Experimental and Machine Learning-Based Optimization of Load Resistance in an MFC Vibration Energy Harvesting System

초록

This study investigates the electromechanical behavior and power generation characteristics of a macro-fiber composite (MFC)-based vibration energy harvesting system by integrating experimental analysis with machine learning-based prediction. A cantilever-type harvester composed of an MFC actuator (P1 type) and MFC energy harvester (P2 type) was fabricated, and vibration was induced using high-voltage excitation under resonant conditions. Open-circuit measurements demonstrated that the output voltage increased monotonically with the actuator input voltage due to enhanced structural vibration and mechanical deformation. Under closed-circuit conditions, the RMS voltage increased with load resistance, while the harvested power exhibited a peak-shaped profile, reaching a maximum at load resistance between approximately 100 Omega and 22 k Omega depending on the excitation amplitude. To predict the harvested power and identify the optimal load resistance, three regression models-random forest, least-squares boosting (LSBoost), and Gaussian process regression (GPR)-were trained using experimentally collected RMS voltage and power data. While random forest and LSBoost captured general trends, they showed considerable errors near the peak power region. In contrast, GPR most accurately reproduced the nonlinear relationship between power and load resistance, achieving the lowest RMSE (0.0926). GPR-based predictions enabled precise estimation of the optimal load resistance for each input voltage, with a maximum predicted power of 4.37 mu W at 21.8 k Omega under 400 V excitation-closely matching the experimental results.

키워드