TY  - EJOU
AU  - Gong, Jianwei 

TI  - A Physics-Driven Surrogate Modeling Framework for the Inverse Design and Optimization of Dielectric Elastomer Tube Energy Harvesters
T2  - Energy Engineering

PY  - 
VL  - 
IS  - 
SN  - 1546-0118

AB  - Dielectric elastomer tube energy harvesters have complex nonlinear electromechanical coupling, making parameter optimisation challenging when relying solely on experiments or finite element simulations. In this study, we present a physics-driven machine learning framework for parameter optimisation of such devices. A vectorised numerical solver based on hyperelastic constitutive relations, Maxwell stress, and gas adiabatic expansion is implemented to generate a synthetic dataset of 5000 physically consistent samples. Using four key geometric and operational parameters as inputs, several machine learning models are trained to predict energy conversion efficiency. A stacking ensemble model achieves a five-fold cross-validation R<sup>2</sup> of 0.992, while the ANN reaches an R<sup>2</sup> of 0.997 on the independent test set. Bayesian optimisation is then applied for single-objective efficiency maximisation and multi-objective trade-off between efficiency and wall thickness. SHAP analysis is used to interpret the contribution of each input parameter. The proposed framework provides a digital design workflow that can accelerate parameter exploration for dielectric elastomer tube harvesters without requiring extensive experimental data.
KW  - Dielectric elastomer; tube energy harvester; machine learning surrogate model; Bayesian optimisation; parameter optimisation; physics-informed modelling

DO  - 10.32604/ee.2026.088854