Modeling of liquid fuel purification by the LTA zeolite using machine learning methods

Impurities have severe negative influences on both the application and energy content of the dimethyl aminoethyl azide (DMAZ) fuel. Therefore, it is necessary to dehydrate the DMAZ liquid fuel using an appropriate separation scenario. This separation is often performed through an adsorption process employing the LTA zeolite (Linde type A). Hence, this study aims to employ different artificial intelligence techniques to simulate the performance of LTA zeolite for dehydration of DMAZ liquid fuel. Initial water concentration, contact time, temperature, and zeolite per solution (DMAZ + water) mass ratio are those features utilized to estimate zeolite capacity for water adsorption. Systematic ranking analyses applied on several statistical criteria approved that the generalized regression neural network is the best smart model for the considered matter. The proposed generalized regression approach predicted a big reliable experimental database with the AARD = 5.68%, RMSE = 11.26, MAE = 6.97%, and R-2 = 0.9675. The influence of different operating conditions on the performance of zeolite adsorption capacity (ZAC) is investigated both experimentally and by using the generalized regression approach. Modeling results justified that the maximum water removal per unit mass of the LTA zeolite is about 0.202 in the optimum conditions.