A Study on Effects of Synthetic Data for Predicting the Remaining Useful Life of Aluminium Electrolytic Capacitors Using Bagging-Based Ensemble Learning

Aluminium electrolytic capacitors (AECs) are crucial components used in power supplies, computer motherboards, and variable frequency drives for ripple filtering and energy storage. Their performance is affected by factors like temperature, frequency, ripple current, and ageing. Despite design advancements, AECs remain reliability weak points, demanding thorough analysis. This research proposes a method to predict the Remaining Useful Life (RUL) of AECs under combined temperature and voltage stress. We develop an RUL prediction model using bagged decision trees by ageing 24 AEC samples and tracking degradation via capacitance and ESR changes. Synthetic data strengthens the model due to limited training data. We assess the impact of different minority oversampling methods like SMOTE, SMOTER, and SMOGN on prediction accuracy, using ANOVA to compare outcomes within and between algorithms.