Modeling and identification of human musculoskeletal walking system

Abstract

Several methods are tested to identify the human musculoskeletal system both as a linear and nonlinear system. For the linear system approach, a MIMO (multiinput, multioutput) ARX (autoregressive with exogeneous inputs) model is first tested to get a rough estimation of the system structure and parameters. A general linear input-output MIMO model is then developed, and parameters are estimated by means of the prediction error identification method. Since the complex human musculoskeletal system is almost certainly a nonlinear system, nonlinear system identification is applied and polynomials are used to approximate the nonlinear system functions. For such a MIMO nonlinear system, the parameters to be estimated will number in the thousands or even millions, depending on the polynomial degrees used and the maximum orders of delays. To overcome such numerical difficulties, a forward-regression orthogonal method is used to select only the most significant terms and estimate the corresponding parameters.<>