Abstract:

This paper presents a data-driven model for accurate, early, and economical fault detection and isolation in vehicle battery systems. The model is based on kernel principal component analysis (KPCA), which maps complex nonlinear data into a high-dimensional feature space. To overcome hyperparameter selection difficulties, KPCA is trained using Bayesian Optimization (BO) iterations with a small amount of labeled data and a large amount of unlabeled data. This approach improves the model's fault detection capability, detecting both early and minor faults. A unified contribution graph based on partial differentiation is adopted for a reasonable isolation scheme. A semi-supervised model of KPCA is developed to reveal the relationship between fault and variable. The method is tested on four fault datasets, proving excellent detection capability in the early stage of faults and accurate fault isolation from occurrence to end.