Diesel engine combustion and mechanical noise separation using an improved spectrofilter

The present paper is dedicated to diesel engine combustion and mechanical noise separation. It is shown that this separation can be performed by a spectrofilter if and only if the spectrofilter is computed using only the random part of the engine signals. This source separation is a difficult task, since the sources are correlated and overlapped both in time and frequency domains. Considering only the random part of the signals virtually uncorrelates the noise sources, making them easier to separate. This approach is validated by the results of a simulation. An experiment based on real engine signals shows that the spectrofilter is more causal and more robust if computed using only the signals random part. This experiment also shows that speed rotation and load condition affect the engine transfer functions.