An autoregressive modeling approach to analyzing wheelchair propulsion forces.

Biomechanical signals collected during wheelchair propulsion are often analyzed by computing averages and/or peak values over several strokes. Due to the complex nature of the signals, this type of analysis may not be specific to identifying factors that may predispose a wheelchair user to joint pain/injury. Hence, a new technique is introduced that uses a system identification approach, autoregressive (AR) modeling, to analyze wheelchair propulsion force waveforms. In this application an AR method was used to create a model force waveform based on current and past values of digital pushrim force data. The feasibility of the AR modeling method over point-wise methods to detect asymmetry among force waveforms was tested with a group of 20 wheelchair users. Subjects propelled at a constant 0.9 m/s on a roller system during which 20 s of force data were collected from the SMART(Wheel)s, force and torque sensing pushrims. Both methods showed that the wheelchair users as a group propelled evenly, however, individual analysis using the AR model error estimates indicated that twenty-five percent demonstrated significant asymmetry in their force waveforms. If only point-wise means and variances of the applied bilateral forces were considered, most subjects would have appeared symmetrical. Thus, the AR modeling approach is more sensitive to detecting anomalies in propulsion technique.