Elsevier

Journal of Molecular Liquids

Volume 265, 1 September 2018, Pages 53-68
Journal of Molecular Liquids

Molecular dynamics, grand canonical Monte Carlo and expert simulations and modeling of water–acetic acid pervaporation using polyvinyl alcohol/tetraethyl orthosilicates membrane

https://doi.org/10.1016/j.molliq.2018.05.078Get rights and content

Highlights

  • Molecular simulations techniques and artificial intelligence modeling was used.

  • Water–acetic acid PV through PVA‑silicone based membranes were investigated.

  • PV separation index (PSI) was investigated under diversity of process conditions.

  • ACOR, DE and GA were employed for improving ANFIS modeling.

  • Errors for all models were too low and the accuracy of the simulation was very high.

Abstract

In this study, molecular dynamics (MD) and Monte Carlo (MC) simulations techniques were employed as well as artificial intelligence knowledge of ANFIS and GP to investigate water–acetic acid pervaporation (PV) separation through poly vinylalcohol (PVA)‑silicone based membranes under a wide range of experimental conditions. For the first time, three new optimization algorithms, namely ant colony optimization for continuous domains (ACOR), differential evolution (DE) and genetic algorithm (GA) were employed for improving ANFIS modeling. The GP creates a mathematical function or model for the estimation of pervaporation separation index (PSI) as a function of the input variables. ACOR-ANFIS and GA-ANFIS and GP had high accuracy (R2 = 0.9831, 0.9792 and 0.9722, respectively) but DE-ANFIS had a lower accuracy (R2 = 0.9610) as compared to other models. On the other hand, molecular simulation methods were used and the results of all simulation models were compared fairly to each other and to the experimental results of the literature. Also, some characterizations were taking place to investigate the characteristics of the simulated membranes with MS such as WAXD, and FFV and glass transition temperature was used to estimate the thermal properties of the simulated membranes.

Introduction

Membrane processes are one of the most important processes in separation science [[1], [2], [3], [4], [5]]. Pervaporation (PV) separation is particularly applied in organic compounds dehydration or separation of azeotropic mixtures or mixtures with a close-boiling point [6,7]. Thus, many experimental researches have been done to develop more selective membranes. However, they must be combined with scale-up and engineering attempts to obtain a real application in industry [8]. For designing integrated PV processes, developing a model that can explain the mass transfer and can be used in prediction of separation performance is essential [[9], [10], [11], [12], [13], [14], [15], [16]]. Mathematical models cannot commonly explain dynamic process behaviors and they are appropriate for certain feeds within particular conditions [[17], [18], [19]]. Conventional PV models characterize the mass transfer inside the membranes selective layer. Because of the intrinsic nonlinearity and complexity of the PV system, PV modeling by mathematical models is conditioned to the employment of many assumptions [20]. Due to the complexity of mass transfer models and lack of enough accuracy of the models in many PV systems, artificial intelligence can be utilized to model the PV system with high accuracy.

Recently, molecular simulation [21,22] has progressed to such a level that it can predict transport and structural properties of membranes [[23], [24], [25], [26], [27], [28]]. However, few works have been done on property–structure relationships of PVA membranes with molecular simulation. Wei et al. [29] explored effects of different ratios of the blend PVA/poly (acrylic acid) membrane on its properties using MD study. There are other studies on TEOS in polymeric membranes by MD [30,31]. But there are no MD simulation studies on pervaporation separation available for PVA/TEOS membranes. Fuzzy systems have particular benefits over conventional methods, especially when obscure data or related knowledge is involved in a process [[32], [33], [34], [35], [36], [37]]. Throughout the last years, fuzzy systems have recognized their popularity as alternative methods to data processing. These methods can minimize the matter of fuzzy modeling by learning the ability of ANN and have been reported since the beginning of 1990s [38,39]. Because the behavior and structure of ANFIS are extremely applicable [40], it has been considered as a fundamental element for interpretation researches [41]. The training procedure of parameters of ANFIS is the crucial issue. Many of them are on the basis of gradient and gradient determination in many steps is not simple. The chain rule that should be employed might trap in local minima. Hence, this study is attempted to develop an approach that can update the primary variables rapidly and conveniently compared to the gradient approach. In the gradient approach, variables convergence is not so fast and relies on primary variables value and consequently, discovering the optimum learning rate is extremely complicated [42]. In this study, different optimization algorithms namely DE, ACOR and GA were used to find the optimum algorithm in ANFIS training stage.

Sargolzaei et al. [43] applied ANFIS, back propagation artificial neural network (ANN) and radial basis function to predict starch removal efficiency from starchy wastewater by a hydrophilic Poly(ether sulfones) membrane in a plate and frame membrane module. Their study investigated the performance of membrane by estimation of optimum input parameters which have an impact on removal percentage of chemical oxygen demand and the permeate flux. Four inputs were considered, including pH, flow, temperature of feed and concentration of permeate. The results displayed a satisfying agreement but the ANFIS prediction was more accurate than two other simulation methods. Many researchers have used black box models including GP [43] for separation processes modeling in order to derive a mathematical function for a system. Shokrkar et al. [44] utilized a GP to obtain a mathematical model for the flux prediction of mullite ceramic MF membrane in the treatment of wastewaters containing oil. The model employed input variables for operating conditions (filtration time and flux) and feed quality (temperature, cross-flow velocity, pressure, and oil content). The results acquired from the GP model show good accuracy having an average error of <5%.

In this study, the grand canonical Monte Carlo (GCMC) and MD simulation approaches was used to examine effects of different TEOS loading content on transport properties of PVA membranes with Materials Studio simulator program. To avoid the influence of other factors such as pressure, all other effective factors are kept fixed. The water and acetic acid transport properties such as diffusivity, solubility, PSI, and permselectivity have been investigated. Also, X-ray diffraction, glass transition temperatures and fractional free volume (FFV) of these simulated membranes have been calculated and discussed. Also, this simulation study delivers a guideline for tailoring TEOS-filled PVA membrane to gain the best water/acetic acid separation properties, including diffusivity, solubility, PSI and permselectivity. Also, the results will be compared with the results of the ANN simulation of this study and also the results of the experimental works. Besides, for the first time, the main purpose of this study is to model PV process with novel modeling methods under various operating conditions. These models can be utilized to analyze the permeation performance through the membrane at various operating conditions. The complex and nonlinear performance of PSI versus concentration of water, temperature of feed and TEOS mass ratio with respect to PVA were modeled. In this study, ANFIS is optimized with ACOR, DE and GA to modify the ANFIS parameters and generates a balance between generalization ability and complexity of models. Therefore, the principal purpose of this study is to propose ACOR-ANFIS, DE-ANFIS, GA-ANFIS and GP models to investigate the PV process.

Section snippets

Data attainment

The feed water content was changed in a range of 10 up to 90 wt% and the experimental temperature was changed from 30 to 50 °C. The TEOS/PVA mass ratio was changed as 0.5 (namely M-1), 1.0 (namely M-2), 1.5 (namely M-3) and 2 (namely M-4). The prepared membranes were equilibrated in the feed container at the predetermined experimental temperature around 2 h prior to executing the PV with a certain amount of water-acetic acid mixture. Following a steady state was achieved, the permeate was

MS theory

Permeation of target molecules through polymeric (or nanomaterial/polymeric) membranes is typically described by solution–diffusion theory. This theory considers three steps: I: on the membrane surface and the high-pressure side, the molecules are adsorbed. II: the molecules penetrate (diffuse) through the polymeric (or nanomaterial/polymeric) membranes, and III, they desorb at the low-pressure side of the surface of the membrane [46,47]. Permeability can be calculated as follow:PI=DI×SIwhere PI

Density

Density of different PVA membranes are indicated in Fig. 5. The density of membranes becomes constant during and at the end of 5000 ps-NPT MD simulation equilibration run. As seen and as mentioned earlier, densities of membranes are between 1.25 and 1.3 g/cm3 and the densities of these membranes become equilibrated after a certain time. In Table 1, the average density of all membranes with different loading content of TEOS has been tabulated. As it reveals, the density of membranes increases

Conclusion

MC, and MD simulation and ACOR-ANFIS, DE-ANFIS, GA-ANFIS and GP models with experimental inputs were compared to predict PSI in water–acetic acid mixtures PV separation. The novelty of GP and usage of optimization methods in ANFIS training stage in obtaining accurate models for water–acetic acid mixtures PV separation were demonstrated. The density of simulated membranes with MS approach increases with the content of the TEOS nanomaterials which is an obvious and desired result. Also, it can be

References (85)

  • S. Chellam

    Artificial neural network model for transient crossflow microfiltration of polydispersed suspensions

    J. Membr. Sci.

    (2005)
  • B. Cosenza et al.

    Nonlinear fuzzy control of a fed-batch reactor for penicillin production

    Comput. Chem. Eng.

    (2012)
  • J. Sargolzaei et al.

    Modeling and simulation of wind turbine Savonius rotors using artificial neural networks for estimation of the power ratio and torque

    Simul. Model. Pract. Theory

    (2009)
  • J. Sargolzaei et al.

    Membrane permeate flux and rejection factor prediction using intelligent systems

    Desalination

    (2012)
  • H. Shokrkar et al.

    Prediction of permeation flux decline during MF of oily wastewater using genetic programming

    Chem. Eng. Res. Des.

    (2012)
  • M.Y. Kariduraganavar et al.

    Pervaporation separation of water–acetic acid mixtures through poly (vinyl alcohol)‑silicone based hybrid membranes

    J. Membr. Sci.

    (2005)
  • K.-L. Tung et al.

    MD and MC simulation analyses on the effect of solvent types on accessible free volume and gas sorption in PMMA membranes

    Desalination

    (2006)
  • F. Mozaffari et al.

    Molecular dynamics simulation of diffusion and permeation of gases in polystyrene

    Polymer

    (2010)
  • M. Asghari et al.

    Molecular simulation and experimental investigation of temperature effect on chitosan-nanosilica supported mixed matrix membranes for dehydration of ethanol via pervaporation

    J. Mol. Liq.

    (2017)
  • M. Pazirofteh et al.

    Molecular dynamics simulation and Monte Carlo study of transport and structural properties of PEBA 1657 and 2533 membranes modified by functionalized POSS-PEG material

    J. Mol. Liq.

    (2017)
  • J. Fried et al.

    The atomistic simulation of the gas permeability of poly (organophosphazenes). Part 1. Poly (dibutoxyphosphazenes)

    Comput. Theor. Polym. Sci.

    (2000)
  • X.-Y. Wang et al.

    Structural, sorption and transport characteristics of an ultrapermeable polymer

    J. Membr. Sci.

    (2008)
  • A.Y. Musa et al.

    Molecular dynamic and quantum chemical calculations for phthalazine derivatives as corrosion inhibitors of mild steel in 1 M HCl

    Corros. Sci.

    (2012)
  • H. Sun et al.

    The COMPASS force field: parameterization and validation for phosphazenes

    Comput. Theor. Polym. Sci.

    (1998)
  • Y. Chen et al.

    Molecular simulation of CO 2/CH 4 permeabilities in polyamide–imide isomers

    J. Membr. Sci.

    (2010)
  • T. Mohammadi et al.

    Wastewater treatment using ultrafiltration at a vegetable oil factory

    Desalination

    (2004)
  • H. Jalalifar et al.

    Application of the adaptive neuro-fuzzy inference system for prediction of a rock engineering classification system

    Comput. Geotech.

    (2011)
  • K. Socha et al.

    Ant colony optimization for continuous domains

    Eur. J. Oper. Res.

    (2008)
  • X. Chen et al.

    A DNA based genetic algorithm for parameter estimation in the hydrogenation reaction

    Chem. Eng. J.

    (2009)
  • A.V. Penkova et al.

    Polyvinyl alcohol membranes modified by low-hydroxylated fullerenol C 60 (OH) 12

    J. Membr. Sci.

    (2015)
  • C.-P. Liu et al.

    Novel proton exchange membrane based on crosslinked poly (vinyl alcohol) for direct methanol fuel cells

    J. Power Sources

    (2014)
  • K.-S. Chang et al.

    Free volume and polymeric structure analyses of aromatic polyamide membranes: a molecular simulation and experimental study

    J. Membr. Sci.

    (2010)
  • F. Pan et al.

    Diffusion behavior of benzene/cyclohexane molecules in poly (vinyl alcohol)-graphite hybrid membranes by molecular dynamics simulation

    Chem. Eng. Sci.

    (2007)
  • A. Dashti et al.

    Recent progresses in ceramic hollow-fiber membranes

    ChemBioEng Rev.

    (2015)
  • V. Zargar et al.

    A review on chitin and chitosan polymers: structure, chemistry, solubility, derivatives, and applications

    ChemBioEng Rev.

    (2015)
  • M. Rezakazemi, A. Dashti, H.R. Harami, N. Hajilari, Fouling-resistant membranes for water reuse, Environ. Chem. Lett....
  • B. Baheri et al.

    Performance of PVA/NaA mixed matrix membrane for removal of water from ethylene glycol solutions by pervaporation

    Chem. Eng. Commun.

    (2015)
  • M. Shahverdi et al.

    Separation of ethylene glycol–water mixtures with composite poly(vinyl alcohol)–polypropylene membranes

    J. Appl. Polym. Sci.

    (2011)
  • S. Shirazian et al.

    Separation of CO2 by single and mixed aqueous amine solvents in membrane contactors: fluid flow and mass transfer modeling

    Eng. Comput.

    (2012)
  • S. Shirazian et al.

    Implementation of the finite element method for simulation of mass transfer in membrane contactors

    Chem. Eng. Technol.

    (2012)
  • S. Shirazian et al.

    Development of a mass transfer model for simulation of sulfur dioxide removal in ceramic membrane contactors

    Asia Pac. J. Chem. Eng.

    (2012)
  • G. Camera-Roda

    Performances of filled and unfilled PTMSP membranes in pervaporation

    Chem. Eng. Commun.

    (1998)
  • Cited by (43)

    • Neural modeling and simulation of molecular separation using amino acid salt solutions

      2021, Journal of Molecular Liquids
      Citation Excerpt :

      They indicated that the very low computational time for simulating the absorption system is the main privilege of the established neural method, which can be utilized as a rigorous simulation instrument for the design and comprehension of the systems with a separation basis. The design of artificial intelligence methods alongside numerical methods can accelerate the overall time of optimization and analysis of the simulation processes [52,53]. This combination framework can also provide a detailed understanding of the process, particularly when many input parameters are involved.

    • Evaluation of the effect of silica nanoparticles, temperature and pressure on the performance of PSF/PEG/SiO<inf>2</inf> mixed matrix membranes: A molecular dynamics simulation (MD) and design of experiments (DOE) study

      2021, Journal of Molecular Liquids
      Citation Excerpt :

      The advantages of computational methods and parameter assessments such as the design of experiment method reduce the required time for developing new products or optimizing an existing product as well as reducing the number of tests and research costs and safe testing without wasting resources [39,40]. Molecular dynamics (MD) and Monte Carlo (MC) simulations are suitable tools to study the morphological and transport properties of MMMs [41–45]. To the best of our knowledge, there is a lack of data on the effect of temperature and pressure on membrane permeabilities and selectivities, not only for mixed matrix membranes but also for neat, hybrid and composite membranes.

    View all citing articles on Scopus
    View full text