Adaptive ignition control using on-line learning of delaunay networks

Due to the extremely nonlinear dynamics of internal combustion engines, standard ignition control systems rely heavily upon calibration and look-up tables. In order to increase engine performance and to minimize fuel consumption and exhaust emissions, new control strategies for the ignition timing are required. Based on cylinder pressure sensors, our approach combines non-linear feed-forward with linear feedback controllers for the individual cylinders. The output of each linear controller is used to adapt the respective feed-forward controller. Because of their computational efficiency and learning capabilities Delaunay networks [12] are employed to represent the feed-forward controllers. Experimental results obtained in a research vehicle show that the proposed control architecture is very effective in learning the engine´s nonlinearities and in compensating for manufacturing tolerances and aging, thus improving efficiency and fuel consumption.