Musculoskeletal computer models are useful for estimating internal quantities that cannot be measured experimentally, designing new medical devices and rehabilitation approaches, and predicting the outcome of surgical procedures. Unfortunately, use of articular contact in such models makes computational speed a limiting factor, rendering dynamic simulations either completely intractable or else so slow that optimization is impossible. Lacking of articular contact will significantly affect many contact quantities predicted from musculoskeletal computer models (e.g., muscle forces, ligament strains, bone loads, and cartilage and implant wear).

Computational limitations in other engineering disciplines have been overcome through the use of surrogate modeling. Surrogate-based modeling involves fitting a model to a model, where the original model is computationally expensive and the surrogate model is computationally cheap. The sample data points used to fit the surrogate model are generated by running the original model repeatedly with different combinations of input variables. Once fitted, the surrogate model can be used in place of the original model in simulations or optimizations to eliminate computational cost as a limiting factor. Though surrogate modeling techniques have successfully eliminated computational bottlenecks in other fields, they have not yet been applied to articular contact problems.

Primary objectives were two-fold. First, we present a computational evaluation and practical application of the proposed surrogate-based modeling approach using dynamic wear simulation of a total knee replacement constrained to both two- and three-dimensional motions in a Stanmore machine. The sample points needed for surrogate modeling fitting are generated by an elastic foundation contact model. Second, we apply the surrogate contact model into the musculoskeletal model to simultaneously calculate medial and lateral tibiofemoral contact forces, patellofemoral contact forces, and the muscle forces. This project provides a computationally efficient way to apply different types of contact models to predict physiologically significant contact quantities.