Development of Back Propagation Neural Network (BPNN) Model to Predict Combustion Parameters of Diesel Engine

Abstract

Effective utilization of fuel in diesel engines is the major challenge posed to the engine designers today, due to large demand of fuel. In this context, control of design, as well as operating parameters for better performance are focused in the present work. Experiments are done on a 4-s, variable compression ratio (VCR) diesel engine and the required data is collected by varying fuel injection timing (IJT), compression ratio (CR), load and fuel injection pressure (IP). The combustion parameters, viz. combustion duration (CD), ignition delay (ID), peak pressure (PP) and heat release (HR) are determined at various operating conditions, as these parameters could influence fuel consumption and performance. For the investigation purpose, an artificial neural network (ANN) with back propagation algorithm is adopted. ANN is trained with the experimental data. The number of nodes in the hidden layer is varied from 3 to 22, to architect a suitable network for the prediction of combustion parameters with good accuracy. Test results show that network with 4-19-4 architecture with trainlm algorithm can predict the four parameters (ID, CD, PP and HR) with correlation coefficients as 0.9892, 09892, 0.9944 and 0.9909 taken in the order.