Title: Analysis of semi-physical neural network structure for model-based ignition timing control of multi-fuel SI engines

Authors: Baitao Xiao; Shu Wang; Robert Prucka

Addresses: International Center for Automotive Research, Clemson University, 4 Research Drive, Greenville SC 29607, USA ' International Center for Automotive Research, Clemson University, 4 Research Drive, Greenville SC 29607, USA ' International Center for Automotive Research, Clemson University, 4 Research Drive, Greenville SC 29607, USA

Abstract: This research investigates the use of a semi-physical neural network for combustion phasing control of a multi-fuel spark ignition engine. The influence of model structure and training data set composition are each analysed for their influence on combustion phasing control accuracy under steady-state and transient engine operating conditions. The model structure developed for this research utilises both fuel sensitive and combustion related physical models as inputs, with an aim to minimise neural network size and increase its generalisation capability. Optimisation of the network structure is also studied to evaluate overall robustness and ensure control stability. Real-time engine test results show satisfactory combustion phasing control performance with multiple fuels for both steady state and transient conditions. Moreover, demonstration of the differences of controller performance brought by the steady state and transient training data samples will be also provided.

Keywords: spark timing control; combustion phasing control; model-based control; artificial neural networks; ANNs; ANN structure design; combustion modelling; CA50; semi-physical neural networks; ignition timing control; multi-fuel SI engines; spark ignition engines.

DOI: 10.1504/IJPT.2016.081796

International Journal of Powertrains, 2016 Vol.5 No.4, pp.358 - 374

Accepted: 20 May 2015
Published online: 26 Jan 2017 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article