Parameter identification and sensitivity analysis of a lower-limb musculoskeletal model

The estimation of joint torque based on wearable sensors is an important content in human-robot interaction research. Despite existing joint torque estimation models providing high accuracy, their application in robotic control is limited due to the number of sensors and real-time output requirements. To address this issue, this paper establishes a knee joint torque estimation model driven by four electromyography (EMG) sensors and proposes a novel method for simplifying musculoskeletal models based on sensitivity analysis. To achieve this, this paper combines multiple advanced Hill-type muscle model components to establish a knee-joint musculoskeletal model that includes four major muscles and employs the genetic algorithm (GA) to identify the model parameters. Then, Sobol's global sensitivity analysis theory is used to analyze the influence of parameter variations on model outputs, and a sensitivity-based model simplification method is proposed. In addition, a lower-limb physical and biological signal collection experiment without ground reaction force is designed for parameter identification and sensitivity analysis. Finally, based on experimental data from several test subjects, the parameters of each individual's musculoskeletal model are identified and evaluated, and the sensitivity index of each parameter is calculated to determine the influence of the number of model parameters on the identification performance. The results indicate that the proposed musculoskeletal model can provide individuals with comparable normalized root mean square error (NRMSE) through parameter identification, and the sensitivity-based model simplification method is effective.