Solar Panel Battery Level Measurement System: Hardware Implementation, Measurements and Instrumentation, and Machine Learning-Based Battery Degradation Analysis

This paper presents a low-cost, solar-charged battery monitoring system developed as a Measurements and Instrumentation course project. A 6 V, 2 W photovoltaic panel charges an 18650 lithium-ion cell through a TP4056-based constant-current/constant-voltage charging module. The prototype has two sensing channels: an INA219 current/power monitor on the I2C bus, which measures bus voltage, charging current and power, and a resistive divider (R1 = 10 kΩ, R2 = 1.5 kΩ) into the 10-bit ADC of an Arduino UNO R3. The firmware reads the INA219 once per second, shows voltage, current and power on a 16 × 2 I2C LCD and the serial monitor, and drives red, yellow and green charging-status LEDs. An instrumentation analysis compares the two voltage channels. The divider resolves about 37.5 mV per ADC step at the battery and needs calibration to meet a ±0.05 V target, whereas the INA219 resolves 4 mV with a datasheet maximum bus-voltage error of ±0.5 %, about ±21 mV at 4.2 V. The hardware and software are linked through the recorded measurements: voltage, current and power readings logged from the prototype were used as the basis for generating a larger battery-ageing dataset, on which machine-learning models were trained to predict the state R2 of 0.9906 and an RMSE of 0.631 %, and predicts that 80 % end of life is reached at 929 cycles. An XGBoost regressor using depth of discharge, charge rate, temperature and cycle count achieves a test RMSE of 2.107 % and an MAE of 1.485 % on unseen cells. The paper also documents the error budget, a calibration procedure against a digital multimeter, and the limitations of the prototype.