Prediction, optimization, and validation of the combustion effects of diisopropyl ether-gasoline blends: A combined application of artificial neural network and response surface methodology

Abstract

This research study mainly focuses on identifying the significant factors to be considered to discover the accuracy and reliability of the predictive models. The experimental results were employed to develop three different models: an artificial neural network (ANN), a response surface methodology (RSM), and a hybrid model. Brake thermal efficiency, specific fuel consumption, and regulated emissions were predicted using ANN, and inputs such as fuel blend concentration, CR, and engine speed were optimized using the RSM and hybrid models. The accuracy and reliability of the model results were validated with the least mean square error, mean absolute percentage error, and a higher signal-to-noise ratio. The higher R2 between 0.99426 and 0.9998 was observed by ANN whereas R2 by RSM and the hybrid model were relatively less. Similarly, the mean square error of ANN was relatively less compared to RSM and hybrid. However, the mean absolute percentage error observed in the validation test results for the optimized input parameters discovered by RSM, was less than 5 % for all the responses and higher in the hybrid model. Thus, the authors concluded that the ANN's predictive ability was much higher and RSM is the best suited for optimizing the engine parameters compared to the hybrid model.