Machine Learning-Driven Surrogate Modeling for Optimization of Triboelectric Nanogenerator Design Parameters

Abstract

Triboelectric nanogenerators (TENGs) offer a promising solution for energy harvesting in wearable devices and sensors. However, their energy output is dependent on process parameters and should be optimized to maximize performance. Due to the absence of effective analytical models for TENG systems, the complex relationship among these variables and the effect of these variables cannot be easily boiled down into a conventional theoretical framework. To address this problem, this study takes four process parameters such as thickness, pore ratio, applied force, and frequency into account and leverages advanced design methods (e.g., Design of Experiment) and machine learning-based regression models to systematically explore the design space. A contact-separation TENG has been designed that includes a tribonegative porous layer of graphene nanoplatelets (GNP) dispersed into polydimethylsiloxane (PDMS) matrix and aluminum as the tribopositive material. Several experiments are conducted to train a support vector regressor (SVR) model, validate the predicted performance, and refine the design that can be further used to obtain an optimized TENG design.