Feature Encoded and Multi-Resolution Physics-Informed Machine Learning for Musculoskeletal Digital Twins

Explore a 47-minute conference talk on advanced machine learning techniques for musculoskeletal digital twin applications. Delve into the development of Feature-Encoded Physics Informed Parameter Identification Neural Network (FEPI-PINN) for simultaneous motion prediction and parameter identification in human musculoskeletal systems. Discover how this approach projects high-dimensional noisy surface electromyography (sEMG) signals onto a low-dimensional noise-filtered embedding space for effective forward dynamic training. Learn about the proposed Multi-Resolution Recurrent Neural Network (MR-RNN) learning algorithm, which enhances time-domain mapping using fast wavelet transform to decompose noisy sEMG signals into nested multi-scale signals. Examine the training process involving lower-resolution input signals and parameter transfer to higher-scale training. Gain insights into how these innovative approaches effectively identify subject-specific muscle parameters, yield accurate motion predictions of elbow flexion extension, and contribute to the construction of subject-specific musculoskeletal digital twin systems for health condition assessment and motion prediction.