Predicting the Kinematic Viscosity and Cetane Number of Diesel-Biodiesel Blend using Neural Network and Empirical Models

Biodiesel, as a renewable and environmentally friendly fuel, is a feasible alternative to fossil diesel, which has gained great popularity in recent years. However, due to some undesirable properties such as higher viscosity, biodiesel must be blended with diesel in order to be utilizable in a diesel engine. Therefore, a reasonable approach is required for predicting the diesel-biodiesel blend properties. This study tries to estimate two substantial properties of blend, i.e. kinemattic viscosity (KV) and cetane number (CN), through neural network (NN) and empirical models which use pure properties of biodiesel (kinematic viscosity, boiling point, evaporation point, flash point, pour point, heat of combustion, cloud point, and specific gravity) as independent variables. In this regard, a three-layer feed-forward network with varying input parameters, training algorithms, transfer functions, and hidden neurons has been examined to predict the KV and CN of the diesel-biodiesel blend. Besides, the prediction capability of thirty empirical equations is investigated to determine the top equations describing blend properties. The result reveals that an ANN with three input parameters of biodiesel concentration (%), the CN of biodiesel, and biodiesel cloud point has the best prediction quality of CN with an R-value of 0.9961. Moreover, NN estimates the KV of blend with the highest correlation coefficient of 0.9985. The results corresponding to empirical equations also indicate that fractional-exponential equations are the best describer of the CN and KV of blend with R-values of 0.9947 and 0.9980, respectively.