Physics-Informed Machine Learning for Dynamic Displacement Reconstruction of Smart Gua Sha Device

Smart nursing devices have garnered significant attention in the field of rehabilitation medicine, with their effectiveness in enhancing treatment precision and enabling clinical-grade home-based rehabilitation care being well-established. Nevertheless, current smart nursing devices face persistent challenges in real-time displacement tracking and trajectory estimation, particularly when employing indirect measurement methodologies. This paper focuses on Gua Sha, a traditional Chinese physical therapy, and proposes a physical information-based deep neural networks (PINN) framework that integrates convolutional neural networks (CNNs) and bidirectional long short-term memory (BiLSTM) architectures for dynamic trajectory reconstruction of professional Gua Sha techniques. The dynamic properties of typical Gua Sha trajectories are systematically characterized, and a methodology for acquiring associated kinematic parameters is established. The proposed approach’s superiority in displacement reconstruction is validated through numerical simulations and experimental studies. The model’s operational capability on microcontrollers was evaluated through edge computing technology. Results demonstrate successful implementation of dynamic displacement measurement and trajectory restoration. Comparative analyses reveal that the model is more robust and makes more reliable predictions than conventional methods.