Impact force identification on wind turbine blades using the ADMM-Net deep unfolding network guided by structural response priors

During the long-term service of wind turbine blades, external impacts are unavoidable. Identifying impact forces is therefore essential for the safe and stable operation of wind turbines. At present, both model-based and deep learning methods have been widely used for this task. Model-based methods are interpretable but sensitive to model mismatch, computationally demanding, and difficult to optimize. Deep learning methods have strong fitting capability but lack physical interpretability due to their “black-box” nature. To address these limitations, this paper proposes a structure-response prior-driven deep unfolding network, termed ADMM-Net, for impact force identification. The method applies algorithm unfolding to transform the iterative steps of the alternating direction method of multipliers (ADMM) into a deep network, and introduces structural response priors obtained from experiments for physics-based initialization. In this way, the framework achieves both interpretability and adaptiveness in an end-to-end manner. Experimental results from impacts experiments on a section of a wind turbine blade show that ADMM-Net not only enables accurate localization and reconstruction of impact forces, but also demonstrates strong noise resistance under complex noisy conditions.