Performances of Hill-Type and Neural Network Muscle Models—Toward a Myosignal-Based Exoskeleton

Abstract

Muscle models are the essential components of any musculoskeletal simulation. In addition, muscle models which are incorporated in neural-based prosthetic and orthotic devices may significantly improve their performance. The aim of the study was to compare the performances of two types of muscle models in terms of predicting the moments developed at the human elbow joint complex based on joint kinematics and neuromuscular activity. The performance evaluation of the muscle models was required to implement them in a powered myosignal-driven exoskeleton (orthotic device). The experimental setup included a passive exoskeleton capable of measuring the joint kinematics and dynamics in addition to the muscle myosignal activity (EMG). Two types of models were developed and analyzed: (i) a Hill-based model and (ii) a neural network. The task, which was selected for evaluating the muscle models performance, was the flexion–extension movement of the forearm with a hand-held weight. For this task the muscle model inputs were the normalized neural activation levels of the four main flexor–extensor muscles of the elbow joint, and the elbow joint angle and angular velocity. Using this inputs, the muscle model predicted the moment applied on the elbow joint during the movement. Results indicated a good performance of the Hill model, although the neural network predictions appeared to be superior. Relative advantages and shortcomings of both approaches were presented and discussed.