Diesel engine spray characteristics prediction with hybridized artificial neural network optimized by genetic algorithm

• Diesel engine spray characteristics were predicted with the ANN (artificial neural network) modeling tools.
• Spray penetration predictive capability has superiority over SMD (Sauter mean diameter).
• LM (Levenberg–Marquardt) training was found to have the least MSE (mean squared error) error with 20 number of neurons.
• ANN-GA prevails ANN in terms of lower MSE and higher R2 determination coefficient.

ANN (artificial neural network) modeling is adopted along GA (genetic algorithm) optimization method in order to investigate spray behavior as function of nozzle and engine variant parameters such as crank-angle, nozzle tip mass flow rate, turbulence, and nozzle discharge pressure. Spray quality is measured in SMD (Sauter mean diameter) and spray liquid tip penetration prospective. Experimental data were used at limited engine condition and elsewhere requisite data was acquired with the aid of curve fitting and extrapolation of CFD (computational fluid dynamics) numerical simulation results. Engine crank-angle, vapor mass flow rate, turbulence, and nozzle outlet pressure were taken as input layer while spray penetration and SMD were used as output layer. It is found out that Levenburg–Marquardt training algorithm has the least mean square error for ANN and ANN-GA (artificial neural network-genetic algorithm) at 24, 30 neurons in hidden layer with the amount of 0.8994, 0.3348, respectively. The coefficient of determination (R2) for penetration equals 0.994 whereas SMD yields lower amount of 0.992. By application of GA to optimize the network's interconnecting weights, R2 values have been enhanced to 0.999 for SMD and to 0.998 for penetration (both values are close to unity). Results indicate that the ANN-GA improved the spray specification modeling simply and with acceptable accuracy.