Comparative Analysis of Evolutionary Algorithms to Improve the Dynamic Performance of a Lower Limb Model

Accurate biomechanical models of the human body were essential for understanding, predicting, and simulating human movement. While complex, multi-segment models offered high fidelity, their computational cost often limited their practical application. This study focused on optimizing a simplified three-link lower limb model to enhance its dynamic performance by incorporating impact of ground reaction forces while maintaining computational efficiency. To address the model's limitations in capturing complex human movements, evolutionary algorithms were employed to refine its parameters. The Ant Lion optimizer, Cuckoo search, Dragonfly algorithm, and Fminsearch algorithm were utilized to determine optimal values for link lengths, mass distribution, and joint stiffness. The model performance was assessed by comparing simulated lower limb with ground reaction forces, joint torques, and angles through experimental data from dynamic tasks. The fitness functions were multi-objective in nature to simultaneously minimize the three lower limb angle prediction errors. The Ant Lion optimizer demonstrated a significant advantage in terms of convergence rate and dynamic model parameter optimization with minimal root mean square error. While the Fminsearch algorithm caused overfitting of parameters, making it unsuitable for the current application, it could be useful in hybrid optimization techniques. This research aimed to identify the most suitable optimization algorithm for improving model accuracy and to explore the trade-offs between model simplicity and dynamic fidelity.