Identification of Diesel Engine Events from Acoustic Signals using Independent Component Analysis and Time-Frequency Analysis 2007-01-2278

In this paper the events of a single cylinder diesel engine are identified from acoustic signals using independent component analysis (ICA) and time-frequency analysis (TFA). The identified engine noise sources and their events can be used for faults diagnosis such as valve clearance check and fuel injection advance detection. Acoustic signal of diesel engine is acquired in time domain with a 12 channels microphone array as blind sources. The mixed signals are filtered to sub-bands by a bandpass filter. The sources of each subband of the mixed signals are separated by independent component analysis. Each separated component is analyzed in time-frequency domain by STFT in order to obtain the exact time for each frequency occurrence. The time is converted to crank angle in the time-frequency domain in order to investigate the engine events. From crank angle-frequency analysis the exhaust and inlet valves open and close, fuel injection, combustion and piston slap are identified. The mixed signals and the estimated independent components (ICs) are converted from dynamic data type (DDT) to listenable audio sounds. As predicted, the results show that the engine events which are identified from acoustic signal are well correlated with the engine design specifications.