TY - JOUR
TI - Artificial neural network based prediction of engine-out responses from a biodiesel fuelled compression ignition engine
AU - Kezrane Cheikh
AU - Habib Houcine S
AU - Bayram Mustafa
AU - Alqahtani Sultan
AU - Alshehery Sultan
AU - Ikumapayi Omolayo M
AU - Akinlabi Esther T
AU - Akinlabi Stephen A
AU - Loubar Khaled
AU - Menni Younes
JN - Thermal Science
PY - 2023
VL - 27
IS - 4
SP - 3433
EP - 3443
PT - Article
AB - Numerical simulations, based on relatively complex physical models developed  for CFD, can accurately predict engine-out responses, but they require huge  memory space and/or computation time. In terms of resources and computer  time, artificial intelligence methodologies are more cost-effective. In this  work, we used an ANN to predict the performance and exhaust emissions of a  single-cylinder Diesel engine running on fossil diesel, biodiesel, and  their blends under various speed and load regimes. To perform the modeling,  we employed multilayer perceptrons and a back-propagation gradient algorithm  with momentum to train the network weights. The modification of the network  weights was done using the second-order method of Levenberg-Marquardt, and  the technique of early termination was utilized to avoid overtraining the  model. The study involved using 70% of the complete experimental data to  train the neural network, allocating 15% for network validation, and  reserving the remaining 15% to evaluate the trained network effectiveness.  The ANN model that was created demonstrated remarkable accuracy in  predicting both engine performance and emissions. This is evident from the  strong correlation coefficients observed, which ranged from 0.987 to 0.999,  as well as the low mean squared errors ranging from 7.44•10-4 to 2.49•10-3.