Artificial Neural Network-Based Algorithm as a Powerful Tool in the Study of Adsorption Processes: Determining Complex Mechanisms in Dye-Reduced Graphene Oxide Systems

Abstract

Despite the widespread application of adsorption processes, identifying the underlying adsorption mechanism remains a challenging task because of the complexity arising from the numerous processes and subprocesses that may occur simultaneously. In this work, a multilayer perceptron (MLP) neural network-based algorithm was proposed to determine the complex mechanism of adsorption processes as a linear combination of isotherms derived from statistical mechanics. In addition, it accurately determines isotherm parameters, supporting statistical thermodynamic analysis. The algorithm was implemented using simulated and experimental data, and its performance was evaluated by comparison with Bayesian model averaging (BMA). Using experimental equilibrium adsorption data previously reported in the literature, no new experiments were performed, and kinetic, time-dependent, and economic analyses were outside the scope of this investigation. Using simulated data, the robustness of the network was tested by varying noise in the data and considering different initial values for the neurons. For the experimental data analysis, the adsorption of the dyes methylene blue (MB) and indigo carmine (IC) was investigated on reduced graphene oxide materials. This algorithm outperforms BMA and allows the analysis of adsorbents with heterogeneous surfaces, which probably present coexisting and competitive adsorption mechanisms. Also, it correlates the structure of the adsorbent along with the properties of the adsorbates through the determination of interactions involved. The results demonstrated that MB adsorption primarily follows the multilayer finite model (MLFM) mechanism associated with the BET type I isotherm. However, a transition to MLFM is verified to be associated with BET type IV at intermediate degrees of oxidation. For the IC adsorption process, the MLFM mechanism and BET type I isotherm are determined even at higher degrees of oxidation.