Dynamic modelling and experimental characterization of an elbow exoskeleton for the assessment of elbow post-stroke impedance

Spasticity is one of the positive symptoms that characterized the Upper Motor Neuron (UMN) syndrome, which occurs in about 30% of post-stroke, in up to 85% of multiple sclerosis and in 71% of spinal cord injured patients. Nowadays, the most effective treatment consists in the pharmacological approach, which requires a precise evaluation of the patient’s condition to define the exact drug dose schedule. However, the current clinical scales adopted in the spasticity assessment do not satisfy the requirements in terms of precision and reliability, mostly because of inherent subjectivity and intra- and inter-rater variability. Therefore, to aid clinicians and improve the spasticity classification and evaluation, the development of a quantitative method for enhancing the quality of spasticity evaluation is mandatory. This thesis project aims at performing a dynamic modelling and experimental characterization of an elbow exoskeleton (the NeuroExos Elbow Module – NEEM) as a starting point for the implementation of a novel method for the impedance estimation, as a proxy measure of spasticity, by means of a robotic device. The capability of the exoskeleton as assessment device has already been demonstrated in previous studies, where a set of biomechanical parameters were extracted from the robot data during imposed flexion/extension movements. By characterizing the dynamic behaviour of the exoskeleton, the user’s contribution to the movement can be isolated, thus improving the spasticity assessment via the extraction of biomechanical parameters that do not reflect the influence of the robot’s dynamics. Starting from Euler-Lagrangian equation, the dynamic model of the NEEM was determined and rewritten in linear form with respect to the parameters. Then, a parameter identification analysis was conducted, also including the contribution of the exoskeleton passive degrees of freedom. The parameter identification was carried out by building a dataset consisting of different trajectories acquired in static and dynamic conditions, for different velocities and different passive degrees of freedom configurations. A Least Square Error (LSE) minimization method was applied to firstly obtain an estimation of the gravity parameters starting from Quasi-Static (QS) data (5 deg/s). Then, the inertia parameter was computed from the dynamic data (> 5 deg/s). A validation procedure was conducted in order to analyse the goodness of fit of the model, on a separate dataset acquired in different dynamic conditions and different fixed spatial configurations of the elbow. The results in terms of percentage goodness of fit range between 98% and 84% showing that the model was able to predict the robot’s dynamics in all the workspace, with some non-predictable effects probably due to additional loads on the mechanical structure dependent on the elbow orientation, which worsen the quality of the torque reconstructions for specific configurations. The maximum error in terms of torque residuals is about 0.09 Nm on a maximum of 1 Nm recorded torque when the exoskeleton is not worn by any user. Considering that the torque range of a recording with the human in the loop is almost between 0 to 4 Nm, the error model of the NEEM can be considered acceptable for a potential clinical study. In conclusion, the system modelling and characterization can provide valuable information to pave the way for implementing a more precise spasticity assessment protocol in clinical applications.