Abstract
The paper presents the results of the researches on nonlinear and short time processing
methods in Diesel engine diagnostic area with the use of vibroacoustic signals. The origin of
the researches is the expected general sharpening of norms on combustion gases emission not
only in cars but in non-road sources also. This fact should induce researches and experiments
on new methods of fault detection which can be applied in new diagnostic systems. In this
area the new signal processing methods have started to play a significant role.
The nonlinear signal processing methods used in the paper base on deterministic chaos theory.
The base of the method is a Takens-Mane theorem which allows to reconstruct the dynamics
of nonlinear system basing on the one-dimensional time series. Then the nonlinear parameters
like Lyapunov exponents can be calculated and used as diagnostic parameters.
The short time analysis used in the paper is applied together with pattern recognition
techniques. The approach consists of using a time window sliding along a signal in time and
of calculation for each window position the given parameters. This parameters create a
multidimensional parameter space where one can observe the time evolution and for detection
of different system states the clustering can be performed.
The researches were performed for misfire detection in heavy Diesel engines in locomotives.
The researches concerned the single cases of misfire and the misfire was also simulated by
disconnection of one cylinder of the engine. The results of the nonlinear analysis demonstrate
that the dominant Lyapunov exponents for the case when the all cylinders are working are
less than for the case of state of a misfire. The results of short-time analysis are showing the
possibility of distinguishing in the parameter space different clusters which can correspond to
different engine states.
Keywords - Diesel engine, misfire, signal processing, short time methods, nonlinear
analysis