An FPGA implementation of an Artificial Neural Network for prediction of cetane number

An artificial neural network (ANN) was implemented on an FPGA to predict cetane number in diesel fuel from its chemical compositions, extracted by liquid chromatography (LC) and gas chromatography (GC). An MLP network is used. To train the MLP, two variants of the backpropagation algorithm are utilized: backpropagation with plummeting learning rate factor and backpropagation with declining learning-rate. By adjusting the ANNpsilas parameters the total sum square error in train phase and average error percent in test phase are reduced to 0.085 and 4.4018%, respectively. The number of occupied slices on the FPGA is 5971 which covers 55% of the chip.