PZT Cantilever for Energy Harvesting and Vibration Sensing

The proposed two-stage ensemble machine learning model aims to bridge the gap between energy harvesting and vibration sensing applications for lead zirconate titanate (PZT) and similar piezoceramic materials by enabling one device to perform both functions simultaneously. Two PZT cantilever configurations were tested: one without a tip mass for maximum linearity at low frequencies and one with a tip mass for maximum energy output. The highest absolute prediction error on the testing set is 19% and 7%, respectively. While the R-2 score remained nearly 1, the PZT cantilever with the tip mass showed an 11% lower mean absolute error (MAE) and 38% lower mean squared error (MSE) compared to the PZT without, suggesting that PZT cantilevers in energy harvesting configurations can be used to predict acceleration with acceptable accuracy.