Misfire detection in spark-ignition engines with the EM algorithm

The government regulations requires the on-board monitoring of misfire occurrences in automotive engines, mainly due to its negative effects on the exhaust emission levels. The paper presents a method of misfire diagnosis based on the analysis of the structure-borne sound. The EM algorithm combined with a parametric pressure model is used to decompose the sound signal in individual components associated with single combustions. This fact leads to a simple statistical test of detecting the misfires by evaluating the shape of the reconstructed pressure. Experimental results with real data measured at the different engine operating points have shown that the method can compete with the already existing approaches.