel equation. ;in, A, B, and C are all oil characteristic parameters. Temperature field values ​​calculated for a thermodynamic model; The oil film pressure distribution calculation unit is used to input the updated oil viscosity into the simplified one-dimensional Reynolds equation to calculate the oil film pressure distribution. p ;in, ; h Oil film thickness; x For oil film analysis, the one-dimensional spatial coordinates are... U For the scrolling speed, The dynamic viscosity of the oil; The simulation unit is used to collect multi-source operating data of the offshore wind turbine gearbox in real time based on the three-dimensional geometric model of the offshore wind turbine gearbox, and input the real-time collected multi-source operating data into the multi-physics coupling simulation model for simulation calculation. The simulation output results are compared, and the key parameters of the model are dynamically adjusted through optimization algorithms to drive calibration and output the calibrated simulation model. The digital twin generation unit is used to verify the simulation accuracy of the calibrated simulation model, construct the final multiphysics simulation model, and form a digital twin. The life prediction module is used to integrate the multiphysics simulation results output by the digital twin, combine them with measured vibration signals and measured oil data, and use an LSTM network with a fusion attention mechanism to learn the life decay law and predict the remaining life; the multiphysics simulation results are the simulation results of the multi-source operating data. The visualization and interaction module is used to display the digital twin through a 3D visualization platform and show the remaining life prediction curve.

2. The offshore wind turbine gearbox life prediction system based on digital twins according to claim 1, characterized in that, Also includes: The error calibration module is used to compare the multiphysics simulation results of the digital twin, the measured vibration signal, and the measured oil data. By adjusting the simulation parameters, the deviation between the digital twin and the physical equipment is dynamically calibrated. The threshold warning module is used to set multi-level warning thresholds and generate warning information based on the remaining life prediction results.

3. The offshore wind turbine gearbox life prediction system based on digital twins according to claim 1, characterized in that, The multi-source operating data includes: vibration signals, oil data, operating speed, and load torque; The vibration signal includes spectral characteristics of each frequency band; The oil data includes abrasive particle concentration, viscosity, and component percentage.

4. The offshore wind turbine gearbox life prediction system based on digital twin as described in claim 1, characterized in that, The simulation unit specifically includes: Comparison sub-element, used to compare the tooth surface contact stress output in the simulation. Bearing vibration acceleration Compared with the measured vibration inversion stress value Accelerometer data Construct the objective function for parameter optimization: ;in, For parameters to be optimized, It is the meshing stiffness scaling factor. T It is the transpose matrix; This is the viscosity correction factor; For simulation output vectors, Include as well as ; This is the measured data vector. Include as well as ; The parameter adjustment subunit is used to optimize the objective function based on the parameters and dynamically adjust the key parameters of the model using a sample group optimization algorithm. ;in, Let be the particle velocity in the i-th stage and k-th iteration; Let be the particle velocity in the (k+1)th iteration of the i-th stage; Let be the particle position in the i-th stage and k-th iteration; Let be the particle position in the (k+1)th iteration of the i-th stage; w Inertial weight; c 1. c 2 represents the learning factor; r 1. r 2 is a random number; This represents the optimal position in the particle's history. The globally optimal position; The simulation model calibration subunit is used to output the calibrated simulation model based on the adjusted key parameters of the model.

5. The offshore wind turbine gearbox life prediction system based on digital twin as described in claim 1, characterized in that, The lifetime prediction module specifically includes: Define the unit to define damage quantification indicators and combine multi-field simulations and real-time data to model key failure modes; The cumulative damage output unit is used to calculate the cumulative damage by combining the simulated physical parameters of the damage quantification index with the real-time monitoring data, and output the cumulative damage. The remaining lifetime prediction output unit is used to predict the remaining lifetime based on the accumulated damage using an LSTM network with a fusion attention mechanism, output the current remaining lifetime prediction value of the gearbox, and perform uncertainty quantization on the remaining lifetime prediction value using a Weibull distribution.

6. The offshore wind turbine gearbox life prediction system based on digital twins according to claim 5, characterized in that, The key failure mode modeling is based on fatigue modeling of the gear surface. The defined unit specifically includes: Allowable contact stress determination sub-element, used for employing a modified ISO 6336 standard contact stress life model. Determine the allowable contact stress; where, Allowable contact stress reflects the gear material's ability to resist contact fatigue; This refers to the lifespan factor. ， ， These are the correction factors for lubricant, speed, and roughness, respectively. The material's basic fatigue limit; The stress spectrum conversion subunit is used to extract the time-varying tooth surface contact stress based on the allowable contact stress. And by using the rainflow counting method Converted to stress spectrum ;in, Let i be the stress amplitude in stage i. This corresponds to the number of loops.

7. The offshore wind turbine gearbox life prediction system based on digital twins according to claim 5, characterized in that, The cumulative damage output unit specifically includes: The alignment subunit is used to align the simulated physical parameters of the damage quantification index output by the digital twin with the characteristic parameters of the vibration signal in the real-time monitoring data according to the time series, and construct the input feature vector. The simulated physical parameters include the cumulative amount of tooth surface fatigue damage. Bearing roller wear depth and changes in oil film thickness The characteristic parameters include gear meshing frequency and sideband energy. Bearing fault characteristic frequency amplitude The rate of increase in abrasive particle concentration in the oil and viscosity change rate ; Cumulative damage calculation subunit, used for calculating damage based on input feature vectors Calculate the cumulative damage to gears and bearings.

8. The offshore wind turbine gearbox life prediction system based on digital twins according to claim 7, characterized in that, The cumulative damage calculation subunit specifically includes: Linear damage accumulation sub-unit, used to calculate linear damage accumulation using Miner's rule: Assuming damage accumulates linearly with stress cycles, determine the linear cumulative damage of gears and bearings; the linear cumulative damage is... , For linear cumulative damage, for The actual number of loops, Nj for The corresponding failure cycle count; The stress amplitude is for stage j. Nonlinear damage correction sub-unit, used to calculate nonlinear damage correction using the Corten-Dolan model: Considering the nonlinear effect of load sequence on damage, the nonlinear cumulative damage of gears and bearings is determined; the nonlinear cumulative damage is... , Non-cumulative damage degree This represents the stress amplitude in the first stage. d is a material constant.

Images