Poster
in
Workshop: D3S3: Data-driven and Differentiable Simulations, Surrogates, and Solvers
Modelling variation in the forward EMG model.
Dimitrios Chalatsis · Alexander Clarke · Dario Farina
Keywords: [ forward emg modelling ] [ neural surrogate solvers ] [ biophysical simulation ]
Traditional finite element methods (FEM) face limitations when simulating dynamic, naturalistic movements due to the need for recalculations as muscle geometry changes. Our approach leverages PINNs to implicitly represent a continuous range of solutions for the volume conduction equation, parameterized by the pinnation angle of muscle fibers. We demonstrate that our method significantly reduces model size and computational time while maintaining high accuracy. The neural network model generalizes well across a range of pinnation angles, offering a promising solution for efficient and dynamic EMG simulations.