Neural network modelling of adsorption isotherms

Abstract

This paper examines the possibility to use a single neural network to model and predict a wide array of standard adsorption isotherm behaviour. Series of isotherm data were generated from the four most common isotherm equations (Langmuir, Freundlich, Sips and Toth) and the data were fitted with a unique neural network structure. Results showed that a single neural network with a hidden layer having three neurons, including the bias neuron, was able to represent very accurately the adsorption isotherm data in all cases. Similarly, a neural network with four hidden neurons, including the bias, was able to predict very accurately the temperature dependency of adsorption data. NomenclatureA constant for Freundlich equation (kg ethanol/ kg adsorbent) (m 3 /kg ethanol) 1/n b constant for Langmuir, Toth and Sips equations; temperature dependent variable for the TD-Toth equation (m 3 /kg adsorbate) b o constant for the TD-Toth equation (m 3 / kg adsorbate) c fluid adsorbate concentration (kg adsorbate/m 3 ) n constant for the Freundlich and Sips equations (dimensionless)

Prediction of the isotherms of human IgG adsorption on Ni(II)-IDA-PEVA membrane using artificial neural networks

“…As expected, the Levenberg-Marquardt algorithm combined with Bayesian regularization technique was able to avoid overfitting problems even for an excessive number of neurons. These results are in accordance with that presented by Morse et al (2011).…”

“…A successful training can result in ANNs that can perform tasks such as predicting an output value, classifying an object, a function approaching, and recognizing a pattern in multivariate data (Dayhoff and DeLeo 2001). Morse et al (2011) indicated the use of neural networks for modeling adsorption isotherms due to the inherent complexities of adsorption process, in which the adsorbentadsorbate interaction may depend on several conditions like temperature, pH, compositions and/or the nature of adsorbent material. As confirmed by De Laurentiis and Ravdin (1994), ANNs are able to deal with these kinds of complex interactions, such as protein adsorption.…”

“…In the last years, artificial neural networks (ANN) proved to be an efficient alternative, since they are computational methodologies that perform multivariate analysis (Yetilmezsoy and Demirel 2008;Morse et al 2011;Ç elekli et al 2013;Dutta and Basu 2013). ANNs have been developed to mimic, mathematically, networks of biological neurons.…”

Machine learning–based optimization for hydrogen purification performance of layered bed pressure swing adsorption

“…Alireza Rahnama et al used the open dataset to train different machine learning models to predict the hydrogen weight percent and classify metal hydrides . By setting different adsorption isotherm (Langmuir, Freundlich, Sips, Toth, and TD Toth) models parameters, Morse et al trained the neural network model by generating data from the adsorption isotherm model formulas . By using Latin hypercube sampling (LHS) with a genetic algorithm to generate data, Anna et al used a multi‐objective optimization to optimize N 2 purity and recovery by an artificial neural network (ANN)…”

Sorption isotherms of ready‐to‐puff preconditioned brown rice: Development of classical models and artificial neural network approach

“…The standard and empirical models cannot adjust for compositional varietal differences, as modeling methods are based only on physical parameters. This shortcoming can be overcome by general class nonlinear models of artificial neural network (ANN) (Morse, Jones, Thibault, & Tezel, ; Myhara & Sablani, ). These heuristic models are recognized as good tools for dynamic modeling and are a useful tool for nonparametric regression.…”