Scott, Ian JamesGoodman-Rendall, Kevin2024-02-222024-02-222024-01-30http://hdl.handle.net/10362/163935Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics, specialization in Data ScienceLithium-ion battery (LIB) cycling data was collected from two publicly available online sources for use in a Gradient Boosting Machine, (LightGBM), and Long-Short-Term-Memory Recurrent Neural Network, (LSTM-RNN) towards predicting capacity decay and remaining useful life (RUL) in a transfer learning format. The two models were compared for their ability to predict state of health (SOH), knee-point (accelerating capacity loss), and second-life regions (< 80% nominal capacity), to inform potential applications of second-life lithium-ion batteries (SLBs). LightGBM proves accurate to an MAE of 3.4% SOH during cross-fold validation while the LSTM-RNN returned an MAE of 4.2% in SOH prediction. Knee-points and second-life regions are more accurately predicted by the LSTM-RNN but a stacked approach of both models is hypothesized as superior. Both models’ transfer learning is limited by cycling regimes and fails to generalize to alternatively cycled LIBs highlighting a need for more empirical LIB cycling data in future research.engLithium-ion batteryLight gradient boostingSecond-lifeKnee-pointLong-short term memory recurrent neural networkTransfer learningSDG 7 - Affordable and clean energySDG 13 - Climate actionmaster thesis203526333