|ijphm_16_015.pdf||5.99 MB||July 19, 2016 - 7:24pm|
This paper presents a system-based method for monitoring a human neuromusculoskeletal (NMS) system. It is based on autoregressive models with exogenous inputs, which link surface electromyographic signals and joint kinematic variables in order to detect changes in system dynamics, as well as to assess joint level and muscle level contributions to those changes. Instantaneous energy and mean frequency of time frequency distributions of electromyographic signals were used as model inputs, while angular velocities of the monitored joints served as outputs. Slow temporal changes in the behavior of the entire system or individual joint models were tracked by analyzing one-step ahead prediction errors of the corresponding models over time. Finally, analysis of the recursively updated models, which tracked the NMS dynamics over time, was used to characterize these changes at the joint and muscular levels. The methodology is demonstrated on data recorded from 12 human subjects completing a repetitive sawing motion until voluntary exhaustion. Statistically significant decreasing trends in the similarities of the NMS models to those observed in the rested state were observed in all subjects. In addition, decreased joint response to muscle activity, as well as changes in the coordination and motion planning have been detected with all subjects, indicating their fatigue.