Engine fault detection using a nonlinear FIR model and a locally regularised recursive algorithm

Strict legislation worldwide on emissions has forced automotive manufacturers to adopt additional control and management techniques. On-board diagnostic (OBD) technology provides an effective way to monitor the engine conditions. However, this method highly relies on an accurate physical model of the process being monitored. This paper utilizes the recently developed locally regularised fast recursive algorithm (LRFR) to build a nonlinear finite impulse response (NFIR) model for an engine intake subsystem. The main advantage of this approach is the simplicity of construction and implementation. The LRFR combines the forward recursive approach and regularisation method to produce a compact parsimonious NFIR model with good generalization performance. The method is applied to a 1.8 litre Nissan gasoline engine to detect air leak fault in the intake manifold, and the test results confirm the efficacy of the proposed approach.