Multiple model control of functional electrical stimulation electrode arrays

Abstract

Functional electrical stimulation (FES) is an upper limb stroke rehabilitation technology that can enable patients to recover their lost movement by assisting intensive and goal-oriented task training. Unfortunately, existing commercial FES devices using single-pad electrode cannot provide selective muscle activation, hence their tracking accuracy is limited. Electrode arrays combining multiple pads in a single structure have recently been developed, and can more accurately assist wrist and finger movements. However, the set-up procedures currently used to locate the best stimulation sites are very time-consuming, and not suitable for a home use scenario. Their accuracy is also limited as they are predominantly open-loop. To date, Iterative Learning Control (ILC) has achieved the best performance for FES array tracking control tasks. Unfortunately it requires a large number of model identification tests that slow down the training, and must be repeated for different desired trajectories. All these drawbacks lead to prohibitive inconvenience for users and prevent translation to clinical or home environments. To address this, an estimation-based multiple model switched iterative learning control (EMMILC) framework is proposed. This combines the most successful adaptive and learning properties of existing FES controllers employed for single pad systems. A novel multiple-model design procedure guaranteeing robust performance is developed,and initial experimental results using single-pad electrode results are then presented to confirm efficacy of the approach. Experimental results show that EMMILC can reduce tracking error to 20% of its initial value within five trials, and maintain the same level of error in the presence of pronounced muscle fatigue. This architecture outperforms the standard ILC approach, and confirms the fundamental proof of concept. The EMMILC approach is then extended for application to electrode array technology, and simulations using a realistic model confirm significant improvement compared with existing controllers.

More information

Identifiers

ORCID for Junlin Zhou: orcid.org/0009-0003-9888-9759

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

ORCID for Ann-Marie Hughes: orcid.org/0000-0002-3958-8206

Catalogue record

Export record