Model Networks as Virtual Combustion Sensors in IC Engines

Abstract

In-cylinder pressure analysis provides valuable information on the combustion process and stands as an invaluable tool in the internal combustion engine research & development. Its implementation in a combustion control algorithms appears as a promising solution for attaining optimal combustion control in IC Engines. Beside direct, pressure sensor based measurement data evaluation approach, there are various attempts to introduce indirect model based measurements which will provide comparable results. This paper deals with the potentials of Local Model Networks and their application in a virtual combustion sensors. High speed processing capabilities, acceptable accuracy and compactness of Local Model Networks based models make them good candidates to become a core component of the future combustion control algorithms. A local linear model (Lolimot) and hierarchical local model (Hilomot) are compared in order to gain some conclusions on optimal network topology, best suited for a job of extracting crucial combustion features. Evaluation, testing and validation of presented virtual sensors is conducted on an extensive set of IC engine data.