Repetitive control and electrode array pattern selection for FES-based drop-foot assistance

Abstract

Drop-foot is a problem resulting from a range of neurological conditions. It is currently diagnosed in over 3 million people worldwide. Current technologies addressing Dropfoot have significant limitations, with 30% of people rejecting the established mechanical solutions due to; muscle wastage, size, noise and damage to dignity. Alternatively, Functional Electrical Stimulation (FES) is a technology, which addresses the condition by directly recruiting the user’s muscles. This overcomes significant issues such as muscle wastage or restriction of freedom. However, it has limitations which include increased rate of fatigue in users, dependence on pad positions, and limited battery life. Further to this, the control of FES commercially is very crude, often leading to jerky or unnatural motion. More advanced controllers exist in the academic domain, but they assume that gait can be modelled as a resetting motion, require large amounts of data, or are unsuitable for clinical application due to problems in demonstrating stability. This thesis focuses on overcoming the limitation of current FES technology for dropfoot. It develops, simulates and applies a new form of closed-loop controller for FES that addresses the limitations of previous systems. In particular, this controller is a form of Repetitive Control (RC), which learns over the periodic nature of gait, does not assume resetting, reduces the data needed between periods, and has clear stability conditions. Simulations show that the ‘point-to-point’ generalisation of RC improves the convergence speed and robustness while only slightly decreasing the tracking accuracy of the entire reference. Experimental validation confirms that tracking accuracy is improved between 52% − 140% compared to existing drop-foot controllers, with a reduction in data used of 94%, compared to traditional RC. To address the challenge of pad placement, a framework to identify an optimal pad ‘pattern’ from thousands of pad combinations has been developed. Iterative learning control is then applied to tune the simulation levels for a static gesture. This was achieved with an error of 2.2 ◦ ±1.23◦ . Applying commercially inspired constraints restricted the pool of patterns and enabled an investigation into hardware and software simplifications for a commercial device.

More information

Identifiers

ORCID for Christopher Freeman: orcid.org/0000-0003-0305-9246

Catalogue record

Export record