Neuro-separated meta-model of the scavenging process in 2-Stroke Diesel engine

The complexity of flow inside cylinder leads to develop new accurate and specific models. Influencing the 2-stroke engine efficiency, the scavenging process is particularly dependent to the cylinder design. To improve the engine performances, the enhancement of the chamber geometry is necessary. The development of a new neuro-separated meta-model is required to represent the scavenging process depending on the cylinder configuration. Two general approaches were used to establish the meta-model: neural networks and NTF (Non-negative Tensor Factorization) separation of variables. To fully describe the scavenging process, the meta-model is composed by four static neural models (representing the Heywood parameters), two dynamic neural models (representing the evolution of gases composition through the ports) and one separated model (the mapping of the flow path during the process). With low reduction errors, these two methods ensure the accuracy and the relevance of the meta-model results. The establishment of this new meta-model is presented step by step in this article.