Model-based control of FES embedding simultaneous voluntary user effort

Abstract

There are over 1.2 million people in the UK with upper or lower limb impairment following stroke. Artificial activation of muscle can be achieved using functional electrical stimulation (FES), which is the most prevalent assistive technology used in the rehabilitation of stroke patients. Significant clinical research shows that effective treatment requires electrical pulses to be delivered to muscles in a manner which supports voluntary effort, during performance of repetitive rehabilitation tasks. This has motivated using electromyography (EMG) to measure the voluntary contraction and then to control electrical stimulation supplied to impaired muscles. However, existing FES control schemes using EMG are predominantly open loop and fail to provide accurate assistance to achieve the intended movement. In this thesis, a model of dynamic interaction between voluntary and evoked muscle activation is initially developed, embedding both nonlinear recruitment and activation dynamics. This enables the proposed model-based, hybrid EMG/FES control scheme to be derived, allowing the dual objectives of tracking and volitional intention support to be optimised. Extension of the model-based EMG/FES structure to embed iterative learning control (ILC) is then undertaken in order to augment the tracking accuracy by learning from experience to update the control action. Experimental results show that the identification scheme is accurate and suitable for clinical application. Further results show that the model-based ILC framework using hybrid activation reduces the tracking error by between 27% and 70% compared to previous FES approaches which neglect voluntary action. Identifying a model of EMG/FES muscle activation is time-consuming and accuracy is degraded by fatigue and other time-varying properties. To address these issues, an adaptive control scheme is then developed termed ‘estimation-based multiple model ILC’ (EMMILC). This control framework automatically selects a controller based on the most appropriate model chosen from an underlying set of ‘candidate models’. A design procedure is proposed to generate a set of models based on distributions of fatigue tests. Results indicate that EMMILC framework has a significant improvement in tracking between 50% and 112% despite the changing model over time due to fatigue effect compared to standard ILC designed using a single model. Additional results confirm the potential of the proposed framework to be applied without the need of model identification.

More information

Identifiers

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

Catalogue record

Export record