In modern engine control applications, there is a distinct trend towards model-based control schemes. There are various reasons for this trend: Physical models allow deeper insights compared to heuristic functions, controllers can be designed faster and more accurately, and the possibility of obtaining an automated application scheme for the final engine to be controlled is a significant advantage. Another reason is that if physical effects can be separated, higher order models can be applied for different subsystems. This is in contrast to heuristic functions where the determination of the various maps poses large problems and is thus only feasible for low order models.
One of the most important parts of an engine management system is the air-to-fuel control. The catalytic converter requires the mean air-to-fuel ratio to be very accurate in order to reach its optimal conversion rate. Disturbances from the active carbon filter and other additional devices have to be compensated. Emission-optimal warm-up strategies require the engine to be run slightly lean without misfires. All these tasks cannot be fulfilled with the feedback controller only but also need a very good feedforward strategy. Such a feedforward strategy usually inverts the known plant dynamics. One important part of the fuel dynamics is the wall-wetting dynamics, i.e., the fact that not all the fuel injected enters the cylinder during the following intake valve opening. First-order approximations of these dynamics are frequently used. Their advantage is that they are easy to implement in an inverse dynamic feedforward strategy and that the parameters can be determined quite simply. However, the wall-wetting dynamics are not sufficiently described by first-order approximations.
New engine management systems increasingly are fitted with wide-range sensors. Due to a smaller time constant, these sensors allow the detection of higher-frequency phenomena and also yield quantitative information on the air-to-fuel ratio (lambda). Mounted closely to the exhaust valve, these sensors also allow a better identification of the wall-wetting dynamics.
A second-order model shows very good performance. It can be fitted into the frequency response measurements with a nonlinear optimization routine. The scheme for the parameter identification presented here does two things: It highlights certain problem areas which require special attention and it describes the parameter maps for the operating region at two different cylinders of an SI engine. Finally, some measurements taken on the cold engine are shown and interpreted.