Polyethylenimine-functionalized halloysite nanotube as an adsorbent for CO2 capture: RSM and ANN methodology

Abstract

In this study, artificial neural networks (ANN) and response surface methodology (RSM) were used for the modeling and optimization of CO2 adsorption in polyethylenimine (PEI)-functionalized halloysite adsorbents. Five-level four-factor central composite design (CCD) using RSM was used to optimize adsorption operational conditions, namely temperature of 20–50 °C and pressure of 1–9 bar, and PEI concentration of 10–40 wt%. The optimum temperature, pressure, and PEI wt% values are 20 °C, 9.00 bar, 29.49 wt% for the input variables, and the adsorption capacity value of 8 mmol/g for the response parameter, respectively. The Bayesian Regularization algorithm optimization technique was used as a learning algorithm. The accuracy of the optimized model was calculated using the mean squared error (MSE) and R2. The MLP and RBF models best MSE validation performances at 100 and 30 epochs, respectively, were 0.00011 and 0.00055. After using the experimental data as training data with the ANNs and RSM approach, the resulting model can yield satisfactory results by considering the effects of independent variables and their interactions on the objective function. The correlation coefficient (R2) and the adjusted R-squared (Adj-R2) are 0.9868 and 0.9846, respectively. Additionally, the CO2 adsorption performances are modeled using ANN for the optimization purpose. Due to the appropriateness of the adequate precision or ratio values of more than 4, the model presented for the system is valid. The SBET and the total pore volume of IMSiNTs/PEI nanocomposites (IMP-30) were 33.62 m2/g and 0.312 cm3/g, respectively. The mass flux, diffusion coefficient, and mass transfer coefficient for carbon dioxide gas in the single system have measured 4.44⨯10−24 mol/m2.s, 3.93⨯10−20 m2/s, and 2.58⨯10−16 m/s in 14 min, respectively.

Topics & Concepts