Multi-modal Diagnostics for Vehicle Fault Detection

On-board vehicle diagnostic systems must have low development and hardware costs in order to be viable. Modelbased methods have shown promise since they use analytical redundancy to reduce costly physical redundancy. However, these methods must also be computationally efficient and function accurately even with simple, low-cost models. The approach presented in this paper uses multiple simple models to analyze dissimilar observable modes of a system. Residuals generated using the models are related and interpreted in a Bayesian network to determine fault probabilities and yield a diagnosis. The technique is demonstrated with a diagnostic system for automobile handling.