Sensorless current estimation in single, series, and parallel Piezoelectric Energy-Harvesting networks using deep learning

The dataset used in the study is a large-scale collection of 78,155,000 entries generated through controlled experimental measurements of single, series, and parallel configurations of two S118-J1SS-1808YB piezoelectric devices. Data were acquired at a sampling rate of 10 kHz under systematic parameter variations: frequency sweeps from 0 to 200 Hz (in 1 Hz increments), and load-resistance sweeps from 10 k$\Omega$ to 300 k$\Omega$ (in 5 k$\Omega$ increments). The foundational variables captured include the external load resistance ($R_{L}$), internal capacitance ($C_{p}$), leakage resistance ($R_{p}$), and the voltage across the elements ($V_{O}$), which were processed using a Savitzky-Golay filter to compute their temporal derivatives ($\mathrm{d}V_{O}/\mathrm{d}t$) with reduced noise. Finally, the theoretical current (ip) was computed analytically from these measurements to serve as the ground-truth target variable for training the predictive regression models. The experimental setup and instrumentation used to acquire the signals include the following equipment: Analog Waveform Generator: A GFG-3015 model is used to supply a controlled sinusoidal excitation signal with an amplitude of 5 $V_{p-p}$. Amplifier: A Class A amplifier is connected to the generator to amplify the signal, ensuring it provides the required voltage and current capability for the load. Piezoelectric Transducer: The setup uses an S118-J1SS-1808YB piezoelectric device (fabricated from PZT-5J) operating at a resonant frequency of 130 Hz. Oscilloscope: A Tektronix TDS 3012 oscilloscope is used to monitor the excitation waveform and capture both the excitation and response signals reliably throughout the experiments. Additionally, the acquisition is performed under strictly regulated environmental conditions, maintaining a temperature between 20 and 22 $^\circ$C and an atmospheric pressure of 1024 hPa.