Novel Image-Based Rapid RUL Prediction for Li-Ion Batteries Using a Capsule Network and Transfer Learning
Jonathan Couture, Xianke Lin
Abstract
The recent popularity boost in electric vehicles created a large demand for lithium-ion batteries, but current recycling methods are not ready to carry the weight of the years to come. Because of this, the main goal of this study is to optimize the speed and accuracy of lithium-ion batteries’ remaining useful life (RUL) prediction methods to make them a viable option for second-life classification applications. This article develops a capsule network architecture for rapid battery RUL prediction using transfer learning techniques. The proposed method can accurately predict the RUL of a cell using a single charging and discharging cycle, making it one of the fastest methods available to date. This novel image-based health prognostic estimation method reduces the preprocessing labor and, consequently, the amount of human-induced bias in the dataset. Not only are complete charging and discharging cycles shown in a single image, but also even numerical data are added and taught to be recognized by the network. This rapid prediction model will have uses in the fast characterization of battery cells for second-life classification purposes, for researchers developing health-conscious charging protocols, and even for battery management system implementations.