Composite adaptive and input observer‐based approaches to the cylinder flow estimation in spark ignition automotive engines

The performance of air charge estimation algorithms in spark ignition automotive engines can be enhanced using advanced estimation techniques available in the controls literature. This paper illustrates two approaches of this kind that can improve the cylinder flow estimation for gasoline engines without external exhaust gas recirculation (EGR). The first approach is based on an input observer, while the second approach relies on an adaptive estimator. Assuming that the cylinder flow is nominally estimated via a speed‐density calculation, and that the uncertainty is additive to the volumetric efficiency, the straightforward application of an input observer provides an easy to implement algorithm that corrects the nominal air flow estimate. The experimental results that we report in the paper point to a sufficiently good transient behaviour of the estimator. The signal quality may deteriorate, however, for extremely fast transients. This motivates the development of an adaptive estimator that relies mostly on the feedforward speed‐density calculation during transients, while during engine operation close to steady‐state conditions, it relies mostly on the adaptation. In our derivation of the adaptive estimator, the uncertainty is modelled as an unknown parameter multiplying the intake manifold temperature. We use the tracking error between the measured and modelled intake manifold pressure together with an appropriately defined prediction error estimate to develop an adaptation algorithm with improved identifiability and convergence rate. A robustness enhancement, via a σ‐modification with the σ‐factor depending on the prediction error estimate, ensures that in transients the parameter estimate converges to a pre‐determined a priori value. In close to steady‐state conditions, the σ‐modification is rendered inactive and the evolution of the parameter estimate is determined by both tracking error and prediction error estimate. Further enhancements are made by incorporating a functional dependence of the a priori value on the engine operating conditions such as the intake manifold pressure. The coefficients of this function can be learned during engine operation from the values to which the parameter estimate converges in close to steady‐state conditions. This feedforward learning functionality improves transient estimation accuracy and reduces the convergence time of the parameter estimate. Copyright © 2004 John Wiley & Sons, Ltd.