Misfire-misfuel classification using support vector machines

Abstract This paper proposes the use of support vector machines to perform classification between different types of missed combustion event in a six-cylinder engine. On-board diagnostics regulations require the detection of missed combustion events, which is possible through interpretation of crankshaft speed information. However, current approaches provide no information on the actual cause of the event, in particular whether it was caused by a misfuel (absence of fuel) or a misfire (absence of spark) event. Whilst the impact on the environment and emission treatment systems due to misfuel is minimal, misfire events are detrimental to both. Consequently information regarding the causes of missing combustion events potentially allows the development of unique recovery strategies particular to the source of the problem. In this paper, an approach is proposed that will provide the potential for, firstly, detection of a missing combustion event and, secondly, real-time classification of the event into either misfuel or misfire events using feedback from a heated universal exhaust gas oxygen sensor. In order to evaluate the potential of such a system in an engine control unit, a computational complexity measure is also presented.