Engine performance, combustion and emission characteristics of African pear (Dyacrodes edulis) seed oil biodiesel-petrodiesel blends (B25, B50, B75 and B100) tested on four-cylinder, direct-injection (DI), four-stroke, water-cooled Perkins 4:108 diesel engine over varying loads (5 Nm, 10 Nm, 20 Nm, 30 Nm and 40 Nm) and speed range of 1500–3500 rpm is presented. Exhaust emissions were recorded using Bacharach, PCA2 Qs1007 model gas analyzer. Brake thermal efficiency (BTE), brake specific fuel consumption (BSFC), carbon monoxide (CO), oxides of nitrogen (NOx) and hydrocarbon (HC) emissions were optimized as response variables considering load, speed and fuel blend as input variables using Nelder-Mead (NM) simplex method and multiple-input multiple-output artificial neural network (MIMO-ANN) of Levenberg-Marquardt algorithm. Multi layer perception (MLP) network back propagation was applied for non-linear mapping between the input and output variables. The developed ANN and NM models produced less deviations and exhibited high predictive accuracy with high correlation coefficients between 0.90 and 0.99. The values of absolute average deviation (AAD), root mean squared error (RMSE) and standard error of prediction (SEP) for all the responses except NOx were very low. Optimal responses of 38.963% BTE, 0.1096 kg/kW-h BSFC, 0.05% CO, 220.9173 ppm NOx and 20.1524 ppm HC at fuel blend of 69.58, 5.0, 40, 38.35 and 38.22, engine load of 10, 20, 10, 10 and10 Nm and engine speed of 980, 2650, 2000, 2000 and 2000 rpm respectively were established as viable routes for reduced fuel consumption, combustion emissions and high thermal efficiency via the NM approach. For these conditions, experimental values of the responses were found to be BTE of 42.00%, BSFC of 0.0506 kg/kW-h, CO emission of 0.046%, NOx emission of 232.01 ppm and HC emission of 20.703 ppm. The developed models produced the idealized results that are useful for predicting the engine performance and emission characteristics.
