Reduced and separated meta-models of the scavenging behavior of a 2-stroke diesel engine for evolutionary exploration of solution spaces.

Summary The use of numerical techniques during the design of a product has become widespread over the last 30 years. However, the slowness of the calculations and the specialization of the numerical models remain problematic. We have therefore chosen to develop reduced models of the scavenging behavior of a 2-stroke diesel engine with lights. These models are analytical, generic, fast to use and allow to eliminate the problems of numerical processing. They are also powerful tools in the search for design solutions. A 2D CFD model was first developed to serve as a database, with the definition of the most important parameters to follow in order to quantify an optimal sweep.The research work unveiled a novel methodology based on a meta-model of the so-called "neuro-separated" behavior including a neural state model, a pseudo-dynamic neural model and a separate variable model. Then, a decision support process exploiting the previous models was implemented through an evolutionary optimization process (based on genetic algorithms) and the rapid behavioral simulation of optimal design solutions by kriging.The multi-point of view, multi-criteria and multi-physics design approach applied to the engine also integrates a cognitive dimension: the evolutionary exploration of the solution spaces was conducted in a free and forced way. In order to validate our approach, we have set up qualification criteria applied to each of our models, allowing to quantify the deviations from the initial CFD base which founded our reduced models.our approach has led to the creation of a modeling and decision support tool using the Python and Matlab modules developed.