<![CDATA[Piezoelectric energy conversion systems are essential renewable energy harvesting solutions for autonomously powering Internet of Things devices. Piezoelectric ceramic elements are highly susceptible to functional degradation due to cyclic mechanical stress exposure that triggers material fatigue. This study proposes a smart computational predictive maintenance approach to detect system anomalies in real time. The experiment acquires transient electrical parameters namely voltage and current, as well as spatial mechanical vibrations from a piezoelectric matrix prototype. Given the high cost of physical destructive testing, a dataset of five thousand samples was synthesized using an empirical data augmentation approach based on hardware ground truth. The Random Forest Classifier algorithm was implemented for binary classification of normal and anomaly system conditions. Independent testing results demonstrated that the model achieved an accuracy of 94.6 percent, precision of 92.49 percent, sensitivity of 94.71 percent, and an F1 score of 93.59 percent. The predictive error margin of approximately seven percent proves the model robustness in accommodating highly realistic physical ambiguous zones. The model is algorithmically consistent with material physics laws by prioritizing voltage features and vertical acceleration, while ignoring microampere current fluctuations. This performance solidifies the viability of ensemble architecture as a structural health monitoring system in edge computing networks. ]]>
Copyrights © 2026
