Concurrent Simulation of Neuromuscular Control and Tissue Deformation Is Possible by Adaptive Surrogate Modeling

Computational biomechanics has largely consisted of two distinct modeling domains, finite element (FE) analysis [1] and multi-body dynamics. Due to computational efficiency, muscle driven musculoskeletal models (or multi-body forward dynamic models) have been the primary method used in prediction of movement patterns [2]. FE methods, while computationally expensive, have the ability to yield important soft-tissue information such as stress [3] and coupling the two domains would yield more complete and useful simulations. Dynamic FE models have been successfully used for passive movement predictions but computational cost makes them unfeasible for movement optimization, where thousands of simulations may be required to find an optimal solution. Surrogate modeling is one possible approach to utilize a database of previous FE results to overcome the relatively high computational cost of FE simulations with different inputs. In general, surrogate modeling methods can be classified as global or local. Global methods fit a regression model to the complete set of input/output pairs whereas local methods include regression within a neighborhood of data points.