Validation and uncertainty of inverse dynamics analysis applied to high acceleration movements.

This thesis is motivated by the lack of knowledge of the uncertainty in the estimation of joint forces and moments derived through inverse dynamics analysis. Previous studies have shown uncertainty bounds can be substantial during slow, simple movements such as gait or lifting however little is known about the uncertainty in inverse dynamics solutions applied to high acceleration, open chain, complex tasks. A three dimensional full body model was used to provide a mechanical basis for evaluating joint forces and moments during the golf swing. Eight male skilled golfers were used; kinematic data was recorded using the Polhemus LIBERTY, an electromagnetic tracker system, using 12 sensors attached to the body with a specially designed jacket. Force plates were used to measure ground reaction forces.Validation of the derived joint forces and moments is problematic since no 'gold standard' is available for comparison. A comparison of the measured with the estimated ground reaction forces, as well as a comparison of the moments at the T8/T9 intervertebral joint that results from bottom up and top down mechanical analysis provided an initial measure of validity. The high acceleration, complex nature of the golf swing resulted in a reduced validity compared to previous studies concerned with lifting, fast trunk rotations and slow speed golf swings. The residuals between the measured and predicted GRF were greatest during the downswing. Similarly, the residuals between the joint reaction forces and moments at the upper trunk joint measured using a top down and bottom up mechanical analysis were greatest during the downswing, exemplified by an increase in joint moment RMS differences of 30.9 Nm, 24.4 Nm and 25.2 Nm for lateral bending, axial rotation and flexion-extension respectively. It was shown that for open chain movements, through periods of high acceleration, inverse dynamics solutions can be subject to errors which have the capacity to significantly affect the interpretation of resultant joint moments depending on whether a top down or bottom up mechanical analysis is used. Top down-bottom up comparisons do not account for two sources of error; the joint centre location and the anatomical coordinate system of the joint where the two models meet. A further drawback associated with these validation methods is that nothing can be learnt about the individual sources of error and how they contribute to the total residual error.A consideration of how errors in measured variables propagate through inverse dynamics equations to produce uncertainties associated with the result was necessary. To analyse this, the Taylor Series Method for error propagation was used. Inaccuracies in body segment parameters, kinematics and external force measurement were determined experimentally. Soft tissue artefact and joint centre location errors were extracted from the literature. Inaccuracies in variables were assumed to be random and uncorrelated and results were representative of the upper bound uncertainty. Uncertainty in joint moment estimations was greatest for downswing where segments were moving with the greatest acceleration. The magnitude of the uncertainty was substantial and ranged from 6-339% of the peak joint moment magnitude.Inaccuracies in proximal moment arms and centre of mass accelerations had the most influence on the joint moment uncertainty and this uncertainty had the capability to alter the timing of peak joint moments by as much as 560ms. The results were critical to the interpretation of inverse dynamics derived joint forces and moments for high acceleration, open chain motions.