A model based approach for evaluating human neuromusculoskeletal system performance

In the thesis, a model based approach is proposed for monitoring the performance of a human neuromusculoskeletal (NMS) system. It utilizes a linear dynamic model with exogenous inputs (ARMAX model) to link multiple features extracted from surface electromyographic (sEMG) signals as model inputs, and measurable physiological outputs, such as forces produced by the limbs or limb velocities, as model outputs. This multiple-input and multiple-output (MIMO) model is then utilized to quantify and track changes in the NMS system dynamics over time. The changes in NMS system dynamics were modeled using distance between the distribution of 1-step ahead model prediction errors observed at the beginning of the exercise, when the subject was rested, and 1-step ahead prediction errors observed at any other time during exercise. The distance, referred to as the Freshness Similarity Index (FSI), was expressed via the Kullback Leibler (KL) divergence measure between the aforementioned distributions of 1-step ahead prediction errors. As the subjects proceeded with their exercises and got increasingly tired, the modeling errors were expected to increase, leading to an increase in FSI. Such behavior of FSIs enables it to act as a quantitative measure of the level of changes in NMS system performance, in other words, as a measure of NMS system performance degradation due to fatigues. The methodology has been evaluated on two data sets, one collected from an activity related to lower limb muscles and the other collected from temporomandibular joint (TMJ) muscles. In both cases, an increasing trend in the FSI clearly illustrated changes in NMS system performance, as exercise progressed. Furthermore, after rest, FSI observed in both exercises recovered to their original levels, quantitatively and meaningfully showing that the corresponding NMS systems of the two subjects indeed rested.