Experimental and Machine Learning‐Based Parametric Study of Free‐Bending Dielectric Elastomer Actuators

Abstract

Recent advances in electroactive polymers and smart materials have established dielectric elastomer actuators (DEAs) as a viable platform for lightweight, large‐deformation soft actuation. This study presents a parametric investigation of dielectric elastomer bending actuators, focusing on the influence of actuator geometry and prestretch ratio (λ) on actuation performance. Bending actuators were fabricated using 3 M VHB 4905 dielectric elastomer films with varying geometric configurations and stretch conditions, and tested experimentally to evaluate bending behavior and identify practical fabrication and operational challenges. A data‐driven classification approach was also explored as a preliminary means of screening actuation reliability. A linear support vector machine (SVM) classifier was trained on 23 fully documented actuators using nine geometric and material features, and benchmarked against four alternative classifiers under leave‐one‐out cross‐validation. The linear SVM identified actuation outcomes with 61% overall accuracy, with feature weights qualitatively consistent with the experimentally observed geometric trends. Given the small, imbalanced dataset, these results are preliminary evidence that geometry carries information about actuation outcome, rather than a validated predictive tool. The present work characterizes only the free out‐of‐plane bending of individual actuators; gripping force, payload capacity, and object manipulation were not evaluated and are identified as directions for future work.