Dynamic measurement of HCCI combustion with self-learning of experimental space limitations

Homogeneous Charge Compression Ignition offers a great potential for increasing the efficiency of combustion engines while simultaneously reducing nitrogen oxide raw emissions. However, the broad application has not yet been realized for production engines, mainly owing to low combustion stability at the edges of the operating range and high sensitivity to changing boundary conditions. Owing to strong cycle-to-cycle coupling by negative valve overlap, the cylinder state of the last combustion has an enormous influence on the subsequent combustion. Therefore, the condition of the previous combustion must be taken into account to control the following combustion event. For this purpose, the interactions between feedback variables and cycle individual control interventions need to be measured in a wide operating range. Against this backdrop, a new measurement methodology is presented in this article, which sets up the transient limitations for Homogeneous Charge Compression Ignition combustion, while maintaining the limits of stability, maximum pressure gradient, and other factors automatically. Hence, an algorithm has been developed that sets the manipulated variables on a cyclic basis, dependent on the previous cycle in several dimensions. The new algorithm was then used to gain dynamic measurement data that were used to train artificial neural networks. It is demonstrated that the models are able to predict misfires under certain conditions. Additionally, a feasibility study regarding the usability of the newly gained models was performed based on a data-driven control algorithm, which was carried out and validated on a single-cylinder test engine.