Deep Learning-Based Optimization of Piezoelectric Vibration Energy Harvesters

Advances in energy harvesting technologies present prospective concepts to capture and store energy from the environment and use it to power sensors used in Structural Health Monitoring (SHM) systems. Among many others, ambient vibrations are a ubiquitous source of energy that has the potential to charge low-powered sensors attached to aircraft structures. This study aims at designing a vibrational-based energy harvesting system consisting of Macro-Fiber Composite (MFC) patches bonded to a cantilever beam with optimal design parameters. As a base model, an electromechanically coupled Finite Element (FE) model is first developed to predict the open-source voltage when subjected to input excitation, which is validated using previous experimental data. Subsequently, the harvested power is found by simulating an electrical circuit consisting of a full-bridge rectifier and an external capacitor, using Electronic Design Automation (EDA) simulation. A Deep learning-based optimization is proposed to calculate the optimal mechanical and electrical parameters, resulting in the maximum number of resonant peaks within a specified frequency range, and also to maximize the power generated from higher-order resonant peaks. Using the developed FE model, a large number of data is generated to train a Deep Neural Network (DNN), which has the capability to find the optimal design parameters for the specified objective. This approach aims at replacing conventional optimization techniques and to obtain an optimal design of broadband vibrational-based energy harvester in a more computationally efficient manner.