Artificial Neural Network Prediction of Diesel Engine Performance and Emission Fueled With Diesel–Kerosene–Ethanol Blends: A Fuzzy-Based Optimization

Abstract

The present study explores the impact of ethanol on the performance and emission characteristics of a single cylinder indirect injection (IDI) Diesel engine fueled with Diesel–kerosene blends. Five percent ethanol is added to Diesel–kerosene blends in volumetric proportion. Ethanol addition to Diesel–kerosene blends significantly improved the brake thermal efficiency (BTE), brake specific energy consumption (BSEC), oxides of nitrogen (NOx), total hydrocarbon (THC), and carbon monoxide (CO) emission of the engine. Based on engine experimental data, an artificial neural network (ANN) model is formulated to accurately map the input (load, kerosene volume percentage, ethanol volume percentage) and output (BTE, BSEC, NOx, THC, CO) relationships. A (3-6-5) topology with Levenberg–Marquardt feed-forward back propagation (trainlm) is found to be optimal network than other training algorithms for predicting input and output relationship with acceptable error. The mean square error (MSE) of 0.000225, mean absolute percentage error (MAPE) of 2.88%, and regression coefficient (R) of 0.99893 are obtained from the developed model. The study also attempts to make clear the application of fuzzy-based analysis to optimize the network topology of ANN model.