A multi-objective optimization model for predictive opportunistic maintenance of lithium-ion batteries

This paper proposes a predictive opportunistic maintenance optimization model for lithium-ion batteries based on remaining useful life (RUL) predictions. First, we introduce a RUL prediction method based on Multi-Head Attention-Temporal Convolutional Networks-Evidential Regression (MA-TCN-ER). The MA-TCN framework captures the global dependencies in multi-dimensional degradation time series of batteries, and evidential regression quantifies the epistemic uncertainty in predictions. Next, the RUL probability density function (PDF) is obtained using kernel density estimation, providing prior knowledge for maintenance decision-making. For individual batteries, we develop a multi-objective maintenance model that considers maintenance cost, availability and reliability measures to determine the optimal preventive replacement time. Based on the optimal replacement time for each component, we calculate the opportunistic maintenance time windows to determine the grouping structure for opportunistic maintenance. The effectiveness of the proposed method is validated using the Oxford lithium-ion battery degradation dataset and the NASA PCoE battery dataset. Experimental results demonstrate that our RUL prediction method provides more accurate RUL estimates than other existing approaches.