Physics-Informed Inverse Neuromechanical Estimation Dataset for Neural Delay Recovery

Description

Dataset Creators and Affiliation: Jakaria Habib Department of Information and Communication Engineering, Pabna University of Science and Technology, Bangladesh. Overview and Purpose: This repo holds the simulation data, trained ML models and associated python analysis scripts to support physics-informed inverse neural activation delay estimation in musculoskeletal dynamics. The data was generated during dynamic gait simulations using OpenSim v4.5 with the 3DGaitModel2354 lower-limb musculoskeletal architecture. Neural activation delay (electromechanical delay, EMD), the time that elapses between the arrival of central motor excitation impulse and mechanical release in pre-contraction period, is a primary control characteristic behind neural coordination and postural stability. This dataset presents a deterministic app state space and machine learning framework that may be applicable to the inverse parameter identification problem: recovering latent neural delay parameters directly from observable biomechanical signals. Dataset Structure and Variable Specifications: The dataset consists of 45,445 observations (lower-limb joint kinematics and Hill-type muscle activation trajectories) across five physiological neural delay conditions varying from 0.015 seconds to 0.060 seconds (15 ms to 60 ms). The input features include seven biomechanical variables including the right knee flexion angle (radians), right knee angular velocity (rad/s), right hip flexion angle (radians), right hip angular velocity (rad/s), rectus femoris muscle activation (0 to 1), vastus intermedius muscle activation (0 to 1) and biceps femoris muscle activation (0 to 1). The target variable being the latent neural delay parameter tau. Perturbation Validation and Trained Artifacts: Besides nominal baseline simulations, a validation dataset of simulated femur shapes generated by applying a structural 7 percent increase in femur mass is also available to test the performance of the continuous mapping approximation and its resilience to anatomical variations. Files are categorized by baseline simulation data, perturbation validation data, and artifacts from trained models such as the Random Forest regressor and associated scaler.

Steps to reproduce

1. Extract the dataset files from the compressed archive. 2. Load the baseline or perturbation simulation Excel files containing the 7 biomechanical state variables. 3. Execute train_tau_inverse_model.py using Python 3.10+ with scikit-learn, pandas, and numpy to train or evaluate the Random Forest inverse regression model.

Institutions

• Pabna University of Science and Technology

  Rajshahi Division, Pābna

Categories

Licence