Application of Artificial Neural Networks (ANN) for Prediction the Performance of a Dual Fuel Internal Combustion Engine

A neural networks (NN) model has been trained to predict the performance characteristics of a dual fuel internal combustion engine (ICE). In the network, back propagation (BP) neural network with Levenberg-Marquardt (LM) and scaled conjugate gradient (SCG) algorithms, single hidden-layer and logistic sigmoid transfer function has been used to optimise prediction model performance. The Neural Networks Toolbox of MATLAB 7 was used to train and test the NN model on a personal computer. In this investigation, a multi cylinder diesel engine was modified for duel fuel system to compare the experimental data with the prediction results obtained from NN model. Engine load, speed (rpm) and Diesel-NG ratio have been used as the input layers, while engine thermal efficiency, break specific fuel consumption (BSFC), exhaust temperature and air-fuel ratio have been used at the output layers. It is found that the RMS error values are smaller than 0.015, R2 values are about 0.999 and mean error smaller then 0.01% which indicate the NN model well matches with experimental results. The results of this investigation will be used to optimise the performance of future NG fueled engine.