关键词: 风电机组, 深度学习代理模型, 齿轮箱多体动力学, 多模态振动, 多时间尺度

Abstract:

To address challenges in characterizing high-frequency vibrations of wind turbine gearboxes, the long computation time of rigid-flexible coupled multi-body dynamics models, and the complexity of configuring gearbox models across multiple scenarios, this study proposed a deep learning modeling method for multi-scale operation using full-condition digital testing. The study proposed a cascaded extended simulation scheme based on stream data-driven OpenFAST and Adams and utilized dynamic mode decomposition technology to construct a multi-scale dataset for the flexible multi-body dynamics characteristics of the gearbox under all operating conditions of the wind turbine. Based on this dataset, a digital surrogate model covering multiple vibration modes and time scales was constructed using the TimeMixer deep learning algorithm. The simulation experiment results show that the deep learning surrogate model for the wind turbine gearbox has the ability to accurately reflect the vibration characteristics, load characteristics, and dynamic behaviors of the gearbox under various typical working conditions. Moreover, the computational efficiency, simulation accuracy, and adaptability to variable working conditions have been improved.

Key words: wind turbine, deep learning surrogate model, gearbox multi-body dynamics, multimodal vibration, multi-time scale