Numerical and Machine Learning-Based Triboelectric Nanogenerator Simulators: Contact-Separation Mode

Abstract

Triboelectric nanogenerators (TENGs) hold great potential as portable, cost-effective, and flexible energy sources. It is essential to understand in depth how the triboelectric properties of materials and operating conditions change TENG performance to improve their electrical outputs. In this study, the effects of various material parameters and operating conditions on the voltage, current, and power outputs of the TENGs are numerically investigated. The surface charge density improves the performance of the TENGs at all load resistances, while dielectric thickness, dielectric constant, surface area, and separation velocity are effective at medium and low load resistances. The separation distance, unlike all these, decreases performance at low load resistances. However, at high load resistances, it has the opposite effect and improves the performance. Furthermore, a broad range of data obtained from numerical simulations is used to train a machine learning-based TENG simulator. This simulator is based on a multilayer perceptron (MLP) model with an input layer of nine neurons, two hidden layers, one with nine neurons and the other with 55 neurons, and an output layer of three neurons for predicting current, voltage, and power. The MLP model, trained using TensorFlow, demonstrates high accuracy with R² values over 0.99 and achieves remarkably low mean absolute percentage error (MAPE) values of 4.22%, 3.35%, and 7.57% for current, voltage, and power predictions, respectively.