Study On Control Application To Automotive Engine

Electronic fuel injection (EFI) systems and three-way catalytic (TWC) converters are widely appled into automotive engine control technoligies (ECT). The three-way catalyst can convert some gases in exhaust such as CO, HC, NOX and so on to CO2, H2O and N2, but it works well when the air-fuel ratio (A/F) is near the stoichimetric value. So in the electronic control system, whether the air-fuel ratio of exhaust is controlled around stoichiometric value is crucial for the engine control system. Too high or too low air-fuel ratio may lead to misfire. Engine misfiring will pollute environment, waste energy and even lead to accident. This dissertation aims to study on individual cylinders A/F control and engine misfire fault detection.Different from traditional air-fuel control, individual cylinders A/F control can control the A/F of exhaust passing around TWC converter much more effectively. For a V6 engine, a model of individual cylinders A/F is constructed and identified by experiment using the recursive least-squares (RLS), and then PI controller within estimated individual A/Fs as inputs is applied to control fuel injection. Experimental results indicate the control system works well. Considering sensor perturbations and actuator perturbations disturbing thecontrol system, the robustness of it to oxygen sensor perturbations and injector gain perturbations are discussed. By comparing the individual cylinders A/Fs under different time constant or different gains, the phenomenon is found that the control system has strong robustness to injector gain perturbations, but weak robustness to the sensor perturbations especially when the time constant up to 0.1s.Experimental results has shown that exhaust components are related with misfire and a 3-layer BP neural network model is constructed for exhaust components and misfire reasons. In order to optimize the model, fuzzy theories and genetic algorithms (GAs) are used to advance the neural networks. Fuzzy theories are used to normalize inputs and judge outputs of neural networks, and genetic algorithms can search globally for global minimum. At last, the proposed misfire detection method is proved to be effective.