Transient Air/Fuel Ratio Control of an S.I. Engine Using Neural Networks 960326
Engine Electronic Control (EEC) systems on spark ignition engines enable a high degree of performance optimisation to be achieved through strategy and calibration details in software, but development times and costs can be high. The range of functions performed by EEC systems, and the level of performance demanded, are increasing and new methods of development are required. In the paper, the use of neural networks in the development and implementation of open-loop control of air/fuel ratio during engine transient operating conditions is described. The investigation has addressed the definition of suitable networks, the procedure and data required to train these, and assessment of real-time performance of the implemented system. The potential benefits of the approach include reduced calibration effort and simplification of the control strategy. The results presented demonstrate that a small number of test data are sufficient to define the transient compensation requirements for the network. Errors in the definition of the training data are reduced by an iterative redefinition and training procedure. Three iterations have proved to be the optimum, with additional iterations failing to improve performance. The control of air/fuel ratio to within one unit is achieved at almost all of the transient test cases evaluated.
Citation: Shayler, P., Goodman, M., and Ma, T., "Transient Air/Fuel Ratio Control of an S.I. Engine Using Neural Networks," SAE Technical Paper 960326, 1996, https://doi.org/10.4271/960326. Download Citation
Author(s):
P. J. Shayler, M. S. Goodman, T. Ma
Affiliated:
University of Nottingham
Pages: 12
Event:
International Congress & Exposition
ISSN:
0148-7191
e-ISSN:
2688-3627
Also in:
Electronic Engine Controls 1996-SP-1149, SAE 1996 Transactions - Journal of Engines-V105-3
Related Topics:
Spark ignition engines
Neural networks
Education and training
SAE MOBILUS
Subscribers can view annotate, and download all of SAE's content.
Learn More »