System Based Monitoring of a Neuromusculoskeletal System Using Divide and Conquer Type Models

This paper presents a system-based methodology for monitoring the performance of a human neuromusculoskeletal system. The algorithm is based on a divide-and-conquer type modeling strategy using distributed autoregressive models with exogenous input to link surface electromyographic signals and joint kinematic variables. Instantaneous energies and mean frequencies of electromyographic signals were extracted over time from their reduced interference time frequency distributions. These features were used as inputs into the model, while angular velocities of the monitored joints formed the vector of outputs of these models. Performance of the monitored system quantified by modeling and tracking changes in prediction errors of the corresponding model over time. The methodology is demonstrated on data recorded from 12 human subjects completing a repetitive sawing motion until voluntary exhaustion. It was found that 100% of subjects displayed statistically significant drifting in the model error distributions, suggesting fatigue was developing within all subjects considered in this study.