27th International Conference on Advances and Trends in Engineering Materials and their Applications, AES-ATEMA RIO-DE-JANEIRO 2016, Rio de Janeiro, Brazil, 2 - 06 May 2016, vol.2016-January, pp.55-60
© 2016, Advanced Engineering Solutions [AES.COM] Ottawa, Canada. All rights are reserved.A machine learning modelling approach to predict nitrogen oxides (NOx) emission was accomplished for turbofan engines of commercial aircraft during the idle condition utilizing International Civil Aviation Organization (ICAO) emission database. Being the first non-conventional emission modelling for the idle phase in the literature, the derived model relates the input parameters consisting of bypass ratio, engine pressure ratio, maximum rated thrust, and fuel flow rate of the considered turbofan engine with the output parameter, NOx emission index during the idle. Multi-layer perceptron neural networks (NNs) with one- and two-hidden-layer architectures were trained by various learning algorithms; namely, conjugate gradient, Levenberg-Marquardt, delta-bar-delta, back-propagation with momentum, and Quickprop algorithms, so as to obtain the optimal model. The estimated NOx emission index values provided a good fitting with the actual emission index values found in ICAO database while the most accurate model was achieved by the Quickprop algorithm trained two-hidden-layer NN.