Park_pro

Embed Size (px)

Citation preview

  • 8/6/2019 Park_pro

    1/1

    Name : Soo-Youl Park

    Title : Development of mean value engine model using ANN

    Objective : To develop a fast-running engine model with sufficient accuracy over wide

    range of operating conditions

    Motivation :

    In case of automotive engine simulation model, there are two requirements. First, ithas high accuracy, second, it should run fast. Especially demands for fast running model

    become increasing. Modern engines have sophisticated control device so that the enginemodel and control model should be connected to each other and both engine simulation

    and control simulation should be performed simultaneously. Furthermore, for theapplication of HIL(Hardware in the loop), the engine model should run at real time.(In

    case of 2000 rpm of engine operating, actual time for one engine cycle is 0.06s, butcalculation time for one cycle with typical engine model is 120s, which is 2000 times

    more than real time.)The engine simulation is a kind of physical simulation which calculates the gas

    dynamics, combustion phenomena and chemical reaction within the engine. Therefore, inorder to guarantee high accuracy results, the model should be complicated. It means that

    these two requirements, higher accuracy and fast running, conflict with each other.However, recent researches show that this conflicting problem can be solved by

    applying ANN to the engine model and this engine model supported by ANN is called asmean value engine model. In this study, I will show the methodology how to build this

    mean value engine model by minimizing sacrifice of accuracy of results.

    Approach :With commercial engine simulation software, detailed engine model will be

    developed. Based on this model, the model regression will be done by applying ANN.This work will be composed of following steps

    (1) Identify important variables which can represent the results;(2) Conduct simulations using a detailed engine model, and train neural networks

    with simulation results;(3) Build a mean value engine model using trained neural networks;

    Reference :

    He, Y., Development and Validation of a Mean Value Engine Model for IntegratedEngine and Control System, SAE Paper 2007-01-1304, 2007.

    Papadimitriou, I., Warner, M., Silvestri, J., Lennblad, J., and Tabar, S., Neural Network

    Based Fast Running Engine Models for Control-Oriented Applications, SAE Paper2005-01-0072.

    The Mathworks Inc., Model-based Calibration Toolbox, Version 3, 2006.

    PDF created with pdfFactory Pro trial version www.pdffactory.com

    http://www.pdffactory.com/http://www.pdffactory.com/