Application of Derivative-Free Search Algorithms for Performance Optimization of Spark Ignition Engines

This paper exploits the possibilities of achieving an efficient performance optimization methodology to be applied to different spark ignition engine configurations. The objective of the task described here is to determine the combination of parameters which provides the highest volumetric efficiency and effective torque. The definition of general strategy requires first the identification and grouping of the geometric and operating variables to be optimized (duct diameters and lengths, valve timing, spark advance, etc…). The high number of possibilities entails critical choices to reduce, from an engineering design point of view before than from a mathematical point of view, the required computational time. Once proper thermo-fluid dynamic decisions are taken, the most efficient optimization methodology is required.

The application of Design of Experiments techniques allows to screen the design space and give a first estimation of the optimal point. Then, a finer optimization strategy needs to be employed. In this paper, the Mesh-Adaptive Direct Search (MADS) method and the Genetic Algorithms are presented and applied. A critical discussion on the opportunity of employing any of those methods rather than relying only the preliminary DoE indications is proposed. Different ideal and real engine configurations are studied, from simple single cylinder to high performance complex 12-cylinder engines.