CBR and Machine Learning for combustion system design

Abstract

Nowadays the automotive industry has to face two major challenges. First products must meet continually increasing government requirements on fuel economy and low exhaust emission. Second the market demands product variety and short production cycles. The automobile's combustion system determines the exhaust emission rate, combustion system engineering is one of the crucial steps in the development process. Cylinder head design is a good example of showing how enhanced AI technologies like CBR and Machine Learning support high-level engineering design tasks.

The work described was coordinated in a joint project between the Daimler-Benz research group on Thermo and Fluid Dynamics and our reasearch group on Machine Learning with the aim of improving of cylinder head engineering. This paper proposes how Machine Learning and specifically Case-based Reasoning (CBR) transform a traditional database containing both geometry and air-motion data into a so called experience memory for cylinder head design. We will present the initial steps of our database analysis in terms of different learning algorithms, then use the extracted knowledge to develop case-based design retrieval and quality prediction modules.