Optimization of a macrofiber piezoelectric energy harvester using artificial neural networks

Abstract

This paper addresses the challenge of finding the optimal design parameters for a piezoelectric energy harvester. It presents an advanced simulation-driven optimization approach to determine the optimal geometric and circuit configuration for maximizing energy harvesting efficiency. Additionally, it enhances power output at an excitation frequency within the 10–100 Hz range, which is commonly found in the environment. The harvester consists of a cantilever beam partially coated with a macrofibre composite (MFC) piezoelectric patch, connected to a resistance load and subjected to base excitation. The optimization platform is built upon both analytical and finite element (FE) models of the energy harvesting system. For beams with a large aspect ratio (length/width), the analytical model based on Euler-Bernoulli beam theory is used, while for those with a small aspect ratio, a 3D FE model is employed to simulate the entire energy harvesting process. This approach enhances the accuracy of piezoelectric energy prediction. Due to the high computational cost and the significant time and memory required for running numerous simulations to evaluate the objective function (OF) used in the optimization, a more efficient solution is implemented based on a Neural Networks (NNs) model. Initially, the NNs is trained using a dataset derived from simulations and its performance and accuracy are rigorously assessed through various statistical methods. Once trained, the NNs serves as a surrogate model for OF evaluation, allowing for more efficient black-box optimization via a Genetic Algorithm (GA). Finally, a thorough analysis of the optimal design parameters obtained from the optimization process is conducted.