A load field reconstruction method and a measuring system of a direct drive wind turbine main shaft bearing

Abstract

The application discloses a load field reconstruction method and a measuring system of a direct-drive wind turbine main shaft bearing, and relates to the technical field of generators. The method comprises the following steps: arranging multiple groups of strain gauges on a preset strain sensitive section of the shafting, wherein the preset strain sensitive section is selected from at least one of the shaft wall of the fixed shaft, the shaft wall of the rotating shaft and a shafting stepped transition section; sequentially performing multi-stage calibration on the shafting arranged with the strain gauges to construct a digital twin agent model; in the operation of the unit, collecting real-time data of the strain gauges and inputting the digital twin agent model to reconstruct the load field of the main shaft bearing. The measuring system for the above method comprises a strain sensing unit, a state monitoring unit, a loading calibration unit and a data processing and reconstruction unit. The application can realize high-precision measurement of the actual operation load of the wind turbine, and provide quantitative basis for unit design verification and operation optimization.

Description

Technical Field

[0001] This invention discloses a load field reconstruction method and measurement system for the main shaft bearing of a direct-drive wind turbine, which relates to the field of generator technology. Background Technology

[0002] Currently, in the field of load measurement of main shaft bearings of direct-drive wind turbines, existing technologies are mainly divided into two core technical solutions, while there are also a variety of related auxiliary technologies, including blade root load measurement, bearing local contact load measurement, and other auxiliary technologies.

[0003] Blade root load measurement: The core idea of ​​this scheme is to attach strain gauges to the blade root, measure the strain response of the blade root section, and then indirectly calculate the load state of the main shaft bearing through a preset "strain-load" relationship model. For example, Chinese patent application CN117606830A discloses a method for measuring the on-site load of wind turbines based on a "strain-load" relationship model. This method achieves load correction by combining blade root strain with the unit's operating torque, thereby indirectly obtaining the bearing load.

[0004] The specific implementation process of this scheme is as follows: attach strain gauges to the blade root → collect blade root strain data → correct the strain data by combining it with the unit's operating torque → calculate the bearing load through a mapping model. This relies on the force transmission relationship between the blades and the shaft system. However, the transmission process from blade root load to bearing load is affected by multiple factors such as blade aeroelastic deformation and flapping-bounce coupling effects. The transmission link is long, with many intermediate links, resulting in significant load transmission errors and making it difficult to accurately reflect the true load state of the main shaft bearing.

[0005] Bearing local contact load measurement: The core idea of ​​this solution is to customize the bearing rollers and integrate strain sensors inside the hollow rollers to directly measure the contact strain of individual rollers, thereby identifying local contact loads.

[0006] The specific implementation process of this scheme is as follows: customizing hollow rollers and integrating sensors → assembling the modified rollers into the bearing → real-time acquisition of roller contact strain → calculation of local contact load. However, this scheme has the following inherent defects: on the one hand, the calibration process of the intelligent rollers is extremely complex, requiring the establishment of a precise mapping relationship between the internal strain of the rollers and the external contact load, and the calibration process has stringent requirements; on the other hand, in actual operation, bearing rollers experience slippage, resulting in uncertain roller motion states, making data processing and analysis extremely difficult, and severely restricting its engineering practicality. In addition, this scheme can only measure the local contact stress of a single roller and cannot reconstruct the overall load field of the unit.

[0007] Other auxiliary technologies: (1) Full-field strain measurement technology: Three-dimensional point tracking technology combined with finite element method is used to realize full-field strain measurement of the rotating structure of wind turbine, but it relies on optical target points and is only suitable for laboratory testing and cannot be adapted to the long-term operation environment at sea. (2) Inverse load reconstruction technology: Based on augmented impulse response matrix, wind load inverse reconstruction is realized. It requires complete shaft system modal parameters and has not solved the calibration error problem. The error is large in actual application. (3) Inverse finite element method (iFEM): It is mainly used for deformation / strain field reconstruction of wind turbine blades and towers, and does not involve multi-stage calibration and load field reconstruction of bearing system. (4) Digital twin virtual sensor technology: The existing load estimation is based on SCADA and CMS data. The model training relies on simulation data and does not adopt step-by-step actual measurement calibration. It is applicable to gearbox bearings and not to direct drive unit main shaft bearings. (5) Strain field / stress field reconstruction technology: The existing stress field reconstruction method based on gap POD adopts the idea of ​​"finite element prior + actual measurement correction", but does not involve multi-stage calibration and state validity determination of wind turbine bearings.

[0008] Common defects of existing technologies: (1) Low calibration accuracy: Existing technologies mostly adopt single zero-position calibration, ignoring thermal strain interference, assembly prestress, and gravity stress differences caused by the difference in attitude between the factory and the field (horizontal assembly vs. vertical operation), resulting in significant systematic errors in the "strain-load" mapping model; (2) Poor long-term reliability: The degradation of bearings after long-term operation, such as raceway wear and increased clearance, is not considered, and the change in force transmission path causes the original mapping model to fail, resulting in distorted load reconstruction results; (3) Limited measurement range: The blade root load measurement scheme has load transmission errors and cannot directly reflect the actual load on the bearing; the intelligent roller scheme can only measure local contact stress and cannot realize the overall load field reconstruction of the unit; (4) Implementation limitations: The intelligent roller scheme requires modification of the core components of the bearing, which is demanding and costly to implement, making it difficult to scale up; some technologies are only applicable to the laboratory and cannot adapt to the complex operating environment at sea. Summary of the Invention

[0009] This invention addresses the problems in existing technologies by providing a load field reconstruction method and measurement system for the main shaft bearings of direct-drive wind turbines. Applicable to large-capacity offshore direct-drive wind turbines, and considering the characteristics of large-capacity main shaft bearings, significant shaft deformation, and strong gravity-stress coupling effects, this invention enables high-precision measurement of the actual operating load of the wind turbine, providing quantitative data for turbine design verification and operation and maintenance optimization. It also provides load input references for bearing fatigue life analysis, supporting the full life-cycle reliability management of offshore wind power equipment. This invention is applicable to main shaft bearing structures with different rotational configurations, including two typical configurations: inner ring rotation configuration (inner shaft, outer fixed shaft) and outer ring rotation configuration (outer shaft, inner fixed shaft). Both the multi-stage calibration method and the load field reconstruction method of this invention are applicable.

[0010] To achieve the above-mentioned objectives, the technical solution of the present invention is as follows: A load field reconstruction method for a direct-drive wind turbine main shaft bearing, wherein the shaft system of the main shaft bearing includes a rotating shaft that rotates with the wind turbine and a non-rotating fixed shaft; the method includes the following steps: Step S1: Arrange multiple sets of strain gauges in the preset strain-sensitive section of the shaft system. The preset strain-sensitive section is selected from at least one of the shaft wall of the fixed shaft, the shaft wall of the rotating shaft, and the stepped transition section of the shaft system. Step S2: Perform multi-stage calibration sequentially on the shaft system with the strain gauges arranged thereon to construct a digital twin proxy model; Step S3: During unit operation, collect real-time data from the strain gauges and input them into the digital twin proxy model to reconstruct the load field of the main shaft bearing.

[0011] Preferably, in step S2, the multi-stage calibration includes: Step S21, First stage constant temperature zero-point calibration: Before assembling the bearing components, the strain gauges are pasted, cured and calibrated to zero stress state in a constant temperature environment; Step S22, Second stage post-assembly state calibration: After the shaft system is assembled, the strain gauge output is collected and updated to the measurement reference value to eliminate the error introduced by the assembly prestress and attitude gravity stress; Step S23, Third Stage Lateral Loading Calibration: Apply multiple known gradient lateral loads to the assembled shaft system, collect the strain response under each load, and fuse the physical prior data obtained through finite element simulation with the strain response data. Train the digital twin proxy model through physical-data co-training, wherein the digital twin proxy model uses a Bayesian regularization algorithm to establish an inverse mapping relationship from strain to load.

[0012] Preferably, in step S21, before assembling the bearing components, the bearing components are placed in a constant temperature environment, and the ambient temperature is controlled to be at room temperature. The temperature difference between the fixed shaft and the rotating shaft is ensured to be within a preset temperature accuracy range to eliminate thermal output interference from the resistance strain gauges. Because strain gauges have significant thermal output characteristics, temperature fluctuations within the preset temperature accuracy range can control thermal strain interference to a very small extent, far less than the working strain of the shaft system, thus eliminating thermal strain interference at its source. Under this constant temperature condition, the strain gauges are pasted, cured, and wired. During pasting, the pasting angle deviation of the strain gauges is strictly controlled to ensure the directional accuracy of strain measurement. After the sensor state stabilizes, the output of all strain gauges is calibrated to a zero-stress state, establishing an initial zero-baseline to eliminate the initial error caused by thermal strain.

[0013] Preferably, in step S22, all assembly work is completed in a vertical state according to the factory assembly process. After assembly, the shaft system needs to be adjusted to a working posture consistent with the on-site operation—that is, the axis is at a certain angle to the horizontal plane, which is the wind turbine elevation angle. During this posture change, the gravitational stress field of the shaft system will change. After assembly, the shaft system is left to stand for a preset time until the assembly stress is completely stable. The output data of the current strain gauge is collected, and a finite element simulation model of the shaft system's gravitational stress field under this posture is constructed to calculate the theoretical strain value of each measuring point. The measured strain and the theoretical strain are checked for deviation to evaluate the working state of the strain gauge. If the deviation exceeds a preset threshold, it is determined that the strain gauge has a bonding failure or a circuit fault. The entire shaft system needs to be disassembled and reassembled. Before assembly, the bonding of the strain gauge and the zero-position calibration in step S21 are completed again. Then, the assembly and calibration of this stage are performed again. At the same time, the strain value after assembly is used as a new measurement benchmark to eliminate the interference caused by the assembly prestress and the posture gravitational stress.

[0014] Preferably, in step S23, the assembled motor unit is hoisted onto a vertical test stand in the factory and adjusted to a vertical posture consistent with the on-site operation; a gradient lateral load is applied to the shaft system through a closed-loop hydraulic loading system, covering multiple typical load levels from no-load to ultimate load; under each load gradient, after the system strain response is completely stable, strain data for a preset duration is collected and averaged to eliminate vibration interference and ensure data stability; before formally training the surrogate model, the strain measurement system whose accuracy has been confirmed in step S22 is first loaded and verified: several verification load points are extracted from the gradient lateral load, and the measured strain values ​​under each verification load are compared with the theoretical strain values ​​calculated by finite element simulation; if the deviations are both within the preset accuracy range, the strain measurement system is confirmed to be working normally within the full range, and the surrogate model training stage can be entered; if the deviation exceeds the standard, the sensor status needs to be checked and the loading verification is re-executed; after confirming that the strain measurement accuracy is qualified, the digital twin strain-load coupled surrogate model of the shaft system is trained by combining the input value of the loaded load.

[0015] Preferably, the digital twin strain-load coupled proxy model adopts a physics-data co-training approach. Specifically, the implementation involves: first, constructing a full-size three-dimensional finite element model of the shaft system based on ANSYS Workbench, using Solid186 hexahedral high-order elements for mesh generation, and obtaining strain-load correspondences under 100 different load conditions as prior physical data through simulation analysis; then, using measured strain-load data collected during the lateral load loading process under seven gradient loads as experimental data, fusing the prior physical data and experimental data at a 3:1 ratio, and employing a Bayesian regularization algorithm for model training to balance physical accuracy and experimental precision; simultaneously, recording the initial baseline values ​​of the displacement sensor and bolt stress sensor under this state to provide a benchmark for subsequent state determination.

[0016] Preferably, in step S3, when the strain data is determined to be valid, the load field reconstruction and evaluation step is performed: Step S31: Collect real-time multi-source strain data of strain gauges under the current operating state. The sampling frequency meets the requirements of dynamic load capture and can capture changes in dynamic load. Step S32: Input the real-time strain data into the digital twin proxy model established in the calibration stage, and reconstruct the actual input load vector of the current unit from the strain response through the Bayesian regularized inverse mapping solution algorithm. The actual input load vector includes radial load, axial load and overturning moment. Step S33: Compare the actual load obtained from the reconstruction with the input load in the design stage to quantitatively evaluate the margin and accuracy of the design load, and provide a basis for structural optimization and life assessment. Step S34: Introduce a deep learning-based intelligent operating condition matching module, using wind speed, wind direction, rotational speed, and output power as input features to automatically identify the current operating condition. The intelligent operating condition matching module adopts a lightweight network and uses a large amount of historical operating data of the same model of unit during training, which can accurately identify a variety of typical operating conditions. For different operating conditions, the corresponding parameterized wind load input model is called to dynamically correct the digital twin proxy model.

[0017] Preferably, the system further includes step S4, a step for determining the validity of the operating status: In step S23, the initial baseline values ​​of the non-contact displacement sensor and the ultrasonic bolt stress sensor are recorded synchronously; after the unit starts operating, the current values ​​of the displacement sensor and the bolt stress sensor are periodically collected and compared with the corresponding initial baseline values ​​to calculate the displacement deviation rate and the stress deviation rate; when both the displacement deviation rate and the stress deviation rate are less than a preset threshold, the digital twin proxy model is determined to be valid, and step S3 is executed; otherwise, the model is determined to be invalid and an early warning is triggered.

[0018] Preferably, when the digital twin agent model is determined to be faulty, if the bearing condition is confirmed to be within the serviceable range, the strain data generated by the shaft system under pure gravity conditions in the stationary state of shutdown is used to perform on-site recalibration of the model in an online incremental update manner without disassembly.

[0019] Preferably, the method also includes an adaptive operating condition step: introducing a deep learning-based operating condition identification module, using the real-time wind speed, wind direction, rotational speed and output power of the unit as input, to automatically identify the current operating condition; calling the parameterized wind load input model corresponding to the current operating condition to dynamically correct the digital twin proxy model, so as to improve the reconstruction accuracy under complex and changing operating conditions.

[0020] Preferably, the process also includes an ultimate load capture step: during unit operation, the unit operating parameters are continuously monitored, and when an extreme operating condition exceeding a preset cut-out wind speed or preset speed threshold is identified, a high-frequency acquisition mode is triggered to record the strain response at the corresponding moment; based on the recorded strain response, ultimate load data is obtained by reconstructing the data through the digital twin proxy model.

[0021] A direct-drive wind turbine main shaft bearing load measurement system, used to implement the above method, includes: The strain sensing unit includes the multiple sets of strain gauges; The condition monitoring unit includes a non-contact displacement sensor arranged between the bearing retaining ring and the mating surface of the fixed axis step, and an ultrasonic bolt stress sensor arranged on the bearing retaining ring fastening bolt. The loading calibration unit includes an in-plant vertical test stand and a closed-loop hydraulic loading system for applying the gradient lateral load; The data processing and reconstruction unit is configured to perform the multi-stage calibration to construct the digital twin agent model, perform the validity determination of the running state, and reconstruct the load field based on the digital twin agent model.

[0022] Preferably, multiple sets of resistance strain gauges are arranged in a preset strain-sensitive section of the main shaft bearing system. The preset strain-sensitive section includes at least one selected from the fixed shaft wall, the rotating shaft wall, and the stepped transition section of the shaft system corresponding to the bearing location. The fixed shaft refers to a fixed shaft that does not rotate with the wind turbine, and the rotating shaft refers to a rotating shaft that rotates with the wind turbine. In the inner ring rotation configuration, the fixed shaft is the outer shaft and the rotating shaft is the inner shaft; in the outer ring rotation configuration, the fixed shaft is the inner shaft and the rotating shaft is the outer shaft. The multiple sets of resistance strain gauges are used to measure the elastic strain response of the shaft system under external loads, realizing the fusion acquisition of multi-source strain data. The multiple sets of resistance strain gauges are high-precision foil strain gauges.

[0023] Preferably, the non-contact displacement sensor is a high-precision eddy current displacement sensor, which is arranged between the mating surfaces of the bearing pressure ring and the fixed-axis step to monitor the relative displacement changes between the two and indirectly characterize the evolution of the bearing clearance; the ultrasonic bolt stress sensor is arranged on the bearing pressure ring fastening bolt to monitor the evolution of the bolt preload stress and indirectly characterize whether the bearing clearance has changed significantly.

[0024] Preferably, the loading calibration unit is used to apply a controllable gradient lateral load to the shaft system during the calibration phase to simulate different loading conditions and provide labeled data for the training of the digital twin surrogate model.

[0025] Preferably, the data processing and reconstruction unit includes a storage module, a state determination module, and a load reconstruction module, wherein the storage module is used to store baseline data during the calibration phase; the state determination module is used to perform operational state validity determination; and the load reconstruction module is used to perform load field reconstruction calculations and to compare and evaluate the actual load with the design input load.

[0026] The beneficial effects of this invention are: I. The present invention provides a load field reconstruction method for the main shaft bearing of a direct-drive wind turbine, which adopts multi-stage step-by-step calibration to eliminate multi-source interference: The three-stage calibration method proposed in this invention eliminates multi-source system errors such as thermal strain interference, assembly prestress, and attitude gravity stress layer by layer, solves the problem of low accuracy of existing single calibration, and significantly reduces the overall calibration error compared with existing single calibration methods, laying the foundation for high-precision load field reconstruction.

[0027] II. The present invention provides a load field reconstruction method for the main shaft bearing of a direct-drive wind turbine, which uses multi-sensor joint judgment to ensure long-term reliability: The present invention realizes online sensing of bearing status through joint monitoring of displacement and bolt stress, which can verify the validity of strain data in real time, avoids the problem of mapping spectrum failure caused by bearing wear, solves the pain point of load field reconstruction distortion after long-term operation, and ensures the reliability of measurement results throughout the entire life cycle.

[0028] III. The present invention provides a load field reconstruction method for the main shaft bearing of a direct-drive wind turbine, which supports design verification by reconstructing the overall load field. The present invention achieves the reconstruction of the overall input load field of the unit through multi-source strain measurement of the entire shaft system, rather than the local contact stress of a single roller. It can directly provide designers with a quantitative comparison between the actual load and the design load, fills the technical gap in the measurement of the overall load of the shaft system of direct-drive units, and provides core support for the design optimization and reliability improvement of the unit.

[0029] IV. The load field reconstruction method for the main shaft bearing of a direct-drive wind turbine provided by this invention is low-cost, easy to implement, and suitable for large-scale applications: This invention does not require modification of the core components of the bearing, but only requires the placement of sensors outside the shaft system. The modification cost is significantly lower than the existing intelligent roller solution, and the implementation difficulty is low. It does not require customized components, is easy to modify and upgrade on existing units, and can be promoted and applied on a large scale.

[0030] V. The present invention provides a load field reconstruction method for the main shaft bearing of a direct-drive wind turbine, which is universally adaptable and covers all types of turbine models: it can simultaneously adapt to the two mainstream direct-drive main shaft bearing structures of inner ring rotation and outer ring rotation, without the need to adjust the calibration and reconstruction algorithm for different configurations, which greatly reduces the technical adaptation cost and can cover the vast majority of mainstream direct-drive wind turbine models.

[0031] VI. The present invention provides a load field reconstruction method for the main shaft bearing of a direct-drive wind turbine, which extends the full range of measurement and covers high and low load conditions: it can accurately capture ultimate load data under extreme wind conditions, providing experimental basis for optimizing the typhoon resistance performance of large-capacity units; it can also improve the measurement sensitivity of small loads by selecting strain-sensitive measurement points, solving the problem of large measurement errors under low wind speed conditions, and effectively extending the lower limit of the effective range of load measurement.

[0032] VII. The present invention provides a load field reconstruction method for the main shaft bearing of a direct-drive wind turbine, which supports the entire life cycle and enables predictive operation and maintenance: Based on the real load data of the entire life cycle obtained by reconstruction, the bearing fatigue life can be accurately calculated and the remaining life can be predicted, replacing the traditional conservative estimation based on the design load. This effectively supports the predictive operation and maintenance of wind farms, can detect the early degradation risk of bearings in advance, avoid unplanned downtime, and significantly improve the operating income of wind farms. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the sensor arrangement of the inner ring rotating shaft system in this invention; Figure 2 This is a schematic diagram of the sensor arrangement of the outer ring rotating shaft system in this invention; Figure 3 This is a schematic diagram of the mesh generation of the simulation model of the motor and shaft system with the inner ring rotating configuration in this invention; Figure 4 This is a schematic diagram of the simulated external load profile of the inner ring rotation configuration in this invention; Figure 5 The simulation results of the fixed-axis stress of the inner ring rotation configuration in this invention; Figure 6 The simulation results of the rotation axis stress of the inner ring rotation configuration in this invention; Figure 7 This is a schematic diagram of the mesh generation of the simulation model of the motor and shaft system with the outer ring rotating configuration in this invention; Figure 8 This is a schematic diagram of the simulated external load profile of the outer ring rotation configuration in this invention; Figure 9 The simulation results of the fixed-axis stress of the outer ring rotation configuration in this invention; Figure 10 The results are simulation results of the rotational stress of the outer ring rotation configuration in this invention. Detailed Implementation

[0034] The present invention will be further described in detail below with reference to embodiments, but the implementation of the present invention is not limited thereto.

[0035] Example 1 This embodiment provides a load field reconstruction method for the main shaft bearing of a direct-drive wind turbine, wherein the shaft system of the main shaft bearing includes a rotating shaft that rotates with the wind turbine and a non-rotating fixed shaft; the method includes the following steps: Step S1: Arrange multiple sets of strain gauges in the preset strain-sensitive section of the shaft system. The preset strain-sensitive section is selected from at least one of the shaft wall of the fixed shaft, the shaft wall of the rotating shaft, and the stepped transition section of the shaft system. Step S2: Perform multi-stage calibration sequentially on the shaft system with the strain gauges arranged thereon to construct a digital twin proxy model; Step S3: During unit operation, collect real-time data from the strain gauges and input them into the digital twin proxy model to reconstruct the load field of the main shaft bearing.

[0036] Example 2 This embodiment provides a load field reconstruction method for the main shaft bearing of a direct-drive wind turbine, wherein the shaft system of the main shaft bearing includes a rotating shaft that rotates with the wind turbine and a non-rotating fixed shaft; the method includes the following steps: Step S1: Arrange multiple sets of strain gauges in the preset strain-sensitive section of the shaft system. The preset strain-sensitive section is selected from at least one of the shaft wall of the fixed shaft, the shaft wall of the rotating shaft, and the stepped transition section of the shaft system. Step S2: Perform multi-stage calibration sequentially on the shaft system with the strain gauges arranged thereon to construct a digital twin proxy model; Step S3: During unit operation, collect real-time data from the strain gauges and input them into the digital twin proxy model to reconstruct the load field of the main shaft bearing.

[0037] In step S2, the multi-stage calibration includes: Step S21, First stage constant temperature zero-point calibration: Before assembling the bearing components, the strain gauges are pasted, cured and calibrated to zero stress state in a constant temperature environment; Step S22, Second stage post-assembly state calibration: After the shaft system is assembled, the strain gauge output is collected and updated to the measurement reference value to eliminate the error introduced by the assembly prestress and attitude gravity stress; Step S23, Third Stage Lateral Loading Calibration: Apply multiple known gradient lateral loads to the assembled shaft system, collect the strain response under each load, and fuse the physical prior data obtained through finite element simulation with the strain response data. Train the digital twin proxy model through physical-data co-training, wherein the digital twin proxy model uses a Bayesian regularization algorithm to establish an inverse mapping relationship from strain to load.

[0038] In step S21, before assembling the bearing components, the bearing components are placed in a constant temperature environment. The ambient temperature is controlled to be at room temperature, and the temperature difference between the fixed shaft and the rotating shaft is ensured to be within a preset temperature accuracy range to eliminate thermal output interference from the resistance strain gauges. Due to the significant thermal output characteristics of the strain gauges, temperature fluctuations within the preset temperature accuracy range can control the interference of thermal strain to a very small range, much smaller than the working strain of the shaft system, thus eliminating the interference of thermal strain at the source. Under this constant temperature condition, the bonding, curing, and wiring of the strain gauges are completed. During the bonding process, the bonding angle deviation of the strain gauges is strictly controlled to ensure the directional accuracy of the strain measurement. After the sensor state stabilizes, the output of all strain gauges is calibrated to a zero stress state to establish an initial zero baseline and eliminate the initial error caused by thermal strain.

[0039] In step S22, all assembly work is completed in a vertical state according to the factory assembly process. After assembly, the shaft system needs to be adjusted to a working posture consistent with the on-site operation—that is, the axis is at a certain angle to the horizontal plane, which is the wind turbine elevation angle. During this posture change, the gravitational stress field of the shaft system will change. After assembly, the shaft system is left to stand for a preset time until the assembly stress is completely stable. The output data of the current strain gauge is collected, and a finite element simulation model of the shaft system's gravitational stress field under this posture is constructed to calculate the theoretical strain value of each measuring point. The measured strain and theoretical strain are checked for deviation to evaluate the working state of the strain gauge. If the deviation exceeds a preset threshold, it is determined that the strain gauge has a bonding failure or a circuit fault. The entire shaft system needs to be disassembled and reassembled. Before reassembly, the strain gauge bonding and zero-position calibration in step S21 are completed again. Then, the assembly and calibration of this stage are re-executed. At the same time, the strain value after assembly is used as a new measurement benchmark to eliminate the interference caused by the assembly prestress and the posture gravity stress.

[0040] In step S23, the assembled motor unit is hoisted onto a vertical test stand in the factory and adjusted to a vertical posture consistent with on-site operation. A closed-loop hydraulic loading system is used to apply gradient lateral loads to the shaft system, covering multiple typical load levels from no-load to ultimate load. Under each load gradient, after the system strain response has fully stabilized, strain data for a preset duration is collected and averaged to eliminate vibration interference and ensure data stability. Before formally training the surrogate model, the strain measurement system whose accuracy has been confirmed in step S22 is first verified by loading: several verification load points are extracted from the gradient lateral loads, and the measured strain values ​​under each verification load are compared with the theoretical strain values ​​calculated by finite element simulation. If the deviations are both within the preset accuracy range, the strain measurement system is confirmed to be working normally across the entire range, and the surrogate model training phase can begin. If the deviation exceeds the standard, the sensor status needs to be checked and the loading verification re-executed. After confirming that the strain measurement accuracy is qualified, a digital twin strain-load coupled surrogate model of the shaft system is trained by combining the input values ​​of the loaded loads.

[0041] The digital twin strain-load coupled proxy model employs a physics-data co-training approach. Specifically, it is implemented as follows: First, a full-size three-dimensional finite element model of the shaft system is constructed using ANSYS Workbench, meshed with Solid186 hexahedral high-order elements. Strain-load correspondences under 100 different load conditions are obtained through simulation analysis as prior physical data. Then, measured strain-load data collected during the lateral load loading process under seven gradient loads are used as experimental data. The prior physical data and experimental data are fused at a 3:1 ratio, and a Bayesian regularization algorithm is used for model training, balancing physical accuracy and experimental precision. Simultaneously, the initial baseline values ​​of the displacement sensor and bolt stress sensor are recorded under this state to provide a benchmark for subsequent state determination.

[0042] In step S3, when the strain data is determined to be valid, the load field reconstruction and evaluation steps are performed: Step S31: Collect real-time multi-source strain data of strain gauges under the current operating state. The sampling frequency meets the requirements of dynamic load capture and can capture changes in dynamic load. Step S32: Input the real-time strain data into the digital twin proxy model established in the calibration stage, and reconstruct the actual input load vector of the current unit from the strain response through the Bayesian regularized inverse mapping solution algorithm. The actual input load vector includes radial load, axial load and overturning moment. Step S33: Compare the actual load obtained from the reconstruction with the input load in the design stage to quantitatively evaluate the margin and accuracy of the design load, and provide a basis for structural optimization and life assessment. Step S34: Introduce a deep learning-based intelligent operating condition matching module, using wind speed, wind direction, rotational speed, and output power as input features to automatically identify the current operating condition. The intelligent operating condition matching module adopts a lightweight network and uses a large amount of historical operating data of the same model of unit during training, which can accurately identify a variety of typical operating conditions. For different operating conditions, the corresponding parameterized wind load input model is called to dynamically correct the digital twin proxy model.

[0043] The process also includes step S4, a step to determine the validity of the operating status: In step S23, the initial baseline values ​​of the non-contact displacement sensor and the ultrasonic bolt stress sensor are recorded synchronously; after the unit starts operating, the current values ​​of the displacement sensor and the bolt stress sensor are periodically collected and compared with the corresponding initial baseline values ​​to calculate the displacement deviation rate and the stress deviation rate; when both the displacement deviation rate and the stress deviation rate are less than a preset threshold, the digital twin proxy model is determined to be valid, and step S3 is executed; otherwise, the model is determined to be invalid and an early warning is triggered.

[0044] When the digital twin agent model is determined to be faulty, if the bearing condition is confirmed to be within the serviceable range, the strain data generated by the shaft system under pure gravity conditions in the stationary state of shutdown is used to perform on-site recalibration of the model in an online incremental update manner without disassembly.

[0045] This also includes an adaptive operating condition step: introducing a deep learning-based operating condition identification module, using the real-time wind speed, wind direction, rotational speed, and output power of the unit as inputs to automatically identify the current operating condition; calling the parameterized wind load input model corresponding to the current operating condition to dynamically correct the digital twin proxy model, so as to improve the reconstruction accuracy under complex and changing operating conditions.

[0046] This includes an ultimate load capture step: during unit operation, the unit's operating parameters are continuously monitored, and when an extreme operating condition exceeding a preset cut-out wind speed or preset speed threshold is identified, a high-frequency acquisition mode is triggered to record the strain response at the corresponding moment; based on the recorded strain response, ultimate load data is obtained by reconstructing the data through the digital twin proxy model.

[0047] Example 3 This embodiment provides a direct-drive wind turbine main shaft bearing load measurement system for implementing the method of Embodiment 1 or Embodiment 2, including: The strain sensing unit includes the multiple sets of strain gauges; The condition monitoring unit includes a non-contact displacement sensor arranged between the bearing retaining ring and the mating surface of the fixed axis step, and an ultrasonic bolt stress sensor arranged on the bearing retaining ring fastening bolt. The loading calibration unit includes an in-plant vertical test stand and a closed-loop hydraulic loading system for applying the gradient lateral load; the closed-loop hydraulic loading system simulates the load through hydraulic loading. The data processing and reconstruction unit is configured to perform the multi-stage calibration to construct the digital twin agent model, perform the validity determination of the running state, and reconstruct the load field based on the digital twin agent model.

[0048] Multiple sets of resistance strain gauges are arranged in a preset strain-sensitive section of the main shaft bearing system. The preset strain-sensitive section includes at least one selected from the fixed shaft wall, the rotating shaft wall, and the stepped transition section of the shaft system corresponding to the bearing position. The fixed shaft refers to a fixed shaft that does not rotate with the wind turbine, and the rotating shaft refers to a rotating shaft that rotates with the wind turbine. In the inner ring rotation configuration, the fixed shaft is the outer shaft and the rotating shaft is the inner shaft; in the outer ring rotation configuration, the fixed shaft is the inner shaft and the rotating shaft is the outer shaft. The multiple sets of resistance strain gauges are used to measure the elastic strain response of the shaft system under external loads, realizing the fusion acquisition of multi-source strain data. The multiple sets of resistance strain gauges are high-precision foil strain gauges.

[0049] The non-contact displacement sensor is a high-precision eddy current displacement sensor, which is arranged between the mating surfaces of the bearing pressure ring and the fixed-axis step to monitor the relative displacement changes between the two and indirectly characterize the evolution of the bearing clearance; the ultrasonic bolt stress sensor is arranged on the fastening bolt of the bearing pressure ring to monitor the evolution of the bolt preload stress and indirectly characterize whether the bearing clearance has changed significantly.

[0050] The loading calibration unit is used to apply a controllable gradient lateral load to the shaft system during the calibration phase to simulate different loading conditions and provide labeled data for the training of the digital twin agent model.

[0051] The data processing and reconstruction unit includes a storage module, a state determination module, and a load reconstruction module. The storage module is used to store baseline data from the calibration phase; the state determination module is used to determine the validity of the operating state; and the load reconstruction module is used to perform load field reconstruction calculations and to compare and evaluate the actual load with the design input load.

[0052] Example 4 A direct-drive wind turbine main shaft bearing load measurement system includes the following functional units: (1) Strain sensing unit: includes multiple sets of resistance strain gauges, arranged in a preset strain-sensitive section of the main shaft bearing system. The preset strain-sensitive section includes at least one selected from the fixed shaft wall, the rotating shaft wall, and the stepped transition section of the shaft system corresponding to the bearing position; wherein, the fixed shaft refers to a fixed shaft that does not rotate with the wind turbine, and the rotating shaft refers to a rotating shaft that rotates with the wind turbine. In the inner ring rotation configuration, the fixed shaft is the outer shaft and the rotating shaft is the inner shaft (e.g., Figure 1 As shown), in the outer ring rotation configuration, the fixed axis is the inner axis and the rotation axis is the outer axis (as shown). Figure 2 As shown, this unit is used to measure the elastic strain response of a shaft system under external loads, enabling the fusion acquisition of multi-source strain data. The strain gauge used in this unit is a high-precision foil strain gauge, capable of achieving high-precision measurement at the micro-strain level, meeting the measurement requirements for small strains in shaft systems. (2) Condition monitoring unit: includes a non-contact displacement sensor and an ultrasonic bolt stress sensor. The displacement sensor is a high-precision eddy current displacement sensor, which is arranged between the mating surfaces of the bearing pressure ring and the fixed-axis step to monitor the relative displacement changes between the two, indirectly characterizing the evolution of bearing clearance. The ultrasonic bolt stress sensor is arranged on the fastening bolt of the bearing pressure ring to monitor the evolution of bolt preload stress, indirectly characterizing whether the bearing clearance has changed significantly. The displacement sensor directly monitors the bearing clearance change in the radial dimension, while the ultrasonic bolt stress sensor indirectly reflects the clearance change in the axial preload dimension. The two form a complementary verification mechanism to avoid misjudgment based on a single parameter. (3) Loading calibration unit: including an in-plant vertical test stand and a closed-loop hydraulic loading system, used to apply controllable gradient lateral loads to the shaft system during the calibration stage to simulate different loading conditions and provide labeled data for the training of the digital twin surrogate model. The closed-loop hydraulic loading system has high loading accuracy and can achieve high-precision loading across the entire range, ensuring the accuracy of the calibration data; (4) Data processing and reconstruction unit: It is equipped with a storage module, a status determination module and a load reconstruction module. The storage module is used to store the baseline data of the calibration stage; the status determination module is used to perform the validity determination of the operating status; the load reconstruction module is used to perform load field reconstruction calculation and realize the comparison and evaluation of the actual load and the design input load. This module has real-time calculation capability and can realize real-time load field reconstruction to meet the needs of real-time monitoring.

[0053] Multi-stage step-by-step calibration method: This invention adopts a three-stage step-by-step calibration method to eliminate multi-source interference layer by layer and establish an accurate digital twin strain-load coupling surrogate model. The specific steps are as follows: (1) Initial zero-point calibration under constant temperature environment: Before assembling the bearing components, the bearing components are placed in a constant temperature environment, and the ambient temperature is controlled at room temperature. It is also ensured that the temperature difference between the fixed shaft and the rotating shaft is within the preset temperature accuracy range to eliminate the thermal output interference of the resistance strain gauge. Due to the significant thermal output characteristics of the strain gauge, the temperature fluctuation within the preset temperature accuracy range can control the interference of thermal strain to a very small range, much smaller than the working strain of the shaft system, thus eliminating the interference of thermal strain from the source. Under this constant temperature condition, the gluing, curing and wiring of the strain gauges are completed. During the gluing process, the gluing angle deviation of the strain gauge is strictly controlled to ensure the directional accuracy of the strain measurement. After the sensor state is stable, the output of all strain gauges is calibrated to the zero stress state to establish an initial zero-point baseline and eliminate the initial error caused by thermal strain.

[0054] (2) Intermediate state calibration after assembly: According to the factory assembly process, complete all assembly work with the axis vertical. After assembly, the shaft system needs to be adjusted to the working posture consistent with the on-site operation - that is, the axis is at a certain angle to the horizontal plane (this angle is the wind turbine elevation angle). During this posture conversion, the gravitational stress field of the shaft system will change. After assembly, let it stand for a preset time until the assembly stress is completely stable; this is because the preload of the bolts will have a stress relaxation process after assembly. The stress state can only be completely stable after the relaxation rate is lower than the threshold. Collect the output data of the current strain gauge, and at the same time construct a finite element simulation model of the gravity stress field of the shaft system under this posture to calculate the theoretical strain value of each measuring point. The measured strain is compared with the theoretical strain to check the deviation and evaluate the working status of the strain gauge. If the deviation exceeds the preset threshold, it is determined that the strain gauge has failed to be glued or has a circuit fault. The entire shaft system needs to be disassembled and reassembled. Before reassembly, the strain gauge is glued again and the zero-position calibration of stage 1 is completed. Then, the assembly and calibration of this stage are repeated. At the same time, the strain value after assembly is used as the new measurement benchmark to eliminate the interference caused by assembly prestress and attitude gravity stress.

[0055] (3) Calibration of the digital twin surrogate model under lateral load: The assembled motor unit is hoisted to the vertical test support in the factory and adjusted to a vertical posture consistent with the on-site operation. Gradient lateral loads are applied to the shaft system through a closed-loop hydraulic loading system, covering multiple typical load levels from no load to ultimate load. Under each load gradient, after the system strain response is completely stable, strain data for a preset time is collected and the average value is taken to eliminate vibration interference and ensure data stability. Before formally training the surrogate model, the strain measurement system whose accuracy has been confirmed in stage 2 is first loaded and verified: Several verification load points are extracted from the gradient lateral load, and the measured strain values ​​under each verification load are compared with the theoretical strain values ​​calculated by finite element simulation; if the deviations of both are within the preset accuracy range, the strain measurement system is confirmed to be working normally within the full range and can enter the surrogate model training stage; if the deviation exceeds the standard, the sensor status needs to be checked and the loading verification is re-executed. After confirming that the strain measurement accuracy is qualified, the digital twin strain-load coupled surrogate model of the shaft system is trained by combining the input value of the loaded load. The model employs a physics-data co-training approach. Specifically, it is implemented as follows: First, a full-size three-dimensional finite element model of the shaft system is constructed using ANSYS Workbench, meshed with Solid186 hexahedral high-order elements. Strain-load correspondences under 100 different load conditions are obtained through simulation analysis as prior physical data. Then, measured strain-load data under seven gradient loads collected during the lateral load loading process are used as experimental data. The prior physical data and experimental data are fused at a 3:1 ratio, and a Bayesian regularization algorithm is used for model training, balancing physical accuracy and experimental precision. Simultaneously, the initial baseline values ​​of the displacement sensor and bolt stress sensor are recorded under this state, providing a benchmark for subsequent state determination.

[0056] Operational status validity determination mechanism: During unit operation, periodic shutdown verification is performed. The validity of strain measurement data is verified using data from the condition monitoring unit. The determination mechanism is as follows: (1) Collect the measured value of the displacement sensor under the current state, compare it with the initial baseline value, and calculate the relative displacement deviation rate. The displacement deviation rate reflects the change of bearing clearance. When the raceway is worn, the clearance will increase and the measured value of the displacement sensor will deviate. (2) Collect the measured values ​​of the ultrasonic bolt stress sensor under the current state, compare them with the initial baseline value, and calculate the relative stress deviation rate. The stress deviation rate also reflects the change in bearing clearance. When the bearing raceway wear causes the clearance to increase, the preload of the bearing pressure ring fastening bolt will decrease accordingly, thus indirectly characterizing the degree of bearing degradation. (3) If both deviation rates are within the preset error threshold range, it is determined that the bearing has not experienced significant raceway wear and clearance changes, the force transmission path of the shaft system has not changed, the current strain gauge measurement data is valid, and the original digital twin proxy model is still applicable. (4) If any deviation rate exceeds the threshold range, it is determined that the bearing condition may have deteriorated and the clearance has exceeded the normal range. An early warning should be triggered immediately, and maintenance personnel should be notified to arrange for inspection and assessment to determine whether the bearing needs to be replaced or repaired. During this period, the use of the current agent model for load field reconstruction should be suspended. The recalibration process should be executed after the inspection is completed and the bearing condition returns to normal.

[0057] This judgment mechanism employs a two-parameter orthogonal judgment method, which effectively avoids misjudgments caused by a single parameter. For example, minor displacement fluctuations caused by temperature changes will not lead to misjudgments because they do not cause changes in bolt stress, significantly improving the robustness of the judgment. The preset error threshold is set based on: bearing design life, material fatigue characteristics, and engineering operation experience, combined with the rated load, operating speed, and operating environment of the spindle bearing. Finite element simulation is used to simulate the changing law of bearing force transmission path under different deviation rates to determine the deviation rate threshold range, ensuring that the judgment results can effectively identify bearing condition degradation while avoiding misjudgments.

[0058] When the bearing condition is confirmed to be within the serviceable range after early warning maintenance, but the mapping relationship needs to be fine-tuned due to slight changes in break-in, a simplified on-site recalibration can be performed: using the strain data under pure gravity conditions when the unit is stationary, the proxy model is incrementally updated to achieve on-site online recalibration without disassembly, without having to transport the unit back to the factory for lateral load loading, thus solving the technical pain point of difficult recalibration in offshore wind farms.

[0059] Load field reconstruction and adaptive tracking driven by digital twin agent model: When the strain data is deemed valid, the load field reconstruction and evaluation steps are executed. (1) Collect real-time multi-source strain data of strain gauges under the current operating state. The sampling frequency meets the requirements of dynamic load capture and can capture changes in dynamic load. (2) Input the real-time strain data into the digital twin surrogate model established during the calibration phase, and reconstruct the actual input load vector of the current unit from the strain response using the Bayesian regularized inverse mapping algorithm, including radial load, axial load, overturning moment and other full-dimensional load components. Since the inverse problem itself is ill-conditioned, small strain errors can lead to large deviations in load calculation; the Bayesian regularization algorithm effectively constrains the solution process by introducing the prior distribution of the load, solves the ill-conditioned nature of the inverse problem, avoids overfitting and ensures the stability of the solution.

[0060] (3) Compare the actual load obtained from the reconstruction with the input load in the design stage to quantitatively evaluate the margin and accuracy of the design load, and provide a basis for structural optimization and life assessment. Through this comparison, designers can accurately grasp the deviation between the actual load and the design load, and provide support for subsequent unit design optimization.

[0061] (4) A deep learning-based intelligent operating condition matching module is introduced, which uses wind speed, wind direction, rotational speed, and output power as input features to automatically identify the current operating condition. This module uses a lightweight network and employs a large amount of historical operating data from the same type of unit during training, which can accurately identify various typical operating conditions. For different operating conditions, the corresponding parameterized wind load input model is called to dynamically correct the digital twin proxy model. Because the dynamic response of the shaft system differs under different operating conditions, dynamic correction can further improve the accuracy of reconstruction and identification.

[0062] (5) By collecting data over a long period of time, the ultimate load under extreme wind conditions is continuously captured, providing a measured basis for the input of ultimate load conditions in the design phase and solving the technical problem of scarce ultimate load conditions data. Since the probability of extreme wind conditions is extremely low, the design phase can usually only estimate the load using standards. However, this method can obtain real ultimate load data through long-term monitoring, providing a more accurate basis for the design.

[0063] Example 5 Simulation results of shaft stress field are as follows Figure 3 — Figure 6 As shown, the strain distribution of the shaft system under rated load conditions and the location of the strain measurement points of the present invention are displayed. The strain gradient is the largest in the high-sensitivity stepped transition section, which is perfectly matched with the measurement point selection strategy of the present invention.

[0064] This embodiment takes the main shaft bearing of a large-capacity direct-drive offshore wind turbine as an example to illustrate the implementation of the present invention in detail. This turbine is the main model of a certain offshore wind farm, and the main shaft bearing adopts a double-row tapered roller bearing.

[0065] (1) Sensor layout scheme and optimization design As shown in Figure 1, before bearing assembly, the sensor arrangement is completed. The optimal arrangement scheme adopted in this embodiment is as follows: On the outer wall of the fixed shaft, four sets of resistance strain gauges are evenly arranged circumferentially. The strain gauges are HBM 1-LY11-6 / 350 foil strain gauges with a resistance of 350Ω, a sensitivity coefficient of 2.0, and a grid length of 6mm, used to measure the circumferential strain response of the fixed shaft side wall; On the inner wall of the rotating shaft, three sets of strain gauges are arranged, two of which are arranged in the bearing body section and one of which is arranged in the stepped transition section of the shaft system to improve the measurement sensitivity under small load conditions; The stress gradient in the stepped transition section is larger, which can more sensitively reflect load changes.

[0066] One set of non-contact eddy current displacement sensors is installed between the mating surfaces of the bearing retaining ring and the fixed-axis step. The sensors are Keyence EX-416, with a measurement range of 0~2mm and a resolution of 0.01μm. They are used to measure the relative displacement between the two and indirectly monitor the evolution of bearing clearance. Four sets of ultrasonic bolt stress sensors are evenly installed on the circumferential retaining ring fastening bolts of the bearing. The sensors are USM Go ultrasonic stress detectors with a resolution of 0.1MPa. They are used to monitor the changes in bolt preload stress and characterize the degradation of the connection condition.

[0067] The method for selecting the preset strain-sensitive section is as follows: The strain distribution of the shaft system under unit load is calculated through finite element simulation, and the section with the largest strain gradient is selected as the high-sensitivity measuring point. This method can maximize the measurement sensitivity under small load conditions. Figure 3 — Figure 6 As shown in the figure, the finite element simulation results of this embodiment show that the strain gradient is the largest in the stepped transition section. Therefore, placing one of the strain measurement points at this location effectively improves the measurement sensitivity of small loads.

[0068] (2) Multi-stage step-by-step calibration process Phase 1: Initial Zero-Point Calibration at Constant Temperature The fixed and rotating shafts, awaiting assembly, are placed in a temperature- and humidity-controlled workshop. The workshop temperature is maintained at a stable 25℃, with temperature fluctuations controlled within ±0.5℃, meeting the preset temperature accuracy range requirements. Under these conditions, all strain gauges are pasted, cured, and wired. During pasting, a special strain gauge adhesive is used, and the curing time is 24 hours to ensure a tight fit between the strain gauges and the shaft system, avoiding thermal strain interference caused by temperature gradients. After the strain gauges are fully cured and their outputs stabilize, the output values ​​of all strain gauges are collected and zeroed, completing the initial zero-point calibration. At this point, the initial output of all strain gauges is 0, effectively eliminating initial thermal strain interference.

[0069] Phase 2: Intermediate Assembly Calibration Following the factory assembly process, the shaft system was assembled with the axis vertical, completing the assembly of bearings, end caps, and connecting bolts. After assembly, it was left to stand for 2 hours to allow the assembly stress to fully release and stabilize, the bolt preload to loosen, and the stress state to stabilize. The current output values ​​of the strain gauges were collected, and a finite element simulation model of the shaft system's gravity stress field under this orientation was constructed to calculate the theoretical strain values ​​at each strain gauge measuring point. The simulation model used second-order tetrahedral elements with a mesh size of 5mm to ensure simulation accuracy. The measured strain values ​​were compared with the theoretical values; a deviation of less than 5% indicated that the strain gauges were functioning normally, without adhesive failure or circuit faults. Simultaneously, the strain values ​​after assembly were used as a new measurement benchmark to eliminate interference from assembly preload and gravity stress.

[0070] Phase 3: Calibrating the digital twin surrogate model under lateral load The assembled motor unit was hoisted onto the vertical test stand in the factory and adjusted to a vertical operating posture completely consistent with the wind farm site. The closed-loop hydraulic loading device was installed at the loading end of the shaft system (e.g., Figure 1 As shown in the loading direction, the maximum loading capacity of this loading device is 600t, and the loading accuracy is 0.5%. The hydraulic loading system is controlled to apply gradient lateral loads of 10t, 50t, 100t, 200t, 300t, 400t, and 500t sequentially (these seven gradient loads serve as experimental data for model training). At each load point, the system is stabilized for 5 minutes. After the system strain response has fully stabilized, strain data for 10 seconds is collected and averaged to eliminate vibration interference. These experimental data are fitted using a Bayesian regularization algorithm, and simultaneously integrated with 100 independent physical prior data points generated from finite element simulation (these 100 conditions do not overlap with the seven gradient load conditions used for training, and the load range also covers 10t~500t, used to supplement the model's generalization ability). This training yields a digital twin strain-load coupled surrogate model: F=f(ε1,ε2,...,ε). i ); where F is the input load vector, including the radial load F x F y Overturning moment M x M y Equal all-dimensional components; ε i These are the measured values ​​of each strain gauge. This model is the core model for subsequent load field reconstruction, and its fitting determination coefficient R² reaches 0.998, demonstrating extremely high accuracy. Simultaneously, the initial baseline value D0 of the displacement sensor and the initial baseline value S0 of the bolt stress sensor are recorded under this calibration state.

[0071] (3) Simulation verification and accuracy evaluation To verify the effectiveness and high precision of the load field reconstruction method proposed in this invention, a full-size three-dimensional finite element simulation model of the main shaft bearing of the direct-drive wind turbine was constructed based on the ANSYS Workbench platform. Solid186 hexahedral high-order elements were used for mesh generation, with a mesh size set to 10mm (this size is the global mesh size of the overall shaft system model). A local mesh refinement strategy was adopted in the critical regions where strain measurement points were located, with a mesh size of 2mm in the refined regions to ensure the accuracy of strain calculation at the measurement points. The 10mm global mesh size was verified for mesh convergence, and the load-displacement response convergence deviation of the overall model was less than 0.5%, consistent with the finite element simulation model during the multi-stage calibration process, ensuring the consistency and reliability of the simulation results. To comprehensively cover the actual operating conditions of the unit, 100 sets of full-range load verification conditions (load input range covering 10t to 500t, with no overlap with the training conditions, used to simulate the shaft system loading state under different conditions) were designed independently of the training data in the calibration phase. This allows for independent and comprehensive verification of the reconstruction performance of this method, avoiding misjudgments of accuracy caused by overfitting. Simulation verification results show that, within the full-range load range, the load reconstruction error of this method is within the high-precision range for engineering applications, significantly better than the error level of existing single-calibration methods.

[0072] Example 6 This embodiment takes a large-capacity direct-drive offshore wind turbine as an example. The turbine adopts an outer ring rotating configuration (such as...). Figure 2 As shown in the diagram, the rotating shaft is located on the outside, the fixed shaft is located on the inside, the inner ring of the bearing is fixed to the fixed shaft, and the outer ring rotates with the rotating shaft. The main shaft bearing also uses a double-row tapered roller bearing.

[0073] Simulation results of shaft stress field are as follows Figure 7 — Figure 10 As shown, the strain distribution of the shaft system under rated load conditions and the location of the strain measurement points of the present invention are displayed. The strain gradient is the largest in the high-sensitivity stepped transition section, which is perfectly matched with the measurement point selection strategy of the present invention.

[0074] (1) Sensor layout scheme In this configuration, the fixed shaft is the inner shaft, and the rotating shaft is the outer shaft. The arrangement principle of the strain gauges is the same as in Example 5, with strain gauges arranged on the fixed shaft wall and the rotating shaft wall corresponding to the bearing location. The specific arrangement scheme is as follows: on the inner wall of the fixed shaft (inner shaft), four sets of resistance strain gauges are evenly arranged circumferentially to measure the circumferential strain response on the fixed shaft side; on the outer wall of the rotating shaft, six sets of strain gauges are arranged, of which four sets are arranged in the bearing body section (two sets near the driving end and two sets near the non-driving end), and two sets are arranged in the high-sensitivity section of the shaft system step transition; a non-contact eddy current displacement sensor is installed between the mating surfaces of the bearing pressure ring and the fixed shaft step to monitor the bearing clearance change; and an ultrasonic bolt stress sensor is installed on the bearing pressure ring fastening bolt to monitor the bolt preload stress change.

[0075] (2) Calibration and Reconstruction Process The multi-stage step-by-step calibration process under this configuration is completely consistent with that in Example 5, including: constant temperature initial zero-position calibration → intermediate state calibration after assembly → transverse load loading proxy model calibration. Since the three-stage calibration method of this invention is aimed at strain gauges on the "fixed-axis shaft wall" and the "rotating shaft wall", it does not depend on which specific ring rotates, so the calibration principle and operation steps do not need to be adjusted.

[0076] During load field reconstruction, the shaft wall strain data collected during operation is also input into the surrogate model, and the full-dimensional load vector is reconstructed through the Bayesian regularized inverse mapping solution algorithm. From the perspective of mechanical principles, whether the load is transmitted through the outer ring → roller → inner ring → fixed shaft (inner shaft) or through the inner ring → roller → outer ring → rotating shaft (outer shaft), the strain response generated on the shaft wall is a function of the external load. Therefore, the mapping relationship of the surrogate model still holds.

[0077] (3) Simulation verification and accuracy evaluation To systematically verify the effectiveness of this method for the outer ring rotating configuration, a full-size three-dimensional finite element simulation model of the outer ring rotating bearing was constructed based on the ANSYS Workbench platform. One hundred independent verification cases were also designed. The results show that the load field reconstruction accuracy under the outer ring rotating configuration is at the same level as that under the inner ring rotating configuration, both meeting the high-precision requirements for engineering applications, thus verifying the versatility of this method for different rotating configurations. (Specific accuracy indicators will be added after subsequent simulation / experimental verification.) Under various simulation verification conditions, strain response data from seven pre-set strain measurement points of this invention are extracted and imported as input parameters into the digital twin strain-load coupling surrogate model constructed by this method. The load vector is reconstructed by the Bayesian regularized inverse mapping solution algorithm. The reconstructed load data is compared and analyzed with the pre-set input load of the finite element simulation to quantitatively evaluate the load field reconstruction accuracy of this invention. At the same time, it is quantitatively compared with the existing single calibration method (the existing single calibration method uses the same sensor arrangement as this embodiment, only performs a single isothermal zero-position calibration, and the other steps are the same).

[0078] Simulation results show that, within the full load range, the radial load and overturning moment reconstruction errors of this method are within the high-precision range for engineering applications, significantly better than the error level of existing single-calibration methods. This verifies the accuracy advantage of this method's three-stage calibration combined with the physical-data collaborative proxy model.

[0079] The simulation results fully demonstrate the effectiveness and reliability of the load field reconstruction method proposed in this invention. It can achieve high-precision reconstruction of the shaft system's load across all dimensions under full-range load conditions, maintaining excellent reconstruction performance even under extreme load conditions. Furthermore, the simulation results further verify the superiority of the physical-data collaborative training mode adopted in this invention—the digital twin surrogate model that integrates prior physical information from finite element simulation. Compared to a purely data-driven surrogate model, this model improves reconstruction accuracy by more than 30%, significantly enhances robustness and generalization ability, and effectively adapts to load reconstruction requirements under different operating conditions.

[0080] The following operational status determination and load field reconstruction test are applicable to the two configurations of Example 5 and Example 6.

[0081] Operational Status Validity Determination Test and Result Analysis After applying the measurement system and validity determination mechanism constructed in this invention to the target offshore direct-drive wind turbine, in order to ensure the long-term reliability of the load field reconstruction results, a static verification test is performed every 3 operating cycles (i.e., every 3 months) according to the preset operation and maintenance verification procedure. The validity determination of the strain measurement data is completed through multi-source data collected by the condition monitoring unit. The specific determination process and results are as follows: After shutdown, once the shaft temperature has dropped to ambient temperature and the stress state has fully stabilized, collect the measured value D from the current non-contact displacement sensor. c Based on the initial baseline value D0, the relative displacement deviation rate is calculated using the following formula to quantify the evolution of bearing clearance: ΔD=(|D c —D0|) / D0×100%.

[0082] Synchronously acquire the measured value S of the ultrasonic bolt stress sensorc Based on the initial baseline value S0, the relative deviation rate of bolt preload stress is calculated using the following formula to characterize the stability of the shaft connection: ΔS=(|S c —S0|) / S0×100%.

[0083] Validity Determination: Based on the allowable error range of the shaft system operation, a preset dual-parameter judgment threshold of 3% was established (this threshold was determined by finite element simulation of the bearing force transmission path variation under different deviation rates, based on the design life of the unit bearings in this embodiment, material fatigue characteristics, and engineering operation experience). Calculations showed that both the relative displacement deviation rate and the relative bolt stress deviation rate were less than this preset threshold. Therefore, it can be determined that: the main shaft bearing did not experience significant raceway wear or abnormal increase in clearance; the internal force transmission path of the shaft system remained stable; the strain data collected by the current strain sensing unit is valid; and the previously calibrated digital twin strain-load coupling proxy model can continue to be used for load field reconstruction calculations.

[0084] Long-term operation verification results show that after one operating cycle (1 year) of continuous and stable operation, the relative deviation rate of the two parameters in each shutdown verification is within the preset threshold range. This fully verifies the reliability and long-term applicability of the operating status effectiveness judgment mechanism proposed in this invention, and can effectively avoid the load field reconstruction distortion problem caused by bearing condition degradation. Based on engineering experience and finite element simulation analysis, setting the threshold too small (e.g., 1%) may lead to a high false alarm rate due to normal fluctuations in ambient temperature; setting the threshold too large (e.g., 5%) may result in missed detections of slight degradation. After comprehensive analysis, this embodiment selects 3% as the recommended threshold, achieving a better balance between early warning sensitivity and false alarm rate.

[0085] Load field reconstruction and long-term adaptive tracking test During normal unit operation, multi-source strain response data of the shaft system are collected in real time through strain sensing units, with a sampling frequency meeting the engineering requirements for dynamic load capture. The collected strain data is input in real time into the digital twin strain-load coupling surrogate model constructed in the early stage of calibration. Based on the Bayesian regularized inverse mapping solution algorithm, high-precision reconstruction of the actual input load vector of the unit is completed. Taking the rated wind speed condition as an example, the actual radial load and overturning moment of the shaft system obtained by reconstruction are recorded and quantitatively compared with the input loads (radial load and overturning moment) preset in the design stage. Based on the measured comparison results, the design margin and manufacturing cost of the main shaft bearings of subsequent units of the same model can be optimized and reduced.

[0086] To further improve the load field reconstruction accuracy under different operating conditions, a deep learning-based intelligent operating condition matching module is introduced. Using the real-time operating parameters of the unit (wind speed, wind direction, rotor speed, and output power) as input feature vectors, the module's feature extraction and pattern recognition functions enable adaptive identification of operating conditions. Simultaneously, the parameterized wind load input model under the corresponding operating condition is invoked to dynamically correct the digital twin strain-load coupling proxy model, further controlling the load field reconstruction error within the preset accuracy range and significantly improving the robustness and reconstruction accuracy of the model under complex and variable operating conditions.

[0087] Furthermore, through long-term continuous operational data acquisition and storage, the system captures and records extreme wind loads throughout the entire lifecycle of the unit. After deployment, this measurement system can continuously acquire long-term operational data to capture load responses under extreme wind conditions. When extreme conditions such as out-of-cut wind speeds occur, the system automatically triggers a high-frequency acquisition mode to record the shaft strain response under these conditions and reconstructs the ultimate load using a surrogate model. The reconstructed ultimate load data can be compared with the ultimate load estimated based on standards during the design phase, providing a measured basis for the precise correction of ultimate load inputs and effectively solving the technical pain point of insufficient extreme load data in traditional design.

[0088] It is understood that the present invention has been described through some embodiments, and those skilled in the art will recognize that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Furthermore, under the teachings of the present invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of the present invention.

Claims

1. A method for reconstructing the load field of a direct-drive wind turbine main shaft bearing, characterized in that: The main shaft bearing system includes a rotating shaft that rotates with the wind turbine and a stationary shaft that does not rotate; the method includes the following steps: Step S1: Arrange multiple sets of strain gauges in the preset strain-sensitive section of the shaft system. The preset strain-sensitive section is selected from at least one of the shaft wall of the fixed shaft, the shaft wall of the rotating shaft, and the stepped transition section of the shaft system. Step S2: Perform multi-stage calibration sequentially on the shaft system with the strain gauges arranged thereon to construct a digital twin proxy model; Step S3: During unit operation, collect real-time data from the strain gauges and input them into the digital twin proxy model to reconstruct the load field of the main shaft bearing; In step S2, the multi-stage calibration includes: Step S21, First stage constant temperature zero-point calibration: Before assembling the bearing components, the strain gauges are pasted, cured and calibrated to zero stress state in a constant temperature environment; Step S22, Second stage post-assembly state calibration: After the shaft system is assembled, the strain gauge output is collected and updated to the measurement reference value to eliminate the error introduced by the assembly prestress and attitude gravity stress; Step S23, Third Stage Lateral Loading Calibration: Apply multiple known gradient lateral loads to the assembled shaft system, collect the strain response under each load, and fuse the physical prior data obtained through finite element simulation with the strain response data. Train the digital twin proxy model through physical-data co-training, wherein the digital twin proxy model uses a Bayesian regularization algorithm to establish an inverse mapping relationship from strain to load.

2. The load field reconstruction method for a direct-drive wind turbine main shaft bearing according to claim 1, characterized in that: In step S21, before assembling the bearing components, the bearing components are placed in a constant temperature environment. The ambient temperature is controlled to be at room temperature, and the temperature difference between the fixed shaft and the rotating shaft is ensured to be within the preset temperature accuracy range to eliminate the thermal output interference of the resistance strain gauge. Due to the significant thermal output characteristics of the strain gauge, temperature fluctuations within the preset temperature accuracy range can control the interference of thermal strain to a very small range, much smaller than the working strain of the shaft system, thus eliminating the interference of thermal strain from the source. Under this constant temperature condition, the bonding, curing, and wiring of the strain gauges are completed. During the bonding process, the bonding angle deviation of the strain gauges is strictly controlled to ensure the directional accuracy of the strain measurement. After the sensor state stabilizes, the output of all strain gauges is calibrated to a zero stress state to establish an initial zero baseline and eliminate the initial error caused by thermal strain.

3. The load field reconstruction method for a direct-drive wind turbine main shaft bearing according to claim 2, characterized in that: In step S22, all assembly work is completed in a vertical state according to the factory assembly process. After assembly, the shaft system needs to be adjusted to a working posture consistent with the on-site operation—that is, the axis is at a certain angle to the horizontal plane, which is the wind turbine elevation angle. During this posture change, the gravitational stress field of the shaft system will change. After assembly, the shaft system is left to stand for a preset time until the assembly stress is completely stable. The output data of the strain gauge is collected, and a finite element simulation model of the gravitational stress field of the shaft system under this posture is constructed to calculate the theoretical strain value of each measuring point. The measured strain is checked against the theoretical strain to evaluate the working state of the strain gauge. If the deviation exceeds the preset threshold, it is determined that the strain gauge has failed to adhere or has a circuit fault. The entire shaft system needs to be disassembled and reassembled. Before reassembly, the strain gauge should be re-adhered and the zero-position calibration in step S21 should be completed. Then, the assembly and calibration in this stage should be re-executed. At the same time, the strain value after assembly should be used as the new measurement benchmark to eliminate the interference caused by assembly prestress and attitude gravity stress.

4. The load field reconstruction method for a direct-drive wind turbine main shaft bearing according to claim 3, characterized in that: In step S23, the assembled motor unit is hoisted to the vertical test bracket in the factory and adjusted to a vertical posture consistent with the on-site operation; through the closed-loop hydraulic loading system, gradient lateral loads are applied to the shaft system, covering multiple typical load levels from no load to ultimate load; Under each load gradient, after the system strain response has fully stabilized, strain data for a preset duration are collected and averaged to eliminate vibration interference and ensure data stability. Before formally training the surrogate model, the strain measurement system whose accuracy has been confirmed in step S22 is first subjected to load verification: several verification load points are extracted from the gradient lateral load, and the measured strain values ​​under each verification load are compared with the theoretical strain values ​​calculated by finite element simulation. If the deviations are both within the preset accuracy range, the strain measurement system is confirmed to be working normally within the full range, and the surrogate model training phase can begin. If the deviation exceeds the standard, the sensor status needs to be checked and the load verification is re-executed. After confirming that the strain measurement accuracy is qualified, the digital twin strain-load coupled surrogate model of the shaft system is trained by combining the input values ​​of the loaded load.

5. The load field reconstruction method for a direct-drive wind turbine main shaft bearing according to claim 4, characterized in that: The digital twin strain-load coupled proxy model adopts a physics-data co-training approach. Specifically, it is implemented as follows: First, a full-size three-dimensional finite element model of the shaft system is constructed based on ANSYS Workbench, and the mesh is generated using Solid186 hexahedral high-order elements. Strain-load correspondences under 100 different load conditions are obtained through simulation analysis as prior physical data. Then, measured strain-load data under seven gradient loads collected during the lateral load loading process are used as experimental data. The prior physical data and experimental data are fused at a 3:1 ratio, and a Bayesian regularization algorithm is used for model training, balancing physical accuracy and experimental precision. Simultaneously, the initial baseline values ​​of the displacement sensor and bolt stress sensor are recorded under this state to provide a benchmark for subsequent state determination.

6. The load field reconstruction method for a direct-drive wind turbine main shaft bearing according to claim 5, characterized in that: In step S3, when the strain data is determined to be valid, the load field reconstruction and evaluation steps are performed: Step S31: Collect real-time multi-source strain data of strain gauges under the current operating state. The sampling frequency meets the requirements of dynamic load capture and can capture changes in dynamic load. Step S32: Input the real-time strain data into the digital twin proxy model established in the calibration stage, and reconstruct the actual input load vector of the current unit from the strain response through the Bayesian regularized inverse mapping solution algorithm. The actual input load vector includes radial load, axial load and overturning moment. Step S33: Compare the actual load obtained from the reconstruction with the input load in the design stage to quantitatively evaluate the margin and accuracy of the design load, and provide a basis for structural optimization and life assessment. Step S34: Introduce a deep learning-based intelligent operating condition matching module, using wind speed, wind direction, rotational speed, and output power as input features to automatically identify the current operating condition; The intelligent operating condition matching module uses a lightweight network and employs a large amount of historical operating data from the same type of unit during training, which can accurately identify a variety of typical operating conditions. For different operating conditions, it calls the corresponding parameterized wind load input model to dynamically correct the digital twin proxy model.

7. The load field reconstruction method for a direct-drive wind turbine main shaft bearing according to claim 6, characterized in that: The system also includes step S4, a step to determine the validity of the operating status: In step S23, the initial baseline values ​​of the non-contact displacement sensor and the ultrasonic bolt stress sensor are recorded synchronously; after the unit starts operating, the current values ​​of the displacement sensor and the bolt stress sensor are periodically collected and compared with the corresponding initial baseline values ​​to calculate the displacement deviation rate and the stress deviation rate; when both the displacement deviation rate and the stress deviation rate are less than a preset threshold, the digital twin proxy model is determined to be valid, and step S3 is executed; otherwise, the model is determined to be invalid and an early warning is triggered.

8. The load field reconstruction method for a direct-drive wind turbine main shaft bearing according to claim 7, characterized in that: If the digital twin agent model is determined to be faulty, and the bearing condition is confirmed to be within the serviceable range, the strain data generated by the shaft system under pure gravity conditions in the stationary state of shutdown is used to perform on-site recalibration of the model in an online incremental update manner without disassembly.

9. The load field reconstruction method for a direct-drive wind turbine main shaft bearing according to claim 8, characterized in that: It also includes an adaptive operating condition step: introducing a deep learning-based operating condition recognition module, which uses the real-time wind speed, wind direction, rotational speed and output power of the unit as input to automatically identify the current operating condition; The parameterized wind load input model corresponding to the current operating condition is invoked to dynamically correct the digital twin proxy model, thereby improving the reconstruction accuracy under complex and variable operating conditions.

10. The load field reconstruction method for a direct-drive wind turbine main shaft bearing according to claim 9, characterized in that: It also includes an ultimate load capture step: during unit operation, the unit's operating parameters are continuously monitored, and when an extreme operating condition exceeding a preset cut-out wind speed or preset speed threshold is identified, a high-frequency acquisition mode is triggered to record the strain response at the corresponding moment; based on the recorded strain response, the ultimate load data is reconstructed through the digital twin proxy model.

11. A load measurement system for the main shaft bearing of a direct-drive wind turbine generator, characterized in that, To implement the method according to any one of claims 1-10, comprising: The strain sensing unit includes the multiple sets of strain gauges; The condition monitoring unit includes a non-contact displacement sensor arranged between the bearing retaining ring and the mating surface of the fixed axis step, and an ultrasonic bolt stress sensor arranged on the bearing retaining ring fastening bolt. The loading calibration unit includes an in-plant vertical test stand and a closed-loop hydraulic loading system for applying the gradient lateral load; The data processing and reconstruction unit is configured to perform the multi-stage calibration to construct the digital twin agent model, perform the validity determination of the running state, and reconstruct the load field based on the digital twin agent model.

12. The direct-drive wind turbine main shaft bearing load measurement system according to claim 11, characterized in that: Multiple sets of resistance strain gauges are arranged in a preset strain-sensitive section of the main shaft bearing system. The preset strain-sensitive section includes at least one selected from the fixed shaft wall, the rotating shaft wall, and the stepped transition section of the shaft system corresponding to the bearing's location. The fixed shaft refers to a fixed shaft that does not rotate with the wind turbine, and the rotating shaft refers to a rotating shaft that rotates with the wind turbine. In the inner ring rotation configuration, the fixed shaft is the outer shaft and the rotating shaft is the inner shaft; in the outer ring rotation configuration, the fixed shaft is the inner shaft and the rotating shaft is the outer shaft. The multiple sets of resistance strain gauges are used to measure the elastic strain response of the shaft system under external loads, realizing the fusion acquisition of multi-source strain data. The multiple sets of resistance strain gauges are high-precision foil strain gauges.

13. The direct-drive wind turbine main shaft bearing load measurement system according to claim 12, characterized in that: The non-contact displacement sensor is a high-precision eddy current displacement sensor, which is arranged between the mating surfaces of the bearing pressure ring and the fixed-axis step to monitor the relative displacement changes between the two and indirectly characterize the evolution of the bearing clearance; the ultrasonic bolt stress sensor is arranged on the fastening bolt of the bearing pressure ring to monitor the evolution of the bolt preload stress and indirectly characterize whether the bearing clearance has changed significantly.

14. The direct-drive wind turbine main shaft bearing load measurement system according to claim 13, characterized in that: The loading calibration unit is used to apply a controllable gradient lateral load to the shaft system during the calibration phase to simulate different loading conditions and provide labeled data for the training of the digital twin surrogate model.

15. A direct-drive wind turbine main shaft bearing load measurement system according to claim 14, characterized in that: The data processing and reconstruction unit includes a storage module, a state determination module, and a load reconstruction module. The storage module is used to store baseline data from the calibration phase; the state determination module is used to determine the validity of the operating state; and the load reconstruction module is used to perform load field reconstruction calculations and to compare and evaluate the actual load with the design input load.

Images