Resolution methods for adsorption models, batch and continuous mode, and its impact on process implementation: An experimental and statistical comparison

• Published in: Journal of water process engineering 2024-02, Vol.58, p.104888, Article 104888
• Main Authors: de Oliveira, Júlia Toffoli, Estumano, Diego Cardoso, Féris, Liliana Amaral
• Format: Article
• Language: English
• Subjects: Adsorption models; Bayesian statistics; Maximum adsorption capacity; Parameter estimation; Process implementation
• ISSN: 2214-7144; 2214-7144
• DOI: 10.1016/j.jwpe.2024.104888

This work studies the application and validation of the Bayesian method proposed by Oliveira et al. (2023) for solving and modeling adsorption breakthrough curves, although, in this research, the method was applied for both batch and continuous mode, through: kinetic, equilibrium isotherms and breakthrough curve models; using experimental data from the literature. In addition, an analysis regarding the estimation and non-estimation of the maximum adsorption capacity parameter was performed and the results found in this research were compared with the ones obtained by the authors, which all applied the minimum square method. Finally, a detailed investigation of the selected models and the parameter values, in terms of practical purposes and process implementation impacts, was carried out. The results shows that the proposed Bayesian technique, applying the maximum adsorption deterministic, was effective for estimating the adsorption parameters based on the convergence of the Markov chains, and the method was able to predict the behavior of the curves. Also, in all cases studied in this work, by leaving the maximum adsorption capacity as a random variable it provoked in the over or underestimation of the parameter (kinetics: an overestimation of 15 %; isotherms: overestimation of 21 %; breakthrough-curves: underestimation of 25 %), showing the importance of its non-estimation. In addition, this research proved that only applying the coefficient of determination (R2) is not sufficient to select the most adequate model, since in all cases some models obtained the same value of R2 (kinetics: 0.993 for pseudo-first and pseudo-second order; isotherms: 0.998 for Langmuir and Sips; breakthrough-curves: 0.998 for all models applied by the authors) and the proposed technique was able to change the selected models by the authors or directly select the best one without additional analysis. Thus, it is possible to conclude that the method is a more reliable tool for solving adsorption models, since this approach can often change the selected model, directly influencing the elucidation of its mechanisms, causing error in the application and improvements of the process in a larger scale. [Display omitted]