A method for controlling the high fatigue performance of wind turbine main shafts

Abstract

This invention relates to the field of industrial program control technology, specifically a control method for high fatigue performance of wind turbine main shafts. The method involves acquiring multi-source sensor parameters of the main shaft during heat treatment and inputting them into a dynamic process module coupled with temperature, phase transition, and stress fields. The state distribution results identified by the module are updated using an assimilation algorithm. Then, a preset performance prediction module is used to predict performance indicators characterizing the fatigue performance of the main shaft. A model predictive controller is employed to generate optimal program control instructions for the heat treatment process based on a comparison between real-time performance indicators and preset optimal performance target values, combined with process constraints. These optimal program control instructions are then sent to the actuator array in the heat treatment equipment to perform zoned and time-controlled heat treatment of the main shaft, achieving an optimized closed-loop. This invention improves the effectiveness and reliability of high fatigue performance control for wind turbine main shafts.

Description

Technical Field

[0001] This invention relates to the field of industrial program control technology, specifically to a control method for high fatigue performance of wind turbine main shafts. Background Art

[0002] In the manufacturing process of large and complex components, traditional production control systems mainly rely on preset process procedures and offline quality inspection methods, lacking an integrated, real-time, and intelligent closed-loop control mechanism. This control method has the following shortcomings:

[0003] (1) Existing control systems are usually based on fixed process parameters, which makes it difficult to make real-time adaptive adjustments according to dynamically changing external conditions. This makes it difficult to effectively cope with variability in the production process and affects the performance consistency between product batches.

[0004] (2) The key performance indicators of a product are affected by its microstructure and internal stress distribution. However, existing systems typically rely on offline testing after molding to assess these characteristics, lacking the ability to monitor and intervene in key quality parameters in real time during the production process.

[0005] (3) Traditional control systems mainly focus on easily measurable macroscopic parameters, but cannot directly target the core factors that affect product quality for precise control, which limits the system’s ability to proactively optimize complex manufacturing processes.

[0006] Therefore, a method for controlling the high fatigue performance of wind turbine main shafts is proposed. Summary of the Invention

[0007] The purpose of this invention is to provide a method for controlling the high fatigue performance of wind turbine main shafts.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] A method for controlling the high fatigue performance of a wind turbine main shaft includes:

[0010] The multi-source sensing parameters of the spindle during the heat treatment process are acquired, and the multi-source sensing parameters are input into the dynamic process module that couples the temperature field, phase transformation field and stress field. The state distribution results identified by the module are updated through the assimilation algorithm.

[0011] Based on the state distribution results, the preset performance prediction module is used to process the data and predict the performance indicators used to characterize the spindle fatigue performance.

[0012] A model predictive controller is used to solve and generate the optimal program control instructions for the heat treatment process based on the comparison results between real-time performance indicators and preset optimal performance target values, combined with process constraints.

[0013] The optimal program control command is sent to the actuator array in the heat treatment equipment. The actuator array performs partitioned and time-controlled heat treatment on the spindle according to the command, so as to drive the real-time performance index to approach the optimal performance target value and perform optimization closed loop.

[0014] Furthermore, the multi-source sensing parameters include at least spindle surface temperature data and acoustic feature data.

[0015] Furthermore, the process of inputting the multi-source sensing parameters into a dynamic process module coupled with the temperature field, phase transition field, and stress field, and updating the comprehensive distribution result identified by the module through an optimization algorithm includes:

[0016] The dynamic process module is constructed in the form of a digital twin model coupled with multi-physics fields;

[0017] The dynamic process module is initialized by loading the geometric model and material properties of the spindle;

[0018] The spindle surface temperature data from the multi-source sensing parameters is used as the thermal boundary condition of the dynamic process module. At the same time, the acoustic feature data from the multi-source sensing parameters are used, combined with optimization algorithms, to correct the temperature distribution, phase ratio distribution, and stress distribution predicted by the dynamic process module.

[0019] Furthermore, the performance prediction module is a proxy model built based on a machine learning algorithm. The input of the proxy model is the feature parameters extracted from the distribution results output by the dynamic process module, and the output is the predicted fatigue life value of the key area of ​​the spindle. The predicted fatigue life value is used to characterize the fatigue performance of the spindle.

[0020] Furthermore, based on the overall condition, the process of using a preset performance prediction module to predict the performance indicators used to characterize the final fatigue performance of the spindle includes:

[0021] Extract characteristic parameters related to temperature distribution, phase ratio distribution, and stress distribution from the distribution results;

[0022] A deep neural network is used as a proxy model for performance prediction, comprising an input layer, a hidden layer, and an output layer. The input layer receives feature parameters, the hidden layer handles the nonlinear relationship between the feature parameters and performance indicators, and the output layer generates predicted fatigue life values ​​for the key region of the main axis.

[0023] The surrogate model is trained using historical data, and the prediction error is minimized through parameter optimization. The feature parameters are then input into the trained deep neural network to obtain the prediction performance metrics.

[0024] Furthermore, the process of using a model predictive controller to reverse-engineer and generate the optimal program control instructions for the heat treatment process, based on the comparison results between real-time performance indicators and preset optimal performance target values, and in conjunction with process constraints, includes:

[0025] Define the objective function, constraints, and solver for the model predictive controller; the objective function minimizes the difference between the target fatigue life and the fatigue life value predicted by the performance prediction module, and includes a penalty term for the rate of change of the control variable.

[0026] The model predictive controller receives the current moment distribution of the principal axis from the dynamic process module, then starts the optimization solver to solve the objective function in the subsequent prediction time domain and performs rolling optimization to obtain the optimal control sequence that minimizes the objective function.

[0027] The model predictive controller extracts the first control instruction of the optimal control sequence and sends it to the actuator as the optimal program control instruction.

[0028] Furthermore, the actuator array includes a high-density digital quenching array and a segmented induction heating array; the high-density digital quenching array consists of multiple electrically controlled nozzles with independently controllable flow rates and on / off states; the optimal program control command defines the injection flow rate and timing of each nozzle at different times; the segmented induction heating array consists of multiple induction coil segments with independently controllable power, used to locally heat the spindle according to the optimal program control command after the quenching process, so as to actively adjust and optimize the stress distribution of the spindle.

[0029] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0030] 1. By constructing a digital twin model that couples the temperature field, phase transition field, and stress field, the complex physical behavior of the main shaft during heat treatment can be accurately simulated, significantly improving the accuracy of state distribution prediction. Acoustic feature data combined with optimization algorithms are used to correct the temperature distribution, phase ratio distribution, and stress distribution in real time, enhancing the model's adaptability to external disturbances and uncertainties and ensuring the reliability of the state distribution results. This digital twin model provides reliable input data for subsequent performance index prediction, thereby improving the effectiveness and reliability of high fatigue performance control of wind turbine main shafts.

[0031] 2. By extracting characteristic parameters from temperature distribution, phase ratio distribution, and stress distribution, the input data of the performance prediction module is ensured to fully characterize the spindle state, improving the reliability and comprehensiveness of performance prediction. A deep neural network is used to process the nonlinear relationship between characteristic parameters and fatigue performance. Historical data is used to train the network and optimize its parameters, thereby improving the prediction accuracy of fatigue performance indicators.

[0032] 3. By defining the optimization objective function of the model predictive controller and adding a penalty term for the rate of change of the control quantity, a smooth and efficient optimal control command can be generated, balancing performance optimization and process stability. By adopting a rolling optimization strategy, the model predictive controller can dynamically adjust the control command according to the current state distribution and the prediction time domain, thereby improving the response capability to dynamic changes in the production process and further improving the effectiveness and reliability of high fatigue performance control of the wind turbine main shaft. Attached Figure Description

[0033] Figure 1 This is a flowchart illustrating a method for controlling the high fatigue performance of a wind turbine main shaft according to the present invention.

[0034] Figure 2 This is a flowchart illustrating the process of obtaining the optimal program control instructions according to the present invention;

[0035] Figure 3 This is a schematic diagram of the actuator array of the present invention. Detailed Implementation

[0036] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0037] Please see Figures 1 to 3 This invention provides a method for controlling the high fatigue performance of wind turbine main shafts, the technical solution of which is as follows: Example

[0038] To improve the effective control of high fatigue performance of wind turbine main shafts during production, a manufacturer used a high fatigue performance control method for wind turbine main shafts proposed in this invention. The flowchart of this method is shown below. Figure 1 As shown, it includes:

[0039] The multi-source sensing parameters of the spindle during the heat treatment process are acquired, and the multi-source sensing parameters are input into the dynamic process module that couples the temperature field, phase transformation field and stress field. The state distribution results identified by the module are updated through the assimilation algorithm.

[0040] Furthermore, the multi-source sensing parameters include at least spindle surface temperature data and acoustic characteristic data;

[0041] Furthermore, thermal image data of the spindle surface is acquired by using an infrared thermal imager to obtain spindle surface temperature data; the sound wave signal acquired by the ultrasonic sensor is analyzed, the sound speed is obtained by calculating the signal flight time, and the sound attenuation coefficient is obtained by analyzing the attenuation of the signal amplitude, thus forming acoustic feature data;

[0042] Furthermore, multi-source sensing parameters can also include electromagnetic characteristic data, deformation characteristic data, etc.; among them, electromagnetic characteristic data is obtained through electromagnetic sensors, and deformation characteristic data can be acquired through visual sensors.

[0043] By combining sensing parameters such as surface temperature data and acoustic feature data, the dynamic state of the spindle during heat treatment can be captured more comprehensively, making up for the limitations of insufficient data from a single sensor and providing a data foundation for subsequent correction of the state distribution of temperature field, phase transition field and stress field.

[0044] Furthermore, the process of inputting multi-source sensing parameters into a dynamic process module that couples temperature field, phase transition field, and stress field, and updating the integrated distribution results identified by the module through an optimization algorithm includes:

[0045] The dynamic process module is constructed in the form of a digital twin model coupled with multi-physics fields;

[0046] The dynamic process module is initialized by loading the geometric model and material properties of the spindle;

[0047] The spindle surface temperature data from the multi-source sensing parameters is used as the thermal boundary condition of the dynamic process module. At the same time, the acoustic feature data from the multi-source sensing parameters are used, combined with optimization algorithms, to correct the temperature distribution, phase ratio distribution and stress distribution predicted by the dynamic process module.

[0048] Furthermore, the dynamic process module includes modeling of the temperature field, phase transition field, and stress field;

[0049] Furthermore, the temperature field is modeled based on the heat conduction equation, and thermal boundary conditions, such as convective heat transfer coefficient and ambient temperature, are defined by combining the principal axis surface temperature data from multi-source sensing parameters. A phase change dynamics model, such as the Kosin-Mell model or the JMAK model, is used to model the phase change field to describe the change in phase change ratio with time and temperature. The latent heat released by the phase change is fed back into the temperature field governing equations, and the volume change caused by the phase change also affects the stress field. The stress field model is decomposed into elastic strain, thermal strain, phase change strain, and plastic strain. Elastic strain follows Hooke's law; thermal strain is caused by temperature changes generated by the temperature field model; phase change strain represents the weighted sum of the phase change field ratios and strain tensors; and plastic strain is modeled based on the plastic flow law and yield criterion.

[0050] Furthermore, a multiphysics coupled finite element method is adopted for coupled solution, that is, the governing equations of temperature field, phase transition field and stress field are combined and the time stepping method is used for dynamic solution;

[0051] Furthermore, the process of correcting the temperature distribution, phase proportion distribution, and stress distribution predicted by the dynamic process module using optimization algorithms includes: training a codec using historical data, with acoustic feature data as input and the output being the internal state combination of the main axis, namely temperature, phase proportion, and stress; inputting the measured acoustic feature data into the pre-trained codec to obtain the inverted state values; simultaneously, the dynamic process module calculates the predicted state values ​​through finite element analysis; then, the inverted state values ​​and predicted values ​​are fused using a Kalman filter algorithm to obtain the optimal estimated state combination, thereby correcting the temperature distribution, phase proportion distribution, and stress distribution.

[0052] Furthermore, a multi-physics coupling model is combined with a microstructure evolution model to construct a multi-scale digital twin model, capturing physical behavior from the grain scale to the overall component scale. Specifically, in phase transformation field modeling, a crystal dynamics model is introduced to simulate microstructures, such as the nucleation and growth of martensite and bainite, and to predict grain size distribution. Then, a homogenization method is used to map microstructure characteristics, such as grain size and phase ratio, to the macroscopic stress field, updating plastic strain and yield criterion. Finally, a multi-scale coupling algorithm is used to simultaneously calculate microscopic and macroscopic field variables in the finite element solution. This improves the accuracy of the principal axis dynamic modeling, thereby improving the accuracy of subsequent fatigue performance prediction.

[0053] By constructing a digital twin model that couples the temperature field, phase transition field, and stress field, the complex physical behavior of the main shaft during heat treatment can be accurately simulated, improving the accuracy of state distribution prediction. Real-time correction of the state distribution is achieved using acoustic feature data combined with optimization algorithms, enhancing the model's adaptability to external disturbances and uncertainties. This digital twin model provides reliable input data for subsequent performance index prediction, thereby improving the effectiveness and reliability of high fatigue performance control of wind turbine main shafts.

[0054] Based on the state distribution results, the preset performance prediction module is used to process the data and predict the performance indicators used to characterize the spindle fatigue performance.

[0055] Furthermore, the performance prediction module is a proxy model built based on machine learning algorithms. The input of the proxy model is the feature parameters extracted from the distribution results output by the dynamic process module, and the output is the predicted fatigue life value of the key area of ​​the spindle. The predicted fatigue life value is used to characterize the fatigue performance of the spindle.

[0056] Machine learning algorithms can effectively capture the complex nonlinear relationship between multi-physics parameters and fatigue performance during heat treatment, improving the applicability and robustness of the prediction model. At the same time, the surrogate model can quickly process the complex distribution results output by the dynamic process module and generate predicted fatigue life values ​​for the key areas of the spindle, improving prediction efficiency and accuracy.

[0057] Furthermore, based on the overall condition, the process of using a preset performance prediction module to predict the performance indicators used to characterize the final fatigue performance of the spindle includes:

[0058] Extract characteristic parameters related to temperature distribution, phase ratio distribution, and stress distribution from the distribution results;

[0059] A deep neural network is used as a proxy model for performance prediction, comprising an input layer, a hidden layer, and an output layer. The input layer receives feature parameters, the hidden layer handles the nonlinear relationship between the feature parameters and performance indicators, and the output layer generates predicted fatigue life values ​​for the key region of the main axis.

[0060] The surrogate model is trained using historical data, and the prediction error is minimized through parameter optimization. The feature parameters are then input into the trained deep neural network to obtain the prediction performance metrics.

[0061] Furthermore, principal component analysis is used to reduce the dimensionality of the distributed data and extract the corresponding feature parameters, such as local temperature, maximum temperature gradient, martensite ratio, bainite ratio, residual stress, and stress gradient in stress concentration areas of the key region of the principal axis. Combined with Z-score standardization and Min-Max normalization, the feature parameters are standardized to the range of [0,1] to ensure consistency with the input of the deep neural network.

[0062] Furthermore, the deep neural network adopts a 3-layer hidden layer structure, with each layer containing neurons and a ReLU activation function. The number of neurons decreases by half with each layer, namely 64, 32 and 16. Among them, the first two layers introduce Dropout regularization to prevent overfitting, with the dropout rate set to 0.2.

[0063] Furthermore, the loss function used during deep neural network training includes the mean squared error between the predicted fatigue life and the actual fatigue life, as well as an L2 regularization term; the Adam optimizer is used, and the initial learning rate is set to 0.001;

[0064] Furthermore, in the feature extraction stage, a mutual information-based feature selection algorithm, such as ReliefF or Boruta, is used to dynamically select features that have the greatest impact on fatigue life prediction based on the current process stage. For example, in the quenching stage, temperature gradient and martensite ratio are given priority, while in the tempering stage, residual stress and grain size are given priority. This can enhance the model's adaptability to different process stages, improve prediction accuracy, and at the same time reduce redundant features and lower model processing time.

[0065] By extracting feature parameters from temperature distribution, phase ratio distribution, and stress distribution, the input data of the performance prediction module is ensured to fully characterize the spindle state, thus improving the reliability and comprehensiveness of performance prediction. A deep neural network is used to process the nonlinear relationship between feature parameters and fatigue performance, and the network parameters are optimized by training it with historical data, thereby improving the prediction accuracy of fatigue performance indicators.

[0066] A model predictive controller is used to solve and generate the optimal program control instructions for the heat treatment process based on the comparison results between real-time performance indicators and preset optimal performance target values, combined with process constraints.

[0067] Furthermore, the process of obtaining the optimal program control instructions is as follows: Figure 2 As shown, it includes:

[0068] Define the objective function, constraints, and solver for the model predictive controller; the objective function minimizes the difference between the target fatigue life and the fatigue life value predicted by the performance prediction module, and includes a penalty term for the rate of change of the control variable.

[0069] The model predictive controller receives the current moment distribution of the principal axis from the dynamic process module, then starts the optimization solver to solve the objective function in the subsequent prediction time domain and performs rolling optimization to obtain the optimal control sequence that minimizes the objective function.

[0070] The model predictive controller extracts the first control instruction of the optimal control sequence and sends it to the actuator as the optimal program control instruction.

[0071] Furthermore, the objective function consists of two parts. The first part is the square of the difference between the target fatigue life value and the fatigue life value predicted in real time by the performance prediction module. The second part is a penalty term for the rate of change of the control quantity, which is expressed as multiplying the square of the difference between the current control quantity and the control quantity at the previous moment by the weighting coefficient to prevent the control command from changing too drastically.

[0072] Furthermore, the process constraints include: the maximum temperature constraint to prevent quenching cracks, the maximum stress at the yield limit, and the physical limits of the actuator; the optimization solver uses a sequential quadratic programming algorithm for solving the problem.

[0073] Furthermore, the prediction time domain is set to 30 seconds, which can be adjusted according to actual needs and is not unique.

[0074] Furthermore, a fixed prediction time domain is not conducive to optimizing the response speed and stability of control commands. To address this, an adaptive prediction time domain mechanism is introduced. This mechanism uses a sliding window to analyze and calculate the statistical characteristics of state changes, such as variance, and then uses a pre-trained LSTM network to predict the optimal time domain length. During each rolling optimization, the time domain length is re-evaluated, and the prediction range of the sequential quadratic programming solver is updated to ensure that the control commands are dynamically matched with the process. This can improve the response speed and stability of control commands.

[0075] By defining the optimization objective function of the model predictive controller and adding a penalty term for the rate of change of the control quantity, a smooth and efficient optimal control command can be generated, balancing performance optimization and process stability. By adopting a rolling optimization strategy, the model predictive controller can dynamically adjust the control command according to the current state distribution and the prediction time domain, thereby improving the responsiveness to dynamic changes in the production process and further improving the effectiveness and reliability of high fatigue performance control of the wind turbine main shaft.

[0076] The optimal program control command is sent to the actuator array in the heat treatment equipment. The actuator array performs partitioned and time-controlled heat treatment on the spindle according to the command, so as to drive the real-time performance index to approach the optimal performance target value and perform optimization closed loop.

[0077] Furthermore, the structure of the actuator array can be referenced. Figure 3 It includes a high-density digital quenching array and a segmented induction heating array; the high-density digital quenching array consists of multiple electrically controlled nozzles whose flow rate and on / off status can be controlled independently; the optimal program control command defines the spray flow rate and timing of each nozzle at different times; the segmented induction heating array consists of multiple induction coil segments whose power can be controlled independently, which are used to locally heat the spindle according to the optimal program control command after the quenching process, so as to actively adjust and optimize the stress distribution of the spindle;

[0078] Furthermore, the controller of the high-density digital quenching array receives a spatiotemporal flux matrix from the model predictive controller. ,in It's the nozzle position. It is time; then, the matrix is ​​parsed into a real-time control curve of the flow rate of each independent valve over time, and the controller drives the valve to regulate the temperature according to the curve;

[0079] Furthermore, the controller of the segmented induction heating array receives control commands from the model predictive controller, including: area location, heating power, and heating time; then, it rapidly and locally heats the corresponding area of ​​the spindle to eliminate or transform harmful stress without affecting the overall fatigue performance.

[0080] By utilizing a high-density digital quenching array and a segmented induction heating array, precise control can be achieved in a timely manner based on optimal program control instructions, thereby improving the control accuracy of microstructure and stress distribution and further optimizing the fatigue performance of the spindle.

[0081] This embodiment proposes a control method for high fatigue performance of wind turbine main shafts. The method first acquires multi-source sensing parameters of the main shaft during heat treatment and inputs them into a dynamic process module coupled with temperature, phase transition, and stress fields. The state distribution results identified by the module are then updated using an assimilation algorithm. Next, a preset performance prediction module is used to predict performance indicators characterizing the fatigue performance of the main shaft. A model predictive controller is employed to generate optimal program control instructions for the heat treatment process based on a comparison between real-time performance indicators and preset optimal performance target values, combined with process constraints. These optimal program control instructions are then sent to the actuator array in the heat treatment equipment to perform zoned and time-controlled heat treatment of the main shaft, achieving adaptive optimization closed-loop. This invention improves the effectiveness and reliability of high fatigue performance control for wind turbine main shafts. Example

[0082] This embodiment takes a wind turbine main shaft manufacturing plant as an example to illustrate the application process of the high fatigue performance control method for wind turbine main shafts proposed in this invention.

[0083] The manufacturing process of this factory follows these regulations: the outer mold is made of metal mold casting, the metal mold material is QT400-18-LT, and the wall thickness of the metal mold is 1.0-2.0 times that of the product; the metal mold is heated to 60-120℃, and the time from mold assembly to pouring is controlled within 6 hours; ultra-high purity pig iron + scrap steel process is adopted, and the smelting process is held at 1500℃ or above for 10-20 minutes; the entire filling speed does not exceed 240s, and the flow rate of the inner gate does not exceed 30cm / s; after filling, the shaft is atomized and cooled, and the cooling time is 4-5 hours.

[0084] Multi-source sensor parameters are collected in real time during the heat treatment process at the wind turbine main shaft manufacturing plant; infrared thermal imager is used to obtain the surface temperature distribution of the main shaft, ultrasonic sensor is used to collect sound wave signals, the signal flight time is analyzed to calculate the sound speed, and the sound attenuation coefficient is calculated by the signal amplitude attenuation.

[0085] In addition, electromagnetic characteristic data and deformation characteristic data are collected to provide a comprehensive dataset to capture the dynamic state of the spindle during heat treatment.

[0086] Multi-source sensing parameters are input into a dynamic process module that couples temperature field, phase transition field and stress field. This module is constructed in the form of a digital twin model to simulate the complex physical behavior of the spindle during heat treatment.

[0087] The dynamic process module is initialized by loading the spindle geometry model and material properties; the spindle surface temperature data from the multi-source sensing parameters is used as the thermal boundary condition of the dynamic process module; the acoustic features are mapped to the internal state estimate using a pre-trained encoder-decoder, and the internal state estimate is fused with the finite element predicted state by combining the Kalman filter algorithm, thereby correcting the temperature, phase transition ratio and stress distribution predicted by the dynamic process module and obtaining the optimal state estimate.

[0088] Key features are extracted from the state distribution, including local temperature, maximum temperature gradient, martensite and bainite ratio, residual stress, and stress gradient in key regions. The features are normalized to the range of [0,1] using Z-score and Min-Max normalization. Based on the preprocessed feature data, the predicted fatigue life of the key regions of the spindle is output using the trained performance prediction module.

[0089] The model predictive controller receives the current state distribution of the dynamic process module, solves the optimization objective function using a sequential quadratic programming solver within a 30-second prediction time domain, performs rolling optimization, generates the optimal control sequence, and extracts the first control command of the sequence to send to the actuator.

[0090] The optimal control command is sent to the actuator array. The high-density digital quenching array adjusts the injection flow and timing according to the spatiotemporal flow matrix. The controller parses the matrix into the real-time control curve of each nozzle and drives the valve to perform precise temperature regulation. The segmented induction heating array performs local heating after quenching to adjust the stress distribution.

[0091] The high-density digital quenching array is set with 48 nozzles, a flow rate of 0.5-5L / min, and a spacing of 50mm. This array can reduce the spindle journal temperature from 800℃ to 200℃ in 180 seconds, with a cooling rate of 10℃ / s, a temperature gradient controlled at 50℃ / cm, and a martensite ratio of 85%. The internal state is then stabilized by slow cooling for 4-5 hours.

[0092] The segmented induction heating array is set with 12 coils and a power of 10-50kW. When the central region of the spindle is heated to 400°C for 20 seconds, the residual stress decreases from 350MPa to 150MPa and the stress gradient decreases from 50MPa / cm to 20MPa / cm.

[0093] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for controlling the high fatigue performance of a wind turbine main shaft, characterized in that, include: The multi-source sensing parameters of the spindle during the heat treatment process are acquired, and the multi-source sensing parameters are input into the dynamic process module that couples the temperature field, phase transformation field and stress field. The state distribution results identified by the module are updated through the assimilation algorithm. The dynamic process module combines a macroscopic multiphysics coupling model with a microstructure evolution model to construct a multi-scale digital twin model, capturing physical behavior from the grain scale to the overall component scale. Specifically, in phase transition field modeling, a crystal dynamics model is introduced to simulate microstructure and predict grain size distribution. Then, the microstructure characteristics are mapped to the macroscopic stress field through a homogenization method to update the plastic strain and yield criterion. Based on the state distribution results, the preset performance prediction module is used to process the data and predict the performance indicators used to characterize the spindle fatigue performance. In the feature extraction stage, a feature selection algorithm based on mutual information is used to dynamically select the features that have the greatest impact on fatigue life prediction according to the current process stage. A model predictive controller is used to solve and generate the optimal program control instructions for the heat treatment process based on the comparison results between real-time performance indicators and preset optimal performance target values, combined with process constraints. The model predictive controller receives the current moment's principal axis distribution results from the dynamic process module, then starts the optimization solver to solve the objective function in the subsequent prediction time domain and performs rolling optimization to obtain the optimal control sequence that minimizes the objective function. The objective function minimizes the difference between the target fatigue life and the fatigue life value predicted by the performance prediction module, and a penalty term for the rate of change of the control variable is added. The optimal program control command is sent to the actuator array in the heat treatment equipment. The actuator array performs partitioned and time-controlled heat treatment on the spindle according to the command, driving the real-time performance index to approach the optimal performance target value and performing optimization closed loop.

2. The method for controlling the high fatigue performance of a wind turbine main shaft according to claim 1, characterized in that, The multi-source sensing parameters include at least spindle surface temperature data and acoustic characteristic data.

3. The method for controlling the high fatigue performance of a wind turbine main shaft according to claim 1, characterized in that, The process of inputting the multi-source sensing parameters into a dynamic process module that couples the temperature field, phase transition field, and stress field, and updating the comprehensive distribution result identified by the module through an optimization algorithm includes: The dynamic process module is constructed in the form of a digital twin model coupled with multi-physics fields; The dynamic process module is initialized by loading the geometric model and material properties of the spindle; The spindle surface temperature data from the multi-source sensing parameters is used as the thermal boundary condition of the dynamic process module. At the same time, the acoustic feature data from the multi-source sensing parameters are used, combined with optimization algorithms, to correct the temperature distribution, phase ratio distribution, and stress distribution predicted by the dynamic process module.

4. The method for controlling the high fatigue performance of a wind turbine main shaft according to claim 1, characterized in that, The performance prediction module is a surrogate model built based on a machine learning algorithm. The input of the surrogate model is the feature parameters extracted from the distribution results output by the dynamic process module, and the output is the predicted fatigue life value of the key area of ​​the spindle. The predicted fatigue life value is used to characterize the fatigue performance of the spindle.

5. The method for controlling the high fatigue performance of a wind turbine main shaft according to claim 1, characterized in that, Based on the overall condition, the process of using a preset performance prediction module to predict the performance indicators used to characterize the final fatigue performance of the spindle includes: Extract characteristic parameters related to temperature distribution, phase ratio distribution, and stress distribution from the distribution results; A deep neural network is used as a proxy model for performance prediction, comprising an input layer, a hidden layer, and an output layer. The input layer receives feature parameters, the hidden layer handles the nonlinear relationship between the feature parameters and performance indicators, and the output layer generates predicted fatigue life values ​​for the key region of the main axis. The surrogate model is trained using historical data, and the prediction error is minimized through parameter optimization. The feature parameters are then input into the trained deep neural network to obtain the prediction performance metrics.

6. The method for controlling the high fatigue performance of a wind turbine main shaft according to claim 1, characterized in that, The process of using a model predictive controller to inversely solve for and generate the optimal program control instructions for the heat treatment process, based on the comparison between real-time performance indicators and preset optimal performance target values, and in conjunction with process constraints, includes: Define the objective function, constraints, and solver for the model predictive controller; the objective function minimizes the difference between the target fatigue life and the fatigue life value predicted by the performance prediction module, and includes a penalty term for the rate of change of the control variable. The model predictive controller receives the current moment distribution of the principal axis from the dynamic process module, then starts the optimization solver to solve the objective function in the subsequent prediction time domain and performs rolling optimization to obtain the optimal control sequence that minimizes the objective function. The model predictive controller extracts the first control instruction of the optimal control sequence and sends it to the actuator as the optimal program control instruction.

7. The method for controlling the high fatigue performance of a wind turbine main shaft according to claim 1, characterized in that, The actuator array includes a high-density digital quenching array and a segmented induction heating array. The high-density digital quenching array consists of multiple electrically controlled nozzles with independently controllable flow rates and on / off states. The optimal program control command defines the injection flow rate and timing of each nozzle at different times. The segmented induction heating array consists of multiple induction coil segments with independently controllable power, used to locally heat the spindle according to the optimal program control command after the quenching process, so as to actively adjust and optimize the stress distribution of the spindle.

Images