Modeling and identification of the combustion pressure process in internal combustion engines.

Abstract

This research bridges the gap between physical spark-ignited engine modeling and advanced signal processing to solve the problem of reconstructing combustion pressure from noisy observations of angular velocity at steady-state. Cyclic variability in pressure is estimated from fluctuations in velocity; they are recognized as a fundamental limitation of engine operation, especially as mixtures are leaned to optimize fuel economy and emission control. This thesis makes several contributions in order to solve this inverse problem. First, a new model for cyclic pressure variability is proposed, using a Bernoulli-Gaussian random sequence indexed by combustion number. Second, this model amplitude-modulates a template of pressure deviation from mean pressure to form a model for continuous pressure, parametrized by the sample modulating sequence. Third, this pressure model is used as the input to a linear, crankshaft-angle-varying engine model which relates combustion pressures to the square of angular velocity, defined in the crankshaft-angle-domain--i.e., crankshaft angle is the independent variable. This model is cascaded with a square root operator to solve the forward problem, computing angular velocity from pressures. It is mathematically equivalent to nonlinear models in the time-domain. Fourth, the pressure-to-velocity model is used to develop a recursion relating samples of the square of angular velocity in every combustion to the Bernoulli-Gaussian sequence. A Kalman-filter stochastic deconvolution algorithm is applied to this recursive model with noisy observations of the square of angular velocity to solve the inverse problem. The simplicity of the problem solution makes it implementable on-line. Computer simulations of algorithms for solving the forward and inverse problems were developed. These simulations show that simulated angular velocity (the forward problem) shows behavior much like real data and that the inverse problem may be solved fairly well at low and moderate noise levels. Experimental verification of these algorithms is made using data acquired from a production six-cylinder engine. The results show that the inverse problem solution is able to detect total misfires and poor combustions, indicating that it is able to classify the severity of misfires as well.