Closed-loop regeneration of spent nickel manganese cobalt oxide cathodes enabled by artificial neural network modelling

Authors: Alyoubi, M., Cetinkaya, D., Zoromba, M.S., Abdel-Aziz, M.H., Bassyouni, M., Mahmoud, E.R.I., Ali, I. and Abdelkader, A.M.

Journal: Journal of Energy Storage

Volume: 149

eISSN: 2352-152X

DOI: 10.1016/j.est.2025.119930

Abstract:

This study presents a machine learning-driven approach to optimise the two-stage direct regeneration of spent LiNi₁/₃Mn₁/₃Co₁/₃O₂ (NMC111) cathode materials. Multiple models were developed within the Automated Network Search (ANS) environment in TIBCO Statistica, with the Statistica Automated Neural Network (SANN) achieving the highest predictive accuracy (R²= 0.994, MAE = 0.491). Unlike traditional stepwise optimisation, which sequentially determine optimal hydrothermal and solid-state conditions, SANN was trained on a combined dataset to simultaneously evaluate both stages, capturing their interdependent effects. The model predicted that a hydrothermal treatment at 190 °C for 4 h followed by solid-state annealing at 730 °C for 4 h would yield optimal results, achieving a first-cycle capacity of 150.2 mAh g⁻¹, a 50th-cycle retention of 98.3 %, and extending the remaining useful life from 324 to 458 cycles. These outcomes approach the upper benchmark for commercial NMC batteries (94–98 % retention at 50 cycles) while reducing thermal energy consumption by 120kJ kg⁻¹, equivalent to ∼$43 per metric ton processed. The model maintained optimisation accuracy despite residual impurities, underscoring its industrial applicability. Combined with previous successes in optimising single-method regeneration for LCO and LFP, these findings demonstrate SANN's versatility and potential to enable scalable, energy-efficient, and sustainable closed-loop recycling of lithium-ion battery (LIB) cathodes.

Source: Scopus