Machine Learning-Based Predictive Modeling of Piezoelectric Energy Harvester for Charging Pacemakers

This article represents a T-shaped beam-mass construction as an alternative to conventional cantilevered beams for harvesting piezoelectric energy. In order to facilitate the design process and determine the optimum physical dimensions, machine learning is used to model the architecture. It could be an effective way to evaluate the parameters of the piezoelectric energy harvester (PEH), which is complicated to do with the finite element method (FEM) because it requires a lot of time and processing power. First, using FEM, a dataset consisting of 325 samples is created for the purpose of training a machine learning model. The validation results indicate that the trained machine learning model can achieve around 97% estimation accuracy of device features, such as resonant frequency and generated power. The tests are conducted by changing the base length( \({L}_{b})\) , base width( \({W}_{b})\) , clamped width \({(L}_{clamped})\) , and clamped depth \({(W}_{clamped})\) . The results reveal that input features have significant effect on vibrational response of T-shaped PEH. The performance of four different ML algorithms including K-nearest neighbor (KNN), decision tree regressor (DTR), random forest (RF), and XGBoost (XGB) on predicting Eigenfrequency ( \({f}_{eigen})\) and power ( \({P}_{gen})\) were compared. Among them, the XGB model provides highest accurate predictions of \({f}_{eigen}\) and \({P}_{gen}\) within the test range of \({L}_{b}\) , \({W}_{b}\) , \({L}_{clamped}\) , and \({W}_{clamped}\) . Using \({f}_{eigen}\) and \({P}_{gen}\) for a piezoelectric energy harvester to power a pacemaker with an R2 value of more than 90% and much less time is a significant alternative to traditional numerical simulation techniques in terms of computational resources and the complexity of rigorous modeling by means of trial and error.