Case Study on Predictive Modeling of UAV Battery Remaining Useful Life Using Machine Learning Regression Techniques

Predicting Remaining Useful Life (RUL) for Lithium-cell NMC-LCO batteries in Unmanned Aerial Vehicles (UAVs) is crucial for safe and efficient operation. Traditional methods rely on offline data analysis, limiting their accuracy in dynamic UAV flight conditions. This work proposes a Machine Learning (ML)-based approach for real-time RUL prediction, integrating online data acquisition. Remaining Useful Life (RUL) prediction is a numerical prediction, so the prediction of RUL fits into the regression technique. All regression models of ML are used in predicting the RUL. Sensor data from voltage, current, temperature, and vibration are fed into regression models of ML which include the KNN algorithm, SVM algorithm, Artificial Neural Networks (ANN), Decision Tree, Random Forest, Gradient Boosting, and RNN networks with LSTM units, capturing temporal dependencies in battery degradation. The model is continuously updated using online data, improving its accuracy. The best fit for RUL prediction is analyzed to be a Random Forest. This approach addresses the challenges of dynamic UAV flight and diverse operating conditions, leading to more reliable and accurate RUL predictions for enhanced UAV safety and performance.