Mostafa Taherian ,Seyed Mahmoud Mousavi*,Hooman Chamani

1 Chemical Engineering Department,Faculty of Engineering,Ferdowsi University of Mashhad,Mashhad,Iran

2 Research Center of Membrane Processes and Membrane,Faculty of Engineering,Ferdowsi University of Mashhad,Mashhad,Iran

Keywords:Agent-based model Forward osmosis(PRO mode)Membranes NetLogo platform Simulation Water flux

A B S T R A C T Forward osmosis(FO),as an emerging technology,is influenced by different factors such as operating conditions,module characteristics,and membrane properties.The generalaim of this study was to develop a suitable( flexible,comprehensive,and convenient to use)computational tool which is able to simulate osmosis through an asymmetric membrane oriented in pressure retarded osmosis(PRO)mode in a wide variety of scenarios.For this purpose,an agent-based model was created in NetLogo platform,which is an easy-to-use application environment with graphical visualization abilities and well suited for modeling a complex system evolving over time.The simulation results were validated with empirical data obtained from literature and a great agreement was observed.The effect of various parameters on process performance was investigated in terms of temperature,cross- flow velocity,length of the module,pure water permeability coefficient,and structural parameter of the membrane.Results demonstrated that the increase in all parameters,except structural parameter of the membrane and the length of module led to the increase of average water flux.Moreover,nine different draw solutes were selected in order to assess the influence of net bulk osmotic pressure difference between the draw solution(DS)and feed solution(FS)(known as the driving force of FO process)on water flux.Based on the findings of this paper,the performance of FO process(PRO mode)can be efficiently evaluated using the NetLogo platform.

1.Introduction

An agent-based model is a promising method to simulate complex systems consisting of autonomous decision-making agents[1].During this kind of simulation,each agent acts as a single entity following its own set of characteristics in response to the conditions that the agent deals with[2–4].Agent-based modeling is an applicable tool for process design which provides a natural description of a system[3].It has been used in various application fields[5–7].

Forward osmosis(FO)is an emerging osmotic ally driven membrane process for water and wastewater treatment which has gained significant interest in recent years[8,9].In addition,FOhas drawn research attention for various applications including power generation[10],pharmaceutical applications[11],liquid food concentration[12],and desalination[13].This can mainly be attributed to its relatively low energy requirement[14,15].In this process,water automatically crosses a highly-selective membrane from a feed solution(FS)of lower osmotic pressure to a draw solution(DS)of higher osmotic pressure[16,17].

Most FO membranes have an asymmetric structure consisting of a thin active layer and a porous support layer.This design results in two membrane orientations,namely the membrane dense selective layer facing the FS(FO mode)and the membrane dense selective layer facing the DS(pressure retarded osmosis or PRO mode)[18,19].Convective water flow from the FS to the DS causes a buildup of solute at the membrane surface referred to as concentration polarization(CP)[20].FO is a complex process due to the CP which makes the process nonlinear.In addition,multiple agents such as the membrane structure,temperature,and cross- flow velocity of both solutions are all responsible for the process performance[21,22].

Many studies have shown that higher water flux can be obtained in PRO mode compared to another membrane orientation.However,the PRO mode is more prone to membrane fouling[19,23].The results indicated that the components of FS and the degree of concentration have a direct impact on the selection of membrane orientation[18].PRO mode may be more favorable when FSs with lower fouling propensities such as brackish water desalination are faced or where intensive concentration is unimportant such as power generation[18,24].

Despite the advantages of PRO mode membrane orientation in FO applications,very few modeling studies have been reported to investigate its performance.Tang et al.[25]presented a model for FO flux and demonstrated that internal concentration polarization(ICP)in PRO mode membrane orientation was less dominant compared to FO mode.Li et al.[26]investigated the effect of porous support layer on ICP using finite element method.In addition,a model was developed by Zhao et al.[27]which showed that physicochemical properties of the solution against the support layer influenced ICP.

One of the most important challenges in the literature is the lack of a modeling and simulation platform for the study of PRO mode that can well predict the water flux for different draw solutes in various conditions.Moreover,taking into consideration the possibility of employing this kind of orientation in future FO applications,a comprehensive investigation needs to be conducted.As mentioned earlier,FO process depends on multi agents which make it complex.Thus,agent-based modeling which is applicable for complex processes can be efficiently used in this area.A literature review showed that no reported study was found to use agent-based modeling approach for PRO mode.The authors' previous work[28]showed the quite well potential of the agent-based simulator for performance prediction of another applicable membrane orientation in FO process(FO mode).The FO performances are totally different in these two membrane orientations,resulting in various outcomes of the process[18,29].In order to fully investigate PRO mode process performance by an efficient method,as a new approach,our previous work was developed for simulation of this membrane orientation using the NetLogo platform as an agent-based model.

Comparing different agent-based modeling tools,it appears that NetLogo is a strong tool that has an appropriate environment for users.It has a special programming language which is easier compared to Java and C(Table S1 in supporting information)[30].

In the present study,the performance of PRO mode module that consisted of FS and DS compartments and a membrane(the FS was placed against the support layer of membrane and the DS was placed against the active layer of membrane)was investigated based on the NetLogo simulation environment.The key governing equations were first introduced to develop this model.Then,an iteration algorithm for solving the equations based on the well-defined multi-agent modeling forms was presented.The modeling was validated using a series of PRO experiments.Armed with the developed model,it was aimed to examine the influences of using various draw solutes,the operating conditions(temperature and cross- flow velocity),the module characteristic(length of the module),and the membrane properties(pure water permeability coefficient and structural parameter)on the PRO mode water flux.

2.Theory

2.1.Governing equations

In osmotic processes,CP may occur at both sides of the membrane.At FS side,the solute is concentrated on the membrane surface and at DS compartment,the solute is diluted on the membrane surface[31].In PRO mode,considering the membrane orientation,ICP occurs at FS side and external concentration polarization(ECP)occurs at DS side,respectively,as shown in Fig.1[31].Therefore,both ECP and ICP have to be considered in the process modeling,each one has to be developed and then coupled with each other to exactly determine the water flux.Both polarizations have a large impact on the process performance,causing the reduction of osmotic pressure on the membrane surface[31,32].

In PRO mode,water enters the porous support layer,being transported into the DS by diffusion through the active layer.However,the solute of the FS,which enters the support layer by the convection flow of water,cannot simply cross the active layer.Thus,the concentration is increased in its vicinity.This phenomenon is called concentrative internal concentration polarization(CICP)[33].The following equation is presented to describe the effect of ICP and its relationship with the water flux and other membrane constants[33]:

Fig.1.Schematic diagram ofCP profile through an asymmetric membrane oriented in PRO mode.

where Jwateris the water flux,B is the solute permeability coefficient from the active layer of membrane,A is the pure water permeability coefficient,πD,mis the osmotic pressure of DS on the membrane surface,πF,bis the bulk osmotic pressure of FS,and KFis the solute resistivity of FS for diffusion within the porous support layer,defined by:

where DFis the diffusivity of solute in the FS and t,τ,ε,and S are the thickness,tortuosity,porosity,and structural parameter of the membrane support layer,respectively.KFis a criterion to determine the extent of ICP,indicating the easiness of solute diffusion into and out of support layer[18].

For membranes having high water flux and solute rejection,B is negligible compared to the other variables of Eq.(1).Thus,ignoring solute permeation from the membrane,the water flux can be obtained from Eq.(1)as following:

Exponential function in the above equation is a modification coefficient to consider the CICP effect defined as[33]:

where πF,mis the osmotic pressure of FS on the membrane surface and positivity of the exponential function also implies that πF,m＞ πF,b.

The water convection flow,on the other hand,transfers the draw solute from the membrane surface to the bulk solution.Therefore,this flow causes the decrease in effective osmotic pressure at DS side.This phenomenon is called dilutive external concentration polarization(DECP).The DECP effect is defined using film theory[33]:

whereπD,bis the bulk osmotic pressure ofDS,and kDis the mass transfer coefficient at DS side.As seen in Fig.1,the concentration on the membrane surface is less than that of the bulk solution in DECP.In the above equation,negativity in the exponential function also indicates this fact.The ratio of solute concentrations is considered equivalent to their corresponding osmotic pressure in Eqs.(4)and(5).

In Eq.(5),mass transfer coefficient,kD,is obtained using:

where ShDis Sherwood number of the DS,DDis the diffusivity of the DS,and dhis the hydraulic diameter which is obtained by the following equation for a rectangular channel:

where H and W are the height and width of the channel,respectively.Sherwood number is also determined in the rectangular channel considering the flow regime governing the system using[33]:

where L is the channel length,ReDand ScDare Reynolds number and Schmidt number at DS side,respectively,obtained from the following equations:

whereρD,μD,andνDare the density,viscosity,and cross- flow velocity of the DS,respectively.By substituting Eq.(5)into Eq.(3)for the osmotic pressure of DS on the membrane surface,Eq.(12)is achieved:

Both ECP and ICP have been considered in the above equation which is used to determine the water flux in PRO mode.

2.2.Thermodynamic properties of the feed and draw solutions

In the recommended model,extensive data library was collected in order to determine thermodynamic properties of different solutes.The thermodynamic properties including osmotic pressure,density,viscosity,and diffusivity at different concentrations of solutions are listed in Table S2(see supporting information)[34–42].In the proposed model,the variation of these parameters during the process was considered,due to the net movement of water and changes in the concentration of FS and DS.

3.Simulation Approach

Various agent-based model tools have been developed for simulating complex phenomena such as Swarm,NetLogo,Repast,and MASON[30].A comparison of different tools used in agent-based modeling is presented in Table S1 in supporting information.NetLogo,an agent-based modeling environment,is an appropriate platform for the modeling of complex phenomena.It is due to the features such as providing a user friendly environment,comprehensive documentation,a large collection of pre-written models in its model library,and graphical visualization abilities[43].Fig.2 indicates a schematic of the process used in the system simulation.The PRO mode simulation was developed using NetLogo 5.0.5 environment(available at http://ccl.northwestern.edu/netlogo/,developed by Northwestern University Center for Connected Learning and Computer-Based Modeling).

The agent-based model consists of a number of elements(agents)characterized by some features.Agents of NetLogo(typically called a‘turtle’)interact with each other and with the model's world(called‘patches’)by means of pre-defined rules written by users[30,44].In the agent-based modeling and simulation,the agents of the process were first specified.Then,the specific properties of agents and the relationships among them were defined in the form of agent-based model.The agents of the process and related specifications of each agent(Table S3)and the schematic of input and output parameters of the provided simulation(Fig.S1)are presented in supporting information.

Fig.2.Schematic layout of the process applied for simulation(baseline data shown in this figure used as basic status of simulations).

The NetLogo platform has three separate tabs,namely code tab,info tab,and interface tab.At the code tab,a program was developed considering the governing equation and based on the NetLogo programming language.The program used in the NetLogo simulator had three main parts including the definition of global variables,setup procedure to start the simulation,and go procedure(run continuously by the system).While preparing the model,initial characteristics of the solutions were determined according to the amounts of input parameters.Pure water flux and other output parameters were then calculated by the algorithm demonstrated in Fig.S2 in supporting information.

In addition,at the interface tab,input and output parameters were planned based on the created slider,input,chooser,monitor,and plot.According to the presented platform,changes in output parameters could be followed using plots and monitors.NetLogo simulation environment is shown in Fig.S3 in supporting information.

During the simulation process,moles of water and solute in the FS and DS had random movements inside their corresponding cells and only the moles of water were transferred to the DS side from the FS side based on the equations governing the process.Thus,the FS gradually became more concentrated and the DS became more diluted until the condition of osmotic equilibrium was established in the process.Osmotic equilibrium conditions in this process were established at equal concentrations(if the solutes of both solutions were the same)or equal osmotic pressures(if the solutes were different)[23,45].

4.Results and Discussion

4.1.Model validation

In order to show the applicability and feasibility of the presented agent-based simulator environment,two different experimental conditions were used to determine the degree of model validation.

Empirical data reported by McCutcheon and Elimelech[33]for dilution of 1.5 mol·L-1aqueous NaCl DS by different concentrations of aqueous NaCl FS ranging from 0.05 to 1.0 mol·L-1with the membrane orientation of PRO mode was used for model validation.Experimental system of the process contained symmetric cells at two sides of the membrane with channel dimensions of 0.026 m width,0.077 m length,and 0.003 m height.Flat sheet cellulose triacetate(CTA)FO membranes(Hydration Technologies,Albany,OR)were applied in this experiment.The cross- flow velocity oftwo solutions in counter current mode equaled 0.458 m·s-1.Other information regarding the process conditions was mentioned in the literature[33].

Moreover,empirical data presented by Zhao et al.[24]was used as another case study.Experimental conditions in this research were as follows:pure water was used as FS and aqueous NaCl solution with different concentrations was used as DS.Experimental system of the process consisted of symmetric cells on both sides of the membrane with module characteristics of 0.06 m width,0.08 m length,and 0.002 m height.The applied membrane was the same as that of the aforementioned study.The cross- flow of two solutions was counter current and their velocity equaled 0.25 m·s-1.More information on the experimental conditions can be found in the literature[24].The empirical data and simulation results are compared in Fig.3.

According to Fig.3,the trend of changes in water fluxes was well predicted by the model.Table 1 indicates the amount of error and standard deviation percentage of data.Based on the obtained results,it can be observed that the developed agent-based model has the ability to appropriately simulate the experimental conditions.

4.2.Influence of the driving force of the process

In order to determine the effect of bulk osmotic pressure difference between the DS and FS(known as the driving force of FO process),nine different solutes were selected having the potential to be used as draw solutes.Most of the studies in the field of FO process were conducted using NaCl as solute[19],while the potential of different solutes was evaluated in this research.

Fig.3.Comparison of the simulation results and empirical data obtained from(a)McCutcheon and Elimelech[33]and(b)Zhao et al.[24]study.

The following simulation procedure was carried out.In the first phase,to investigate the ECP effect,the NetLogo platform was operated using deionized(DI)water as feed with a gradual increase in the DS concentration to a certain degree(depending on the available data for each solute).In the next phase,to examine simultaneously the ICP and ECP effects,a gradual increase was applied in the concentration of aqueous NaClsolution as feed by keeping constant the specified DS concentration.The effect of osmotic pressure difference of FS and DS on the water flux is shown in Figs.4 and 5 for draw solutes producing a higher pressure difference(NaCl,KCl,NH4HCO3,MgCl2,and CaCl2)and those generating a lower pressure difference(MgSO4,Glucose,Na2SO4,and Sucrose).Similar water flux trends are observed in the previous studies[22,23,33].

Each solute produced different water fluxes equivalent to the driving force that it could create.Among different solutes,NaCl and KCl had high potential to produce water flux with respect to their corresponding osmotic pressure.MgCl2and CaCl2generated the highest osmotic ressure compared to the other solutes(due to high solubility in water),although the water flux produced by these two solutes wasonly comparable to NaCl and KCl.This was probably because of the effect of CP for MgCl2and CaCl2.The other draw solutes except for those aforementioned produced lower water flux owing to the low solubility in water and the strong effect of CP.

Table 1Statistical parameters for comparison of the simulation results and empirical data

Fig.4.The influence of osmotic pressure difference of FS and DS on water flux for NaCl,KCl,NH4HCO3,MgCl2,and CaCl2 as solute(1 atm=101325 Pa).

Fig.5.The influence of osmotic pressure difference of FS and DS on water flux for MgSO4,glucose,Na2SO4,and sucrose as solute.

On the other hand,a logarithmic increase in water flux in the first phase was only attributed to DECP effect because of using DI water as feed.However,in the next phase,water flux was dropped because of the decrease in the effective osmotic pressure difference and creation of CICP[33].

4.3.Influence of the effective parameters on water flux

By developing the presented model based on the baseline data(shown in Fig.2),the graph of average water flux for dilution of 3 mol·L-1aqueous NaCl DS was plotted(Fig.6).The results obtained from studying the effect of operating conditions,module characteristics,and membrane properties are reported based on the comparison with the baseline outputs.

Fig.6.Average water flux versus concentration of FS using the baseline data.

Fig.7.The influence of cross- flow velocity on water flux.

4.3.1.Influence of operating conditions

In order to study the effect of flow velocity on water flux,the simulation was carried out at three different levels of DS cross- flow velocity(0.2,0.4,and 0.6 m·s-1).As shown in Fig.7,with the increase in cross flow velocity from 0.085 to 0.2,0.4,and 0.6 m·s-1,water flux was increased by about 13.12,29.38,and 44.76 percent,respectively,in the 0.1 mol·L-1FS.Similarly,change in water flux percentage was increased for other concentrations with the increase in the DS cross- flow velocity.This was due to the increase in mass transfer coefficient of the DS(kD)and the decrease in ECP effect.Decreasing trend of change in water flux percentage at all cross- flow velocities with the increase in FS concentration was due to the rise in ICP at feed side.Furthermore,the change in cross- flow velocity could not compensate for the adverse effect of ICP.These results are corroborated by the findings of other research[19,21,46,47].

The percentage of changes in water flux compared to the baseline values was investigated at different DS and FS temperatures(TF=303 and TD=298 K,TF=298 and TD=303 K,TF=303 and TD=303 K).As can be seen in Fig.8,temperature had a positive effect on water flux.Generally,temperature may affect the thermodynamic properties of solutions such as osmotic pressure,dynamic viscosity,and diffusivity.According to the outputs which controlled the simulation parameters,dynamic viscosity was decreased with the increase in temperature.However,diffusivity and osmotic pressure were increased.

Effective osmotic pressure difference was reduced with the increase in FS temperature.However,KFwhich implies the extent of ICP was reduced because of the increase in solute diffusivity of FS according to Eq.(2).On the other hand,the increase in DS temperature led to the increase in effective osmotic pressure of the process as well as the increase in kDdue to the increase in diffusivity of the DS.Accordingly,the increase in kDcaused the decrease in ECP effect at DS side.

The degree of improvement in water flux with the increase in DS temperature exhibited only a little difference compared to the 303 K isotherm process.In addition,the total mass of the DS requiring thermal energy to increase temperature was less than that of FS.Therefore,the most effective way was to increase DS temperature because improvement in water flux was achieved with less input energy.These results support the findings of previous research regarding the effect of temperature on water flux[19,22,33].

Fig.8.The influence of temperature on water flux.

4.3.2.Influence of module characteristic

In order to study the effect of module length,different lengths were selected(0.1,0.5,and 1 m).Fig.9 indicates the changes in water flux compared to the baseline values.For example,with the increase in module length from 0.077 to 0.1,0.5,and 1 m at 0.1 mol·L-1FS,water flux was decreased by around 2.06,16.99,and 24.17%,respectively.This decreasing trend was also observed in other FS concentrations.The decrease in water flux might be attributed to the decrease in concentration difference of FS and DS and consequently the decrease in the driving force of the process.Furthermore,with the increase in FS concentration,increase in the percentage of change in water flux was observed in different lengths.This was due to the negative impact of ICP in higher FS concentrations.The influence of module length found in a similar research[48],accords with the results of the present study.

Fig.9.The influence of length of module on water flux.

4.3.3.Influence of membrane properties

The percentage of changes in water flux compared to the baseline values was determined by changing the pure water permeability coefficient at different concentrations of FS(Fig.10).At 0.1 mol·L-1FS,water flux was increased by about 4.02,7.81,and 11.29%with the increase in pure water permeability coefficient from 2.82×10-12to 3.5×10-12,4.5 × 10-12,and 6 × 10-12m·Pa-1·s-1,respectively.This could be proved according to the direct relationship of flux and pure water permeability coefficient based on Eq.(12).In addition,the percentage of change in water flux was less in higher concentrations of FS due to the negative effect of ICP.These findings are in agreement with other previous research results[21,47].

In order to study the effect of structural parameter of the membrane,three different amounts of S(1×10-4,2×10-4,and 5×10-4m)were investigated compared to the baseline values(Fig.11).The percentage of change in water flux was decreased with the increase in the amount of structural parameter of the membrane.Since the solute resistivity for diffusion into the support layer had a direct relationship with structural parameter of the membrane according to Eq.(2),the decrease in flux could be attributed to the increase in KF.Moreover,the decrease in the value of structural parameter of the membrane could be introduced as a remedy for minimizing ICP effect.There is a good agreement between the results from investigating this parameter and the findings of similar studies[21,33,47,49].

Comparing the amounts of change in water flux due to the effect of this parameter and other studied parameters,structural parameter of the membrane was considered as a key parameter in improving the process performance.Thus,selecting an appropriate membrane is of significant importance.

Fig.10.The influence of pure water permeability coefficient on water flux.

Fig.11.The influence of structural parameter of the membrane on water flux.

5.Conclusions

The main purpose of this study was to examine the feasibility of using agent-based model and its ability to analyze FO process in PRO mode membrane orientation.Therefore,NetLogo environment,as a novel approach,was applied for simulation of this process.After validation of the modeling,the effective parameters including cross- flow velocity,temperature,length of the module,pure water permeability coefficient,and structural parameter of the membrane were examined.The simulation results showed that the increase in all parameters,except the length of the module and structural parameter of the membrane had a positive impact on average water flux.Amongst all studied parameters,structural parameter of the membrane played an important role in influencing the performance of the process.Moreover,compared to the effect of ICP,the improvement of water flux by declining the effect of ECP,although appreciable,was marginalized.Changing the internal structure of the porous support layer of the membrane is the most efficient method to lessen the ICP effect.Therefore,a new type of FO membrane consisting of only a thin dense selective layer without porous support layer,which has an adequate structural strength for FO operation,would completely eliminate the unfavorable effect of ICP on water flux reduction.Results of this study suggest that the proposed NetLogo platform which is adjustable through graphical user interface can easily predict the behavior of FO process(PRO mode)in different scenarios to investigate the parameters affecting the process performance,which helps further advancements in this field.

Nomenclature

A pure water permeability coefficient,m·Pa-1·s-1

a effective membrane area,m2

C0initial solute concentration of solution,mol·L-1

D diffusivity,m2·s-1

dhhydraulic diameter,m

H channel height,m

Jwaterwater flux,L·m-2·h-1

K solute resistivity for diffusion within the membrane support layer,s·m-1

k mass transfer coefficient,m·s-1

L channel length,m

MWwatermolecular weight of water,g·mol-1

Re Reynolds number

S structural parameter of the membrane support layer,m

Sc Schmidt number

Sh Sherwood number

T temperature,K

t thickness of the membrane support layer,m

V0initial volume of solution,L

v cross- flow velocity,m·s-1

W channel width,m

xwaterweight fraction of water in solution

ε porosity of the membrane support layer

μ dynamic viscosity,Pa·s

π osmotic pressure,Pa

ρ density,kg·m-3

τ tortuosity of the membrane support layer

?watervolumetric fraction of water in solution

Subscripts

D draw solution

F feed solution

Supplementary Material

Supplementary data to this article can be found online athttps://doi.org/10.1016/j.cjche.2018.01.032.