Professional Documents
Culture Documents
Dynamic Modeling of A Simple Reverse Osmosis Desalination Plant For Advanced Control Purposes
Dynamic Modeling of A Simple Reverse Osmosis Desalination Plant For Advanced Control Purposes
3
Marriott Marquis Hotel at Times Square
New York City, USA, July 11-13, 2007
Abstract— Reverse Osmosis (RO) has become an important process dispose of a dynamical model in order to be able to apply
for desalting water. It requires an efficient control system to maintain advanced techniques of model-based control.
costs at acceptable level and therefore dynamic models are essential. Reverse osmosis transport models can be divided into three
Although it is possible to find in the literature steady-state models main groups: (i) irreversible thermodynamics models ([18]), (ii)
and some dynamic models obtained by parameter identification, porous models (such as the finely-porous, preferential sorption-
there are no reports about lumped parameter dynamic models for
capillary flow, and surface force-pore flow models) and (iii) non-
control purposes obtained by application of physical laws. Such
porous or homogeneous membrane models (such as all models
models are useful not only to design model-based control systems
but also for the implementation of fault tolerant systems based on based on diffusion). Transport models centers on the top thin skin
fault detection and isolation (FDI) methods, as well as to analyze of asymmetric membranes or the top thin skin layer of composite
transient characteristics of the plant. In this paper, models from the membranes since these determine fluxes and selectivities of most
literature are shortly analyzed and a simple lumped parameter model membranes ([5]). There exists a fundamental difference between
for control purposes, which is derived from physical laws, is proposed. the assumptions about if the membrane is porous or not. Models
Moreover, a block-oriented library for MATLAB/SIMULINKTM is based on homogeneity assume that transport takes place between
presented, so that different plant configurations can be implemented the interstitial spaces of the polymer chains or polymer modules,
as block diagram to simulate the system and to test control algorithms. typically by means of diffusion. Porous models assume that
transport occurs through the pores that run the length of the
I. INTRODUCTION membrane barrier layer; so that transport occurs not only by
diffusion but also by convection through the pores. However, it
R EVERSE Osmosis (RO) is a process used to demineralize
water, to clean brackish water or to desalt seawater. The
process consists in recovering water from a saline solution
is still not clear if RO membranes are porous or not and it consti-
tutes a current point of discussion. The models mentioned before
have the common characteristic that all assume equilibrium in
pressurized by pumping it into a closed vessel to a point grater than the membrane. A simplified model for an industrial plant, which
the osmotic pressure of the solution. Thus, the solution is pressed is based on diffusion equations, has also been reported in [3].
against a membrane so that it is separated from the solutes (the This model integrates partial differential equations to obtain
dissolved material). The portion of water that passes through the the final equations those are finally also static.
membrane reducing strongly the solute concentration is called Several linear dynamic models for RO plants can also be
permeate. The remaining water (brine) is discharged with a found in the literature (see for example [2], [4], [14], [15]). All
high salt concentration. these models are obtained by system identification using real data
so that they are only valid for a particular plant, working at the
In the last years, significant advances in the membrane
selected operating point. Based on such models advanced control
technology have allowed an essential improvement in the algorithms have been used: For example in [4], a CMPC (Con-
filtering quality and simultaneously a general reduction of strained Model Predictive Control) algorithm is presented and in
costs. Hence, RO plants have today lower energy consump- [15], a DMC approach (Dynamic Matrix Control) is compared
tion, investment cost, space requirements and maintenance with standard PID control. An overview about control of desali-
than other desalination processes. nation plants, which includes RO processes, is presented in [1].
On the other hand, RO desalination plants are energy intensive The main contribution of the present paper is a lumped parame-
and require fine tuned components. Therefore, a good control ter dynamic model based on physical laws and its implementation
design is necessary to maintain water-production costs at as blockset for Simulink. This model can be linearized at
acceptable level and to elevate the plant availability, particularly different operating points such that algorithms for advanced
in regions with high water scarcity. In this case, it is crucial to model-based linear control systems can be implemented.
Authorized licensed use limited to: ENSTINET. Downloaded on December 05,2021 at 11:10:16 UTC from IEEE Xplore. Restrictions apply.
FrA19.3
Because of the high pressure, pure water permeates through the models show very different dynamic characteristics, as it can
the membranes and the salty water becomes very concen- be observed from the transfer function of Table I and open loop
trated (brine). The water product flows directly from the step responses of Fig. 3.
permeators into a storage tank, where pH is adjusted and
Transmembrane
minerals are added. The brine (at high pressure) is discharged Pressure Permeate flux
G11(s)
(Fig. 1) or sent to an energy recovery device (see for example
[20]).
pretreatment postreatment G21(s)
chemical RO Unit chemical
P
additives C additives Permeate
Inlet pH + conductivity
Feed water pH
Minerals G22(s)
+
Filter F
fresh
HP pump
pH water Fig. 2. Block diagram for the RO plants obtained from the literature.
brine TABLE I
LP pump
DIFFERENT MODELS FOR RO PLANTS FROM THE LITERATURE
Fig. 1. Schematic diagram of a RO plants TRANSFER FUNCTIONS
Pretreatment is important in RO plants because the membrane Inputs
Reference Pressure control valve Acid inlet valve
surfaces has to be continuously clean. The water should be
−0.155 (0.375 s +1)
pretreated, so that salt precipitation or microbial growth on the F G11(s) = G12 (s) = 0
membranes is avoided. Hence, the pretreatment consists of (0.22 s +1)(2.51 s +1)
fine filtration and the addition of chemicals to inhibit pre- 2.48 e−120 s 0.45e −120s
Outputs
The high-pressure pump supplies the pressure needed to [4]
−4.74
enable the water to pass through the membrane and to have P G31(s) = G32 ( s) = 0
the salts rejected. This pressure range is from 15 to 25 bars (1.45 s +1)
for brackish water and from 54 to 80 bars for seawater. −0.077
pH G41 (s) = 0 G42 (s) =
The membrane assembly consists of a pressure vessel and (21.2 s +1)
several membrane units that permit the feed water to be P pH
pressurized against the membrane. The membrane must be 0.0045 (0.104 s +1)
F G11(s) = G12 (s) = 0
able to resist the entire pressure drop across it. The semi- 0.012 s2 + s +1
Outputs
[15]
−0.51(0.35 s +1) −57(0.32 s +1)
III. EXPERIMENTAL MODELS C G21(s) = G22 (s) =
0.213 s2 + 0.7 s +1 0.6 s2 +1.8 s +1
Several linear models can be found in the literature (see
Notation: F: permeate flow rate, C: permeate conductivity, P: transmen-
for example [2], [4], [14], [15]). In [2], the model was obtained brane pressure, pH: inlet pH
by fitting second order transfer functions of the form y1: Permeate flow rate in m3/d
0.04
K (τi s +1)
Gi (s) = i
(1) 0.02
τ s + 2ςτ
2 2
i i i s +1 0
model was also used in [15]. The model of [14] was obtained -0.04
plants, it is not possible to verify if models are incorrect. However, Fig. 3. Step responses for the models of Table I.
4855
Authorized licensed use limited to: ENSTINET. Downloaded on December 05,2021 at 11:10:16 UTC from IEEE Xplore. Restrictions apply.
FrA19.3
seawater
Notice that the step response of the model given in [4] is not Fs, Ts, Ps, Cs
Mixer
analyzed because of the different definition of inputs and outputs.
Moreover, the structure of Fig. 2 only matches the first two HP Pump
outputs for the model of [4]. The model of [15] has a negative Fs, Ts, Ps, Cs RO unit
gain for transfer functions G21 and G22, i.e. there is an inverse
Fp, Tp, Pp, Cp
relationship between inputs and outputs. For the model [14] Brine Permeate
subsystem Permeate
the gains G21 and G22 are also negative but the transfer functions subsystem
membrane
exhibit non-minimal phase behaviour. Because of these differences, Fb, Tb, Pb, Cb
it is important to have a physical model in order be able to Fp, Tp, Pp, Cp
Energy
analyze the reason of such diverse behaviors. recovery
Permeate
brine
The modeling process will be divided in several steps. The o the RO membrane has a homogeneous, nonporous
first one consists in decomposing the plant in subsystems. Hence, surface layer
assumptions will be carried out and variables will be defined. o solute and solvent dissolve in this layer and then each
After that, the subsequent subsections are devoted to present the diffuses across it
model equations for each subsystem. o solute and solvent diffusion is uncoupled and due to its
own chemical potential gradient across the membrane
A. Plant decomposition
o gradients are the result of concentration and pressure
At the present time, there is no suggestion in the literature
differences across the membrane.
about how to decompose RO plants. According to its characteris-
tics, a spatio-temporal decomposition has advantages in the plant • Inside the RO module, flow rates are assumed to be laminar.
organization. This technique consists in a combination between • RO unit is completely fed with water.
the physical decomposition and the functional one. Thus, a
physical decomposition is used for the main plant and functional Finally, variable names and subindices are summarized in
decomposition for the obtained subsystems. Table II. Parameters are given in Table III.
A basic RO system normally consists of five main subsystems: TABLE II
power generation, pretreatment, pumping and energy recovery, NOTATION AND UNITS
membrane assembly (RO unit) and post-treatment (see Fig. 5). VARIABLES
t time s
Electrical seawater
chemicals
Desalination Unit F mass flow rate kg/s
Unit chemicals
q volumetric flow rate m3/s
Power Pre- Pumping
Desalination Pos- m mass kg
Subsystem
Generation treatment
with recovery Subsystem treatment Q b heat flow rate W
Subsystem Subsystem (RO Units) Subsystem
energy device
T temperature °K
brine fresh water
C salt concentration g/kg
electrical energy supply
p pressure MPa
Fig. 5. Schematic geographical decomposition of a RO plant C *p , b average heat capacity kJ/(kg °K)
Subsystems can also be broken down according to physical, π osmotic pressure MPa
functional or both concepts. Pre and post treatments subsystems Subindices
can be modeled as stirred tank-systems ([16]) and models for f Feed water (seewater)
pumping systems can also be found e.g. in [19]. Therefore, p Permeate
the focus of this work is the model of a simple RO unit. A b Brine
possible general decomposition for a RO unit is shown in Fig. 4. v Valve
m Membrane
It consists of the brine subsystem, the permeate subsystem and
the membrane. C. Modeling of the brine side
Balance equations for mass, momentum, energy and salt
B. Modeling Assumptions
of this subsystem yield
Nowadays, there are two technologies that dominate the
mark of membrane production: hollow-fiber and spiral wound. dmb dt = Ff − Fb − Fm (2)
Although modeling principles and general balance equations are dCb dt = 1 mb ⎡⎣ Ff (C f − Cb ) − Fm (C p − Cb ) ⎤⎦ (3)
independent of the membrane technology, some equations will
change from a technology to the other one in order to accom- dpb dt = 1 Lb ⎡⎣ pb − p f − φ F f ⎤⎦ (4)
modate geometrical aspects and filtering procedures. Moreover, with
several assumptions are taken for the model building:
Lb = 32 ηb lb2 (d i2 Pmax ) and φ = 128 ηb lb / πρ b d i4 (5)
• It is assumed that the membrane responses to an ideal and finally
pore-free solution-diffusion model. Thus, following conditions
are implicitly considered in this model: dTb dt = 1 mb ⎡⎣ Ff (T f − Tb ) − Fm (Tp − Tb ) − Q b C *p ,b ⎤⎦ , (6)
4856
Authorized licensed use limited to: ENSTINET. Downloaded on December 05,2021 at 11:10:16 UTC from IEEE Xplore. Restrictions apply.
FrA19.3
where C*p,b is the average value of Cp,b between Tf and Tb. An 128η d 0 A
H = . (19)
equation for the specific heat capacity C *p ,b is given in [7]. d i4
The brine flow rate can be obtained from V. SOFTWARE FOR MODELING AND SIMULATION
Fb = Kbo pb − pbo . (7) A simulation environment and a library in the form of
Kbo and pbo are defined by the external valve. The heat flow rate Simulink blockset are currently being developed in order to
model and simulate desalination plants. Only blocks from the
Qb(t) represents heat losses in the system and is calculated from
standard Simulink library are used, so that the minimal configu-
Q b = α t Ab ΔTx , (8) ration of Matlab/Simulink is necessary to model such processes.
Examples of these blockset are presented in Fig. 6.
where Ab is the heat transfer area, αt is the overall heat transfer
coefficient and ΔTxb is given by Feed water stream Feed water stream
(Ff,Pf,Tf,Cf) (Ff,Pf,Tf,Cf)
at permeate at permeate
line exit Brine stream line exit Brine stream
D. Modeling the permeate side (Fb,Pb,Tb,Cb) (Fb,Pb,Tb,Cb)
IC IC
For the permeate side, following equations are obtained:
RO Unit (hollow-fiber module) RO Unit (spiral wound module)
dm p dt = Fm − Fp = 0 , (10)
T
dC p dt = Fm (Cm − C p ) m p ,
Library
(11) Valve
travel (x)
F(x)
Valve
opening
of seawater Prop
Library
of heat exchangers
C properties
dp p dt = (1/ Lm )[− pb + φm qm ] , (12)
Valves Seawater Properties Heat exchangers
dT p dt = 1 m p ⎡⎣ Fm (Tm − Tp ) − Q p C *
p, p
⎤⎦ , (13) 1
Feed water stream
Feed water stream
(Ff,Pf,Tf,Cf)
Membrane
stream
that can also be calculated from [7]. The heat flow rate Q p (t) (Fm,Pm,Tm,Cm)
Brine
Subsystem
Brine
stream 2
Pressure (Fb,Pb,Tb,Cb) Brine stream
IC
Q = α A ΔT ,
permeat line exit
(14) 3 Pressure
Brine Subsystem at permeate line exit
p t p xp Permeate
Permeate 1
stream
where Ap is the heat transfer area, αt is the overall heat transfer
Subsystem
(Fp,Pp,Tp,Cp) Permeate stream
Feed water Membrane (Fp, Pp, Tp, Cp)
Membrane Membrane stream
stream stream
coefficient and ΔTxp is given by
Subsystem (Fm,Pm,Tm,Cm)
(Ff,Pf,Tf,Cf) (Fm,Pm,Tm,Cm)
Brine Subsystem
Ta has been defined above. [Ff ]
[Pf ]
1
Feed Water
[Tf ] Kbo Fb
E. Modeling the membrane Stream [Cf ] sqrt
3
1
For the membrane, special equations are used from the [Fm]
[Pm]
Momentum Balance Pob
Brine stream
(Fb, Pb, Tb, Cb)
l/Lb pb
solution-diffusion model for hollow-fiber modules, namely
2 [Pf ] 1
[Tm]
Permeate [ICp] xo s
Stream [Cm]
Mass Balance Phi
tanh( Hl ) mb
Fm = Aρ p ( Δp − Δπ * ) + B [C f − C p ]
[Ff ] 1
H + H 2 ls tanh( Hl )
[Fm]
4
[ICp] Energy Balance
[ICT] [mb]
IC [Ff ]
[ICc] 1 Tb
with [Tf ]
[ICT] xos
[Tb]
T − T0 T −T0 [Tf ] Tf
aT bT [Ff]
A = A0 e T
, B = B0 e T
. (17)
[Tb] Tb
Qp [Tp]
[Tm] Tp
Ta Ta
Specif ic Heat T
Heat Excahge Capacity
where A0 and B0 are membrane coefficients at the reference
Cp
of Seaw ater C
Cp of Seaw ater
Concentration Balance
temperature T0 = 291K. aT and bT are dimensionless empirical [Ff] [mb] Cb
[Cb]
[Ff]
[Cm]
π i* = 0.5 RC i Ti [2 − PRR ] /[1 − PRR ] . (18) Fig. 6. Selected blocks from the Desalination blockset
R is the universal gas constant and PRR (Permeate Recovery The block diagram for the system simulation has been designed
Rate) is computed as the quotient between permeate and brine following exactly the plant decomposition scheme. Moreover, it
volumetric flow rates in percent (100% qb/qp). Finally, l is is color coded (blue for permeate, green for feed water and brine
the fiber length, ls the adhered fiber length and H is obtained from and orange for the membrane). Notice that subsystems are also
4857
Authorized licensed use limited to: ENSTINET. Downloaded on December 05,2021 at 11:10:16 UTC from IEEE Xplore. Restrictions apply.
FrA19.3
4858
Authorized licensed use limited to: ENSTINET. Downloaded on December 05,2021 at 11:10:16 UTC from IEEE Xplore. Restrictions apply.
FrA19.3
Finally, valve opening and feed water temperature are changed was implemented and a RO plant using the parameters of a RO
dynamically as it shows Fig. 10. Hence, it is possible to module of the company DuPont was simulated. Simulation
observe that the permeate flow rate increases almost linearly while results show that the model behaves qualitatively similar to a
permeate salt concentration decreases exponentially with the feed real plant. This is a promising result if it is taken into account
water temperature. A similar effect produces a continuously that the model was simplified to obtain lumped parameters
closing of the discharge valve between 100 and 50%. by integrating the spatial equations only in one direction.
Permeate flow rate qp [m3/h] However, this model can be considered a first step in the
0.7
α: Valve opening: 50 % modeling of RO plants and it will be refined in the near
0.6
Tf: Feed water temp. Tf = 23°C future, for example to include the dynamic behavior of pH-
0.5
Tf = 21°C value. Moreover, parameter estimation has to be carried out
0.4
in order to adjust the model to a real plant.
0.3 Tf = 19°C
4859
Authorized licensed use limited to: ENSTINET. Downloaded on December 05,2021 at 11:10:16 UTC from IEEE Xplore. Restrictions apply.