Development of mean value engine model using ANNProblem DefinitionHow to make itTraining Neural NetworkResults & ConclusionDevelopment of mean value engine model using ANNSoo-Youl Park (Mechanical Engineering)Problem DefinitionEngine simulation model Physical model (Gas dynamics, Combustion, thermodynamics)Generally, More complex more accurate and more time consuming in calculationNeed for fast running model guaranteeing high accuracy hopefully, running at real timeMinimizing the use of physical model by incorporating ANN Mean value engine modelHow to make itBuild a fully physical based model Using DOE (Design of Experiment) with above model, generate training and test data. Replace some parts of physical model with ANN modelTraining Neural Network Self Organizing Local Linear (Unsupervised Learning) njjmijiijCompcuay1 1)( uWjRMS training errorRMS testing errorWith Core2 CPU of 2.4GHz, calculation time for acceleration cycle from zero to 80 km/h is140s in case of fully physical model 21s in case of ANN supporting model close to real time (real time is about 7.36sAccuracyResults & ConclusionANN supporting vs. fully physical model Results for acceleration
View Full Document