Statistical Processing of Car Engine Signals for Combustion Diagnosis

We report on analysis of car engine signals as cylinder pressure and vibration signals for combustion diagne sis. Combustions have to be observed for controling efficiency and pollution as well as protecting against knock and can be affected, e.g. by controling the angle of ignition. We first model pressure signals by nonstationary stochastic processes characterized by the compression cycle and a stochastic resonance model. Vibration signals are modeled as time-variant filtered versions of pressure signals superimposed by noise. Wigner-Ville time-frequency estimates applied to measured data that average over many combustion cycles provide evidence of the models. Because only vibration signals can be easily measured in cars, we show that pressure signals can be reconstructed by time-variant filtering of vibration signals. Enhanced knock detectors are discussed which test resonance powers estimated from vibration signals via non-equidistant sampling. Finally, we report on a signal-processor based test-bed diagnosis-system for real-time operation.