Misfire detection in a diesel engine using clustering in a short-time analysis of vibroacoustic signals

Abstract

The paper presents some results of the research on new diagnostic methods in combustion engines. It describes the application of short-time signal analysis together with pattern recognition techniques in the diagnosis of misfire in Diesel engines through vibroacoustic signals. One considered Diesel locomotive in particular. In the area of the nonroad sources of combustion gases the locomotives rate relatively high as air polluters. There are some regulations in the area of locomotives (e.g. Cart UIC 623 1-2-3 in Europe) but we still observe a lack of obligatory requirements for systems monitoring emission critical damage. Such obligatory on-board diagnostic systems were introduced for passenger cars (OBD II, EOBD). The OBD system performs a continuous monitoring of basic system parameters and one of its most important tasks is misfire detection. All these facts inclined the author to research the new relevant detection methods. The main aim of the research is to distinguish between two states: normal engine operation and the state of misfire. The general idea of the method was taken from the short-time Fourier analysis. The method is based on calculation of the values of some selected parameters in the time window sliding along the signal. For each window position one has a set of parameter values which gives the point in a corresponding multidimensional parameter space. Hence, the time evolution of the signal can be observed as the evolution plot in the parameter space. We suspect that the different system states (misfire) can be distinguished by the different position of points in the parameter space. In order to detect them, the clustering in the parameter space was performed. The first results show the possibility of distinguishing some different clusters within the parameter space which may correspond to different engine states.