Due to more stringent emission standards as well as customer requirements on performance improvement, model-based controls in diesel engines are becoming more and more common and necessary. In fact, as diesel engines becoming increasingly complicated with additional hardware components such as electonic throttle, EGR, VVT, VGT, as well as aftertreatment devices, the dynamics of the systems with more freedom of multiple actuators become much more sophisticated. With such complexity in the diesel engine systems, the traditional simple PI control, single-input single-output type of controls will not be good enough to address the multivariable interactions among subsystems, instead the advanced model-based, multi-input multi-output and coordinated supervisory controls almost become the only effective ways to improve system performance and achieve emission standards.
In most of the model based approaches, feedforward plus feedback controls or together with integrated estimation methods based on real-time dynamics are the key techniques. Such kind of advanced controls are effective for transient conditions, and also robust under varying measurements and un-modeled dynamics. However if the sensors and actuators have large part-to-part variations, or if the the system or components are significantly degraded over the useful life, then any identification of those slow changes and compensation of the controls over the system life cycle will be critical for improving long term control performance, system reliability and diagnostic monitoring capability. In this paper, we proposed a new architecture of using both observer bank and event based estimation strategy for control design, and applied them into the air-path controls and oxygen concentration estimation. The approach utilizes both short term real time estimation and long term statistics adaptation/ optimization to achieve the improved accuracy of certain state estimation for transient conditions and static compensation over useful life. Effectiveness is illustrated through the simulated results with mean-value engine model.