On the Use of Mesh-Based Joint Contact Models Within Simulations Using Automatic Differentiation

Abstract

Biomechanical simulations integrating computations of joint contact pressures within optimizations typically use mesh-based contact models. However, these models are usually not continuous, which could make convergence difficult. Moreover, they can be difficult to tackle when using expression graphs to calculate derivatives (e.g., when using automatic differentiation) during the resolution of optimization problems. This is due to the computational need to use branches when dealing with conditionals. This study presents a mesh-based contact model adapted to deal with gradient-based optimizations and automatic differentiation. A tracking example of a knee prosthesis contact, integrated into a full-body muscle-driven model, is presented. Kinematics and dynamics data were tracked during an overground gait trial. The results show that the accuracy of knee contact force estimations is comparable to other studies (RMSE medial contact force = 50.8 N, RMSE lateral = 81.6 N), and the computational time (3 h and 20 min) is acceptable to be run in a conventional PC. Future work is oriented to formulate predictive simulations of muscle-driven full-body models to predict kinematics and dynamics data of subjects post-surgery simultaneously.