Impact force identification in engineering structures: from traditional regularization to hybrid neural networks

Accurate impact force identification is essential for ensuring the structural integrity and safety of engineering systems; however, addressing this inverse problem remains a formidable challenge due to its inherently ill-posed nature. Traditional regularization techniques, such as Tikhonov regularization and singular value decomposition, have long been employed to stabilize solutions and mitigate uncertainties. Yet, recent advancements in machine learning, particularly hybrid neural network frameworks, are transforming the field by offering robust, data-driven alternatives. This paper presents a comprehensive exploration of impact force identification methodologies, ranging from classical inverse problem-solving approaches to cutting-edge hybrid deep learning techniques. To demonstrate the practical applicability of these methods, we include a case study on a fiberglass-reinforced plastic (FRP) cooling tower fan blade, evaluating the effectiveness of the proposed machine learning approach under realistic conditions. By bridging physics-based modelling with artificial intelligence, this study highlights the advantages, challenges, and future prospects of hybrid approaches in structural health monitoring. These techniques not only enhance precision and reduce computational costs but also pave the way for real-time impact force identification in complex engineering systems.