Enhancing Fault Detection in Hybrid Electric Vehicles using Kernel Orthonormal Subspace Analysis

Abstract

Hybrid electric vehicle (HEV) performance and safety are critical areas of focus in modern automobile technology, with fault detection being a key challenge. Traditional methods often fall short when it comes to detecting complex faults in the HEV powertrain system, as these faults exhibit nonlinear behaviors. This paper introduces a novel data-driven approach for fault detection in HEV powertrains using Kernel Orthogonal Subspace Analysis (KOSA). The KOSA method addresses the limitations of linear Orthogonal Subspace Analysis (OSA) by mapping nonlinear problems to a higher-dimensional space through a kernel function, thereby enabling more effective fault separation. This transformation, combined with the dimensionality reduction capabilities of OSA, allows KOSA to detect complex faults in the powertrain system more efficiently. Experimental results from both a nonlinear model and real-world data from the HEV demonstrate that KOSA outperforms both OSA and Kernel Principal Component Analysis (KPCA) in terms of fault detection accuracy and robustness.