In Connolly and Yagle (1992, 1993) we presented a new model relating cylinder combustion pressure to crankshaft angular velocity in an internal combustion engine, primarily the fluctuations in velocity near the cylinder firing frequency. There are three aspects to this model. First, by changing the independent variable from time to crankshaft angle, a nonlinear differential equation becomes a linear first-order differential equation. Second, a new stochastic model for combustion pressure uses the sum of a deterministic waveform and a raised-cosine window amplitude-modulated by a Bernoulli-Gaussian random sequence, parametrizing the pressure by the sample modulating sequence. This results in a state equation for the square of angular velocity sampled every combustion, with the modulating sequence as input. Third, the inverse problem of reconstructing pressure from noisy angular velocity measurements was formulated as a state-space deconvolution problem, and solved using a Kalman-filter-based deconvolution algorithm. Simulation results in Connolly and Yagle (1992, 1993) show that the parametrized pressure can be deconvolved at low to moderate noise levels, and combustion misfires detected, all in real time. This paper presents and discusses experimental results that confirm this model, at least at the relatively low-speed, low-to-moderate load operating conditions analyzed. They show that cyclic combustion pressure variation is fairly well modeled and may be directly estimated from angular velocity measurements. They also show that the deconvolution algorithm is able to detect misfires and possibly classify their severity. Since the experimental data are taken from an actual V-6 automobile engine, and the algorithms are simple enough to be implemented in real time, these results are directly applicable to real-world combustion pressure identification.
Read full abstract