University of Tasmania

File(s) under permanent embargo

SI engine modeling using artificial neural networks as virtual sensors

posted on 2023-05-26, 19:55 authored by Fu, Si Hua
Spark ignition (SI) engines have been developed for more than one century. In today's modern vehicles, the SI engine has been demonstrated to be efficient and reliable. However, automotive engineers are still seeking to improve and optimize SI engine performance. The key to optimize the SI performance is the setup of the engine control unit (ECU). The setup includes various engine parameters, such as engine speed, brake torque, and air intake mass flow etc. Problems can arise in how to acquire the data of desired engine parameters for constructing the SI engine control system. Engine sensors and some special equipment are employed to acquire the desired data. The equipment includes the engine dynamometer and gas analyzer that are much more expensive than engine sensors. In addition, the setup process of the ECU is a time consuming procedure by testing the engine on the dynamometer. The engine system is highly non-linear, as simple mathematical equations have difficulty expressing the relationship between each engine parameter. Assumptions and simplifications must be made in conventional engine models. Artificial neural networks have the ability to model and optimize non-linear dynamic systems. Thus, the neural network may be an alternative approach for engine modeling if there is a lack of understanding of the engine system. This study aims to apply of artificial neural networks as virtual sensors in SI engine modeling. A Holden Vectra engine was employed for this study. The selected engine input parameters depend on the availability of the engine sensors. Power, fuel consumption and emissions are the output parameters which must be measured by the engine dynamometer and the gas analyzer. A total number of 8 inputs and 9 outputs were selected in the engine modelling. Two different artificial neural network structures namely: BackPropagation (BP) and Optimization Layer by Layer (OLL) neural networks were compared based on output parameters prediction. Two data sets were obtained for model training and prediction respectively. This work is a step towards establishing application of artificial neural networks as virtual sensors for IC engine modeling.


Publication status

  • Unpublished

Rights statement

Copyright 2007 the Author - The University is continuing to endeavour to trace the copyright owner(s) and in the meantime this item has been reproduced here in good faith. We would be pleased to hear from the copyright owner(s). No access or viewing until written permission from the University of Tasmania is obtained. Thesis (MEngSc)--University of Tasmania, 2007. Includes bibliographical references

Repository Status

  • Restricted

Usage metrics

    Thesis collection


    No categories selected


    Ref. manager