# Design and optimization of soft finger actuators for rehabilitation applications: A combined finite element and neural network approach

## TL;DR

This paper uses simulations and neural networks to optimize soft finger actuators for rehabilitation gloves, improving their bending performance.

## Contribution

A combined finite element and neural network approach is introduced to optimize soft finger actuator design for rehabilitation.

## Key findings

- Increasing actuator height and bellows number significantly improves bending angles.
- Greater foot and surrounding thicknesses restrict bending, requiring careful design.
- Neural networks predict bending angles with 99.26% explained variance, validating the simulation approach.

## Abstract

This study presents a comprehensive analysis of soft finger actuators using finite element modeling to assess their performance in various structural configurations. By conducting detailed numerical simulations, we explore how variations in structural parameters influence the bending angle, thereby guiding iterative design improvements. Specifically, the research examines the impact of critical design factors, such as the number of bellows, actuator height, surrounding thickness, and foot thickness, on the bending behavior of soft actuators. The objective is to optimize these actuators for use in rehabilitation training gloves, where precise motion control is para-mount. Our findings reveal that increasing both the height and the number of bellows significantly enhances the achievable bending angle, facilitating more effective rehabilitation exercises. In contrast, greater foot and surrounding thicknesses exhibit a restrictive effect on bending, underscoring the need to carefully consider these parameters in design processes. These insights are instrumental in formulating design guidelines that aim to optimize actuator performance in therapeutic applications. Crucially, the manuscript presents a rigorous comparison between the experimental results and simulation results, demonstrating a high degree of concordance that validates the FEM approach and the predictions of the neural networks. This close match between the observed and predicted data not only confirms the reliability of the simulations, but also enhances the credibility of the design recommendations for rehabilitation applications. Furthermore, the study uses artificial neural networks to predict bending angles with high precision. With a residual variance of just 0. 74% and an explained variance of 99. 26%, the neural network model demonstrates exceptional predictive capacity, highlighting its potential as a tool for further refinement of the design and optimization of the performance of soft actuators. This research not only advances our understanding of soft actuator mechanics, but also contributes to the development of more effective rehabilitation technologies.