Cyclic variability of the combustion process in internal combustion gasoline engines has long been known to produce an undesirable increase in emission levels, decrease in engine efficiency, and reduced driveability. These variations are closely reflected in the cyclic variation of cylinder pressure and/or ionization measurements. However, the majority of the information from these sensors has traditionally been discarded, either by considering only the ensemble mean or by analyzing only a few scalar quantities such as peak pressure, peak pressure rise, or mean effective pressure. This loss of information, and the lack of a more complete description of the combustion process, significantly limits the generality and capability of monitoring and control of the combustion process at the individual cylinder event level.

In this work, a control-relevant stochastic combustion model is developed to describe the cyclically varying time history of the combustion process during the combustion period within the cylinder. The resulting set of model parameters, which can be efficiently estimated from sensor data, provide a much more complete description of the cyclic combustion process and are capable of characterizing not just point values, but variations in the entire evolution of the combustion period.

This work describes how the generality of the stochastic combustion model can be exploited. Examples include simulation of cyclic data sets with similar statistical properties to actual engine data and derivation of the traditional monitoring statistics, such as standard deviation of maximum pressure and start of burn angle, directly from the parameters of the model. The same approach also provides insight into the phasing of the cyclic process because a number of these monitoring statistics can be computed directly as a function of crank angle. It is also straightforward to evaluate the contribution from each of the physically meaningful model parameters to these quantities and, thereby, obtain some physical understanding of the mechanism involved.

The presentation begins with the framework of the stochastic combustion model and the variety of possible parameterizations. These parametric forms are fitted to engine data and the model validated by comparing the model simulations to verification data sets. The model is then used to estimate the stochastic properties of undesirable engine events such as knock through an evaluation of residuals. Finally, a stochastic control algorithm is proposed to maximize combustion efficiency.