Model-based adaptive air/fuel ratio control for an automotive gasoline engine

Abstract

This thesis presents the design of an adaptive air/fuel ratio control system incorporating a λ sensor with varying dynamical properties. As a baseline, the design of a nominal non-adaptive air/fuel ratio controller based on the internal model control principle is presented. For the adaptation of the λ sensor dynamics two different adaptive methods tailormade for the internal model control structure have been developed. The first method suggests a sequential tuning of the parameters of the internal model and the internal controller, respectively. For both parameter groups the parameter values are found by minimizing a predefined cost function. The optimization is run with a gradient-based minimization procedure where, analogously to the well-known iterative feedback tuning scheme, the gradients are computed from signals obtained from closed-loop experiments. Thus for the calculation of the gradient, the unknown plant is utilized, whereas other “local” tuning methods suggest to replace the real plant by its model to calculate the gradient. The suggested algorithm inherently operates under closed loop conditions and allows for the different roles of the disturbances that occur during the two tuning steps. As a consequence of this, the adapted internal model can be shown to be unbiased whereas for the tuning of the internal controller the disturbances are rightly taken into account. The second concept is based on direct adaptation of the internal model. To that end, the internal model is formulated in an output-error identification model structure that can be adapted by a linear regression on filtered versions of the input and output signals of the internal model. The parameter vector of the regression is adapted by using a standard adaptive law. The output-error approach has advantageous properties concerning the parameter convergence and the control performance when disturbances are present. To guarantee the global stability of the adaptive systems, the denominator polynomial of the transfer function of the unknown plant needs to be known, however, it is shown in this thesis that only a rough initial internal model in the control-relevant frequency range is necessary for stability. Both methods are described in a general framework, such that, they can