Alsadaie Mujtaba JPC
Alsadaie Mujtaba JPC
Alsadaie Mujtaba JPC
This work is made available online in accordance with publisher policies. Please refer to the
repository record for this item and our Policy Document available from the repository home
page for further information.
To see the final version of this work please visit the publisher’s website. Access to the
published online version may require a subscription.
Citation: Alsadaie SM and Mujtaba IM (2016) Generic Model Control (GMC) in Multistage Flash
(MSF) Desalination. Journal of Process Control. 44: 92-105.
Copyright statement: © 2016 Elsevier Ltd. Reproduced in accordance with the publisher's self-
archiving policy. This manuscript version is made available under the CC-BY-NC-ND 4.0 license.
Generic Model Control (GMC) in Multistage Flash (MSF)
Desalination
Salih M. Alsadaie and Iqbal M. Mujtaba*
E-mail: I.M.Mujtaba@bradford.ac.uk
ABSTRACT
Keywords: MSF Desalination, Dynamic model, Optimization, PID Control, GMC Control
1. Introduction
The lack of potable water in most of countries of the world is one of everyday challenge;
therefore highest priority should be given in the efforts toward solving this global problem.
Due to the limitations of the underground water, low rainfall, rapid economic growth, etc., in
several regions of the world, MSF is used to convert saltwater by evaporation to potable
water or make-up water that is free of impurities. MSF accounts for approximately 22% of
commercial desalination worldwide (Miller et al. 2014). However, MSF has higher yields
than other desalination methods, such as multi-effect desalination (MED) and Reverse
Osmosis (RO) separation (Darwish, 2014). Moreover, the capability of coupling the MSF
plant to a power generation plant as the heat source makes the process increasingly important
for future drinking water and power production.
However, desalination plants are large and complex. They are also energy-and cost intensive
and above all, crucial to life support in several regions of the world. Consequently,
desalination plant must meet high standards of performance, including optimality, cost
1
effectiveness, reliability, and safety. Many of these criteria can be satisfied by improved
design and control.
Most industrial plants are non-linear in nature; the complexity of their non-linearity varies
according to the physical function of each process. MSF desalination is a highly complex
nonlinear process (Ismail, 1998; Ali et al, 1999 and Lior et al, 2012); however, its non-
linearity is represented in some operation conditions such as limitation on the brine
temperature at the brine-heater outlet. Furthermore, the need for continuous monitoring of
liquid levels in the flashing chambers is necessary to avoid loss of efficiency due to blow-
through or loss of boiling due to flooding in the flash chambers. Therefore, an efficient and
accurate control system in the plant to maintain the operation at optimum conditions is
required
Most of the MSF plants are currently operated under conventional PID controller due to its
simplicity and well recognition by the industry (Al-Gobaisi et al. 1991). However, PID
controller is linear and cannot efficiently control highly sophisticated systems which contain
nonlinear variables. Moreover, the tuning of the PID parameters is being the main concern by
engineers due to the time consuming and inefficiency. Despite of a lot of work to improve the
tuning of PID parameters, there is no adequate single method to obtain optimum values for
these parameters. Nevertheless, due to the change of the operating conditions of the MSF, the
fixed tuning of PID controller for one condition would not be optimal due to change in sea
water temperature seasonally and variable water demand and thus, new optimum values of
PID parameters are required which can be consider as time consuming (Al-Gobaisi et al.
1994). The availability of powerful computer tools opened the way to implement the
advanced process control (APC) strategies.
A number of researches have been conducted in the past decades to implement APC strategy
in the MSF desalination process. Maniar and Deshpande (1996) applied Constrained Model
Predictive Control (CMPC) for MSF process. The manipulated variables for the controllers
were calculated by solving an optimization problem with respect to the operating constraints.
Though the authors obtained reasonable results, the nonlinearity of MSF process cannot be
controlled well using linear CMPC. Later, Ali et al. (1999) utilized a reduced model to
implement a robust control of MSF process using Nonlinear Model Predicted Control
(NLMPC) which was able to drive the plant to its steady state with less computational time.
Dewei et al. (2012) proposed a Cascaded Quadratic Dynamic Matrix Control (QDMC) as one
of the MPC strategies for a Reverse Osmosis (RO) desalination process. Compared to PID
control, the results revealed that the QDMC outperform the traditional PID control. Although
it was developed four decades ago, Fuzzy Logic Controller (FLC) remains to have a lot of
attention due to its ability to control very complex systems (Alatiqi et al,. 1999). Jamshidi et
al. (1996) designed and implemented fuzzy controllers for MSF process to control TBT. A
genetic algorithm was applied to fuzzy control of a brine heater unit in MSF plant. The
simulation results of the controlled TBT showed a significant improvement in convergence to
the desired set point and reducing oscillations and overshoot. Ismail et al. (1996) combined a
set of fuzzy rules to introduce a controller that look like Proportional-Integral-Derivative
(PID-like FLC). The controller was then introduced to MSF process to control TBT. Ismail
2
(1998) studied the capability of Fuzzy Model Reference Learning Control for the TBT. In
comparison with the conventional PID and direct fuzzy logic, the results showed the
outperformance of the learning system over the other two types. For the same purpose of
controlling TBT, Olafsson et al. (1999) designed and applied simple fuzzy control to brine
heater in MSF process. In most of their study cases, the results showed that FLC can perform
better or equally well as the conventional PID controller.
Neural Network System (NNs) is another technique used as APC to handle complex and
nonlinear process. Ali et al. (2015) provided an excellent review on the application of NN
based control (state observers) in many engineering systems. After successful implementation
of NN techniques as optimization control strategy for seawater-desalination solar-powered
membrane distillation unit by Porrazzo et al. (2013), Tayyebi and Alishiri (2014) proposed
nonlinear inverse model control strategy based on neural network for MSF desalination plant.
Using three-layer feed forward neural network, three loops were designed for controlling the
TBT, the brine level in the last stage and salinity.
Generic Model Control (GMC) is a well-known advanced control technique that has been
used widely in the past and was developed by Lee and Sullivan (1988) as a result of the
intense desire to develop a model that can handle nonlinear processes like most of the
chemical processes. Cott and Macchietto (1989) applied GMC strategy as controller to track
the reactor temperature set point. Vega et al. (1995) applied GMC controller experimentally
and by simulation to a batch cooling unseeded crystallization process to control crystallizer
temperature. Aziz et al. (2000) used GMC to design a controller for a batch reactor to track
the optimal temperature profiles. Ghasem et al. (2003) implemented GMC controller to the
two-phase model of a non-isothermal fluidized bed catalytic reactor to control the
temperature inside the reactor by tracking new set point and handling the disturbance. In
tracking the optimal temperature set point profile of batch reactor, Arpornwichanop et al
(2005) applied GMC algorithm to drive the temperature of the batch reactor to follow the
desired profile. Mujtaba et al. (2006) coupled GMC with NNs as controller to estimate the
heat release due to exothermic reaction. Karacan et al. (2007) proposed multivariable generic
model control (MGMC) to control the top and bottom product temperatures of the packed
distillation column. Ekpo and Mujtaba (2008) used GMC controller in batch polymerisation
of methyl methacrylate to track the set point optimal temperature profile with neural
networks as an online heat release estimator for the system. Kamesh et al. (2014) used GMC
to track a set point of reactor temperature of an industrial multiproduct semi-batch
polymerization reactor.
The aforementioned publications used GMC algorithm to control the temperature in their
systems. However, GMC is used widely to control other type of variables such as pH,
concentration and purity. For instant, Sousa et al. (2004) proposed GMC-fuzzy algorithm for
the pH control of the enzymatic hydrolysis of cheese whey proteins. Kathel and Jana (2010)
implemented GMC algorithm in two different forms, namely real and ideal GMC, to control a
high-purity reactive batch distillation column. Du et al. (2013) applied GMC algorithm in
sewage processing to control the concentration of dissolved oxygen based on the hybrid
3
model. Fu and Liu (2015) implemented GMC controller in heat integrated air separation
column to control the purity of the Nitrogen and Oxygen products.
From the foregoing, the GMC has been proved to be simple, robust and strategic in
controlling various types of process parameters, hence, the decision to use it to control the
TBT and BL in MSF plant.
There is no known use of GMC as a controller strategy in MSF plants. In this work therefore,
the GMC control strategy is designed and introduced to the MSF process to control and track
the set points of the two most important variables in the MSF plant; namely the output
temperature of the brine heater (TBT) and the Brine Level (BL) in the last stage. To do so, an
optimization problem is solved first to obtain different values for TBT and BL set points for
four different seasons throughout the year. For a comparison purpose, the PID and PI are
used to control the TBT and BL respectively. Note, all the past work on the control of MSF
process restricted to one particular season (for a single seawater temperature). Also, they
were restricted to track a set point change without simultaneously disturb any other systems’
parameters. We have relaxed both of these in this work.
2. Process description
4
Cw
Recovery Section Rejection Section
T2 T1 Ws
Ts
D
LC
TC BD
T4
T3 T5
Brine F
Heater
Rec
3. Mathematical model
To achieve the objectives of this study, a detailed dynamic model of MSF process is required.
The dynamic model is developed and implemented using gPROMS model builder. The
model is based on Reddy et al. (1995) and Alfulaij et al. (2011). The actual data used to
validate this model were obtained from Alasfour and Abdulrahim (2009) of Azzour
desalination plant. The Azzour desalination plant is located in Kuwait; it has 24 stages with
capacity of 6 MGD per unit and has eight units with total output of 48 MGD (Al-Shayji et al.,
2005). The model has been already run dynamically and in steady state, and results were
examined against actual data by Alsadaie and Mujtaba (2014).
Stage Model
Salt balance
𝑑𝑋𝐵
𝜌𝐵 𝐴𝑆 𝐿𝐵 = 𝐵𝑖𝑛 × (𝑋𝐵𝑖𝑛 − 𝑋𝐵𝑜𝑢𝑡 ) + 𝑉𝐵 𝑋𝐵𝑜𝑢𝑡 (2)
𝑑𝑡
5
ℎ = 𝐶𝑝 × (∆𝑇) (6)
𝐶𝑝 = 𝑓(𝑋𝐵 , 𝑇) (7)
Total distillate
Last Stage, N
6
Bout in equation (19) is BD in Figure 1.
𝑅𝑒𝑐 = 𝐵0 = 𝑊𝑅 (20)
Salt balance
𝑑𝑋𝐵
𝜌𝐵 𝐴𝑆 𝐿𝐵 = 𝐵𝑖𝑛 × (𝑋𝐵𝑖𝑛 − 𝑋𝐵𝑜𝑢𝑡 ) + 𝐹 × (𝑋𝑓 − 𝑋𝐵𝑜𝑢𝑡 ) (21)
𝑑𝑡
The material and energy balances for distillate and vapour are similar to those for single
stage. In rejection stages, WR is replace by WS and XR is replaced by XS.
4. Optimization problem
The wide difference in seawater temperature during the day (also between summer and
winter seasons) has great impact on TBT and BL, consequently, product rate and plant
performance are affected. The seawater temperature depends on the locality and the time of
the year and it can be varies between 15 oC and 35 oC (Hawaidi and Mujtaba, 2010). Darwish
et al, (1996) reported that it can be as low as 10 oC in Kuwait. At low temperature, its mass
flow rate has to be reduced to achieve reasonable flashing brine temperature in the bottom
stages. However, the decrease in the cooling seawater flow rate can result in a decrease in its
velocity as low as lower than the acceptable minimum (about 1.5 m/s) (Darwish et al, 1996).
For this reason, most MSF plants operate in summer and winter mode, when the set point of
the intake sea water temperature varies between 25 oC in the winter mode and 32 oC in the
summer mode (Alatiqi et al. 1999).
For fixed operating conditions, the MSF plants produce more fresh water in winter (low sea
water temperature) than in summer. However, this production pattern goes counter to the
demand of fresh water (Hawaidi and Mujtaba, 2011). Tanvir and Mujtaba (2008) minimised
the operating cost by optimizing the number of stages based on seasonal variation of the sea
water temperature. For fixed fresh water production and TBT, Hawaidi and Mujtaba (2010)
studied the effect of sea water temperature on the operating cost of the MSF process. Hawaidi
and Mujtaba (2011) conducted an optimization study to demonstrate the optimum design and
operation of MSF process to meet the variable demand of fresh water through the day and the
year at fixed TBT.
For control purpose, an optimization study is conducted to obtained different optimum values
for TBT and BL based on four different seasons. Based on seawater temperature profile
presented by Hawaidi and Mujtaba (2010), a four different values of the sea water
temperature are considered; 20 oC, 28 oC, 32 oC, and 24 oC for winter, spring, summer and
autumn respectively (Figure 2). To obtain different values of the TBT, a fixed number of
stages and fixed fresh water product are considered. Moreover, to obtain different values of
7
BL, more constrains are introduced to maintain the brine level in all stages at reasonable
level, and thus optimal values for BL are obtained for each season.
40
Average Sea Water Temperature (oC)
36
32 32 oC
28 28 oC
24 24 oC
20 20 oC
16
12
8
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Time (Month)
Sea water temperature Set point change
Figure 2: Variations of sea water temperature and set point profiles for four seasons (Hawaidi and
Mujtaba, 2010)
For the given design configurations; fixed number of stages, fixed fresh water demand, fixed
steam temperature and four values of the intake sea water temperature to determine the
optimum TBT, BL, Rec and WS by minimizing the Total Seasonal Operating Cost (TSOC).
OP Min TSOC
TBT, BL, Rec,, Ws
(0.3 𝑚) ≤ 𝐵𝐿 ≤ (0.8 𝑚)
𝑡𝑜𝑛 𝑡𝑜𝑛
(11500 ) ≤ 𝑅𝑒𝑐 ≤ (17200 )
ℎ𝑟 ℎ𝑟
𝑡𝑜𝑛 𝑡𝑜𝑛
(7600 ) ≤ 𝑊𝑆 ≤ (13600 )
ℎ𝑟 ℎ𝑟
8
Where Dend is the total capacity of the plant and D*end is the fixed water demand (=1296
ton/hr). The boundary values of Rec and WS are chosen based on the minimum and maximum
allowable values of the water velocity in the condenser tubes between 1.5 and 2.3 m/s (Helal,
2003)
The objective function equation (TSOC) is obtained from Hawaidi and Mujtaba (2011) and
defined total annual cost (TAC) as
Where CPC is the annualized capital cost, STC is the storage tank cost and TOC is the total
operating cost. Since the CPC and STC are function of the plant configuration and constant
for all seasons, then the only variable cost here is the TOC. Hawaidi and Mujtaba (2011)
defined the TOC as following:
Where
More details on the calculations of TOC can be found in Hawaidi and Mujtaba (2010, 2011).
5. Controller strategy
For safety purpose, most MSF plants have many control loops to maintain steady state and
overcome the instability caused by the start-up of the plant or failure in one of the plant
components. Maniar and Deshpande (1996) and Ismail (1998) mentioned nine controlled
variables with nine corresponding manipulated variables as the main process variables to be
controlled. Al-Gobaisi et al. (1994) mentioned that most existing MSF plants could be
controlled by 4 to 6 primary loops. However, in these studies two main control loops were
TBT control loop and BL control loop as without these two loops the plant cannot be
controlled at all. In this study we also implemented GMC control in these two control loops.
1. Top brine temperature (TBT). The temperature of the recirculation brine after it is
heated by the low pressure steam in the brine heater. It plays an important role in
describing the performance of MSF and has direct effects on the distillate production
9
and the levels in each flash chamber. It can be used to control the whole plant in
addition to load control. This means for each plant production, there is a certain top
brine temperature which depends on the seawater inlet temperature
2. Last stage brine level (BL): The brine levels in the flash stages are quickly affected by
the steam supply temperature or flow rate (Husain et al., 1994). Brine levels in all
stages should be high enough to seal the interstage orifices and prevent blow-through.
However, the high BL increases the thermodynamic non-equilibrium losses and
should be low enough to ensure less equilibration losses. An adjustable level
controller is required with high sensitivity over the permissible range of BL. This
controller is one of the most important control loops in the MSF plant since the level
in all stages is controlled by adjusting the BL in the last stage (Darwish et al., 1996).
The PID controller is widely used and recommended for a variety of problems. It can be used
for many industrial systems. The controller parameters can be tuned by using trial and error
methods, or any of the classical tuning techniques such as Zeigler Nicholas. For many
process control problems, good results can be achieved by tuning PI, or PID using
conventional methods, which rely on the knowledge and skill of the control engineer
However, due to the change in conditions of the MSF plant during its operation, tuning PID
parameters are always considered as time consuming and it is a quite challenging.
Where kc is the proportional gain, τi is the integral time, τD derivative time constant, e is the
error (controller input), and C is the controller command (controller output).
As mentioned above, different methods can be used to tune the PID controller parameters.
The most common method is the integral performance criterion. In this work, an
optimization based method is used to minimize the Integral Absolute Error (IAE), the Integral
Time Absolute Error (ITAE) and Integral Square Error (ISE) and the PID parameters (kc, τi,
τD) are optimized to give minimum error. Since initial values of PID parameters are required
to conduct the optimization problem, Ziegler-Nichol’s method is used to obtain the initial
values for PID parameters.
subject to:
10
−100 ≤ kc ≤ 100
0.0 ≤ i ≤ 100
0.0 ≤ D ≤ 100
The results of the optimization problem for TBT and BL loops are presented in Table 2.
Figure 3a and Figure 3b show the performance of both controllers, TBT and BL respectively,
using the three types of optimization criterion (ISE, IAE and ITAE). The optimum values
from best method are used later in the control comparison. It is to be mentioned that two
optimization functions are used here. One is to optimize the parameters of the TBT controller
loop and another is to optimize the parameters of the BL controller loop.
Since its development by Lee and Sullivan in 1988, there has been growing interest in the use
of GMC, which has been demonstrated to have certain robustness for a wide range of process
nonlinearity against model mismatches. GMC is relatively easy to implement and does not
require linearizing the nonlinear process (Aziz et al., 2000).
Where y is the measured variable and ysp is the desired value of the control variable.
Similarly to PI controller, the first expression of the above equation k1 (ysp-y) is required to
bring the process from a large distance towards steady state, but some offset would exist. The
second expression k2 ʃ(ysp-y)dt however, is required to eliminate the offset of the controller.
The values of k1 and k2 are tuning parameters to obtain the desired response. More details of
the model can be found in Lee and Sullivan (1988).
In the brine heater of MSF process, the dynamic model equation relating the TBT as
controller variable to the steam flow rate (Ms) as a manipulated variable can be written as;:
11
Solving for the manipulated variable, Ms, the following equation can be obtained.
Ms gives the amount of steam flow rate required to control the outlet temperature of the brine
heater.
Similarly, the above procedure can be followed to implement the GMC method to control the
brine level in the last stage. First, process model equation relating the brine level, LB, as
controller variable to the brine flow rate leaving the last stage (Bout) as manipulated variable
must be defined. Equation (36) is the material balance equation in the last stage, and can be
used here to calculate the change of the brine level, LB, in the last stage.
Where Bin is the brine flow rate leaving the previous stage, F is the makeup flowrate fed to
the last stage, Bout is the blow down flow rate leaving the last stage, VB is the vapour leaving
the brine pool, and Rec is the recycle brine flow rate. To solve for the control, Equation (36)
must be equalised to Equation (32) and substituting LB for y and LB_sp for ysp.
Solving for the manipulated variable, Bout, Equation (38) can be obtained.
Bout gives the amount of blow down flow rate required to maintain the BL in the last stage at
the desired level.
Lee and Sullivan (1988) provided a figure that outlines the relation between two variables, ξ
and τ. Tuning GMC can be obtained by choosing a better combinations of ξ and τ. The
choices should be reasonable and require understanding of the system’s natural dynamic
response. By choosing reasonable values of ξ and τ, the two tuning parameters k1 and k2 are
obtained using Equations (39 and 40).
2𝜉⁄
𝑘1 = 𝜏 (39)
k2 1 (40)
2
12
It is important to mention that different values of k1 and k2 are obtained for different control
loops. More details of the procedure in choosing ξ and τ can be found in Lee and Sullivan
(1988).
Simulations with optimization of the MSF process for four different seasons, optimization of
PID controller parameters, TBT and BL controls were carried out using gPROMS builder
model. First, the MSF process was optimized at fixed plant capacity and four different values
of sea water temperature by minimizing the TSOC. For the sake of stability, other variables
such as (Rec) and the intake sea water flow rate (WS) were relaxed to fluctuate for limited
values. Since the steam is coming from different source, its temperature is fixed and only the
steam flow rate valve is varied to achieve the optimum TBT. The results of the optimization
are shown in Table 1. The table also includes the optimum brine recycle and intake sea water
flow rate at fixed capacity for four different seasons. Therefore, the operator has to change
these values to their next values after every season. It should be mentioned that the optimum
values for TBT and BL for four seasons are developed for control purpose and cannot be
relied on to make accurate performance. More parameters must be considered to draw final
design evaluation.
The main objective of this study is to evaluate the performance of GMC controller comparing
to conventional PID controller by tracking the set points change of TBT and BL respectively.
The PID controllers are introduced to the model and their parameters are tuned (Table 2).
13
(a) (b)
Figure 3: Step response of the optimally tuned PID parameters (a) TBT loop (b) BL loop.
To select the best technique that used to minimize the error and thus giving optimum values
of PID parameters, the results presented in Table 2 are plotted in Figure 3a and 3b
respectively (Only for Spring operation conditions). It is to be mentioned here, the optimum
values obtained by optimization techniques were very aggressive in some cases. As it can be
seen from Figure 3a, the values obtained for TBT loop using ISE and IAE criteria are very
aggressive and take large time to settle down while these from ITAE, though have
overshooting, seem to be close to optimum and less aggressive. For, BL loop, however, the
values obtained using ITAE seem to be less aggressive and giving very smooth curve. Thus
the optimum values of PID parameters obtained by ITAE criteria are considered to be our
choice for both loops. This choice applies for all seasons.
For tuning GMC parameters for TBT loop, Cott and Macchietto (1989) recommended a value
of 10 for ξ to eliminate the overshoots. However, Lee and Sullivan (1988) mentioned that the
selection of GMC parameters depends on the system’s natural dynamic response. In this
work, the value of 10 for the ξ that gives less overshoots is selected. τ is calculated using the
graphical method proposed by Lee and Sullivan (1988) which gives 16 sec for TBT loop and
8 sec for the BL loop.
For each controller loop, three case studies were performed to examine the performance of
each type of controller in the set points tracking, disturbance and constraint handling.
Figure 4 presents the control performance of the PID and GMC controller for tracking the set
point change of TBT based on different seasons. For each season’s data, the model run for 40
seconds to reach steady state before changing the new set of data for the next season. For
reader interpretation convenience, the results of process variables and manipulated variables
were plotted together in one figure (Figure 4).
14
Figure 4: Tracking the TBT set points for four different seasons using GMC & PID controllers
In all cases, GMC controller was performing smoothly and reach the set point in less time.
The PID controller, on the other hand, expressed oscillatory response more than GMC before
returning to the set point while the GMC controller did not reveal any sluggish response and
move smooth towards the new set point and provide better performance over PID in tracking
the set point. Similar behaviour can be observed for the manipulated variables (steam flow
rate). For PID, the steam flowrate looks unstable in attempt to bring the process variable
(TBT) back to the set points for all seasons. However, the steam flow rate behaviour for the
GMC controller was smooth and stable while controlling the process variable.
Case 2: Disturbance
Disturbance normally occurs in MSF plants due to the pumps or valves failure. In order to
examine the capability of the controller in handling the disturbance, a change in the brine
recycle flow rate was introduced in this case at a regular interval of 50 s by increasing its
value 6%, decreased 14% and then increased by 8%. The process was assumed that it runs in
the autumn season when the disturbance occurred. The recycle flow rate was chosen as the
disturbance because it affects the TBT and BL at the same time. Figure 5 shows the
performance of both controllers in handling the disturbance. As it can be seen, the GMC
controller acts vary fast and provide better performance in returning the temperature to steady
state. Also as expected, a perfect GMC (with no modelling error) should not have significant
change in the PV when disturbances enter the system. However, the PID controller exhibits
some oscillatory response and couldn’t reach the set point fast and takes larger time to reach
steady state. The manipulated variables reacted simultaneously as their process variables.
When increasing the recycle brine flow rate by 6%, the steam flow rates increased to provide
enough heat to keep the TBT constant. Similarly when the recycle flowrates decreased by
14%, the steam flow rates dropped to maintain constant TBT. For PID controller, the
behaviour of the steam flow rate follows the same behaviour of the process variable with
some oscillatory response while the steam flow rate using GMC controller behaves smoothly
and fast to keep in the TBT constant.
15
90.4 60
Controller Output
90.3 90.0
90.2 40
Process Variable TBT (oC)
89.8 50
90.1
Controller Output
90.0 30
89.6
89.9 40
89.8 20
460 480 500 520 540 560 580 600
89.7
Time (Sec) 30
89.6
The availability of steam depends on external source and thus it is limited to a certain
amount. Here, similar to the first case, the set points of the TBT was changed based on the
four different seasons however, the steam flow rate was assumed to be limited and hit the
lower and higher limits to bring the controlled variable to its set point. As it can be seen in
Figure 6, the set point was raised in spring season and thus more steam than required was
needed to raise the TBT to its new set point. Thus, the steam hit the constraint of 100% for
short time resulting in delay of the PID controller to reach the set point compared to the first
case. The same results can be seen when the set point was further raised to 94.09 oC in
summer. This is due to the reason that when the set point was increased, the controller sent
signal to the steam valve to fully open. However, due to the lack of available steam PID
controller struggle to bring the process back to its steady state. In autumn, when the set point
was changed to 90.06 oC, the steam flow rate was constrained by 0% and thus again the
controller took large time in attempt to bring the TBT to its new set point. In comparison with
the GMC controller, it seemed that GMC controller performs similarly in handling the
constraints because the availability of the steam that control the process and thus both
controllers behave similarly and slowly. Figure 6 shows that the steam hit the constraint of
100% and 0% for the same time as it was shown for PID. However, when the available steam
is adequate for the appropriate temperature, the GMC controller performs faster and smoothly
and exhibit less oscillatory or sluggish response compared to PID controller when
experienced large overshooting in particular when the set point was further increased in
summer period.
16
Figure 6: Tracking the TBT set points using GMC and PID controllers with constraints.
Although, there was no large difference in the BL set points for different seasons, the
difference was quite reasonable to examine the controller’s performance. Figure 7 shows that
the set points were changed based on four seasons. The PID and GMC controllers were
implemented to track the new set points. In all intervals (Season interval), the GMC
controller over performed the PID controller and reached the set point faster. The PID
showed slight sluggish response and took some time to reach the set point. When the set
points were increased from winter to spring and again from spring to summer, the GMC
controller reached the set point at the same time with PID, however, while PID continues
slugging, the GMC remains constant and kept the BL stable. The reason of the both controller
crossed the set point at the same time is that the tuning of GMC parameters were tuning
based on the time that PID cross the set point as it was mentioned before. The behaviour of
the manipulated variables (Blow down) were identical to the performance of the process
variables (brine level). The manipulated variable of GMC was smooth while the manipulated
variable of the PID experienced slight overshooting to bring the level of the stage back to its
set point.
17
0.48
Controller Output
800
Process Variable BL(m)
Controller Output
0.42 0.40 0.38 600 900
0.46 400
0.40 0.38 0.36 800
Process Variable BL(m)
0.44 400
0.36 0.34 700 200
Controller Output
0.38
0.42 600 0 10
40 50 60 70 80 90 100 110 12 20 30
0.36 0.34 200
500 Time (Sec) 0
0.40 0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160
0.34
Time (Sec) 0 400
0 0.3820
10 30 40 50 60 70 80 90 100 110 120 130 Set
140 Point
150 (left)
160 PID PV (left)
PID MV (right) 300 GMC MV (right)
Time (Sec)
0.36 Set Point (left) PID PV (left) 200 GMC PV (left)
PID MV (right) GMC MV (right)
Set Point (left) PID PV (left) GMC PV (left) 100
0.34
PID MV (right) GMC MV (right)
0
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160
Time (Sec)
Figure 7: Tracking the BL set points for four different seasons using GMC & PID controllers
Case 5: Disturbance
Similar to the case 2 in TBT loop, the same disturbance of the brine recycle was introduced to
the process and the behaviour of both controllers were observed. Both loops (TBT and BL)
work simultaneously and any set points change or disturbance affects both loops at the same
time. Thus, the same step change in the brine recycle was introduced to BL loop. As it can be
seen in Figure 8, the GMC controller over performed the PID controller in bringing the
process back to its steady state very fast. While PID controller showed some overshoots
before reaching0.437
the set point, the GMC work perfectly in handling the disturbance and look
like no change occurred to the process. 650
0.436
Process variable BL (m)
0.435
Controller Output
0.437
660 600
0.434
0.436 640
0.433
550
Process variable BL (m)
620
0.435
Controller Output
0.432 600
0.434 580
0.431 500
0.433 560
460 480 500 520 540 560 580
540 600
0.432 Time (Sec)
520
0.431 500
PID PV (left) GMC PV (left)
PID MV (right) GMC MV (right) 480
Figure 8: Handling the disturbance to control the TBT using GMC & PID controllers
18
Case 6: Constraint handling
The valve of the blowdown flow rate is assumed to open and close to limited positions to
study the performance of the controllers under the constraints. Thus, the set points of the BL
was changed based on the four different seasons and the valve position was assumed to be
limited and reached the lower and higher limits to bring the controlled variable to its set
point. As it can be seen from Figure 9, though both controllers worked well in controlling the
process, GMC looks better in bringing the process to steady state fast at the start-up of the
plant. Both valves were hitting 0% and 100% to bring the process steady, however, the valve
controlled by GMC started to be stable first to maintain the BL constant. When the set point
of BL was increased from 0.36 m in winter to 0.429 m in spring and raised again in summer
up to 0.742 m, the valve of the blowdown were closed completely. Due to the constraints, the
valve position reached its lower limit in attempt to increase the BL to its set points. Here,
both controllers. PID and GMC look behave similarly in controlling the process well. Again,
in the final season (autumn) when the set point was changed from 0.472m to 0.433 m, the
valve position reached the higher limit for few seconds to bring down the BL to its new set
point. GMC controller worked better here in autumn (last interval) in reaching the set point
fast. Regarding to the manipulated variable behaviour, the manipulated variables for both
controllers behave similarly as it was in Figure 7, however, PID manipulated variable react
few seconds behind GMC manipulated variable. Despite their close performance in
controlling the BL, the GMC has more stability over PID controller and could easily
accommodate all the process changes.
Figure 9: Tracking the BL set points using GMC and PID controllers with constraints.
19
7. Conclusions
The work presented in this paper focused on the implementation of GMC control in MSF
desalination plants. Since most MSF plants are operated under conventional PID control, the
proposed GMC control can improve the control process in MSF plants. To carry out the
control process, detailed dynamic model of MSF process was developed and implemented
using gPROMS model builder. Two controller loops, namely TBT and BL, were designed to
investigate the performance of GMC controller. For each loop, three cases were carried out;
tracking the set points without constraints, tracking the set points with constraints and
handling the disturbance. Different values for TBT and BL set points were selected for four
seasons in the year based on optimisation process. The disadvantage of PID controller is its
linearity and time consumption in tuning its parameters. However, GMC is easy to use and
can handle nonlinear systems. Also, the tuning of the GMC parameters is very simple.
In comparison to the PID controller, the results indicated that the GMC is powerful and
robust tool in controlling MSF plants and outperformed the PID in all cases. In handling the
disturbance for example, the GMC control the process easily without showing any oscillatory
or overshoots. In TBT loop, although both controllers reached the set points nearly at the
same time, the GMC reached the set points with less overshoots and more smoothly.
However, in BL loop, GMC controller appeared to be fast and more robust in controlling the
level with and without the present of the constraints and over performed the PID controller.
In the BL loop, it is important to mention that both controllers were not just used to track the
set point but to overcome the change of other variables such as the recycle flow rate, intake
sea water temperature and intake sea water flow rate. Here, the GMC controller looks even
better in tracking the set points. While the PID controller exhibits some oscillatory, the GMC
controller reaches the set point fast and remain constant for the whole period. This behaviour
was monitored for all four seasons.
Most importantly, it is the tuned procedure of the two controllers. While PID parameters took
large time to be tuned and waste of time, the GMC parameters were tuned fast and easily
based on known plant speed and graphical method.
Although most of the applications of the GMC algorithm were in controlling the temperature,
here, the GMC was used successfully to control the level of the brine in MSF as well as the
temperature of the brine heater and has revealed its controllability to handle nonlinear system
under different set points change with and without constraints.
Nomenclature
AS - Stage area (m2)
20
A - Heat transfer surface area of a stage (m2)
Ah - Heat transfer surface area of the brine heater (m2)
AD - Distillate tray area (m2)
B0 - Flashing brine mass flow rate leaving brine heater (kg/s)
Bin - Brine inlet flow rate to a stage (kg/s)
Bout - Brine outlet flow rate from a stage (kg/s)
BPE - Boiling point elevation (oC)
Cp - Specific heat at constant pressure (kJ/kg oC)
DC - Total condensate flow in a stage (kg/s)
Din - Distillate flow rate to a stage (kg/s)
Dout - Distillate flow rate to a stage (kg/s)
Dtotal - Total distillate product flow rate 9kg/s)
hBo - Enthalpy of flashing brine leaving the brine heater (kJ/kg)
hBin - Enthalpy of flashing brine entering a stage (kJ/kg)
hBout - Enthalpy of flashing brine leaving a stage (kJ/kg)
hDC - Enthalpy of condensate distillate around tube bundle (kJ/kg)
hDin - Enthalpy of distillate entering a stage (kJ/kg)
hDout - Enthalpy of distillate leaving a stage (kJ/kg)
hFin - Enthalpy of cooling water entering the brine heater (kJ/kg)
hNCGs - Enthalpy of NCGs leaving the flashing brine in a stage (kJ/kg)
hVB - Enthalpy of vapour below demister in a stage (kJ/kg)
hV - Enthalpy of vapour around cooling tubes (kJ/kg)
hVin - Enthalpy of vapour entering from previous stage (kJ/kg)
hVout - Enthalpy of vapour leaving a stage (kJ/kg)
LB - Height of the brine (m)
LD - Height of the distillate (m)
Ls - Height of the stage (m)
MV - Manipulated variable
NEA - Non-equilibrium allowance (oC)
21
PV - Process variable
TBT - Top brine temperature (oC)
TBo - Temperature of flashing brine leaving the brine heater (oC)
TBin - Temperature of flashing brine entering a stage (oC)
TB - Temperature of flashing brine leaving a stage (oC)
TF1 - Temperature of cooling brine entering the brine heater (oC)
TFin - Temperature of cooling brine entering a stage (oC)
TFout - Temperature of cooling brine leaving a stage (oC)
Tsteam - Steam temperature (oC)
TV - Temperature of flashed vapour in the vapour space (oC)
Uh - Overall heat transfer coefficient in the brine heater (kW/m2 oC)
Vbrine - Volume of the cooling water inside the brine heater (m3)
VB - Vapour release flow rate from brine in a stage (kg/s)
Vin - Vapour flow rate entering a stage (kg/s)
Vout - Vapour flow rate leaving a stage (next stage or vent) (kg/s)
Vtube - Volume of the cooling water inside the tube bundle (m3)
WR - Cooling brine flow in the heat recovery stages (kg/s)
WS - Cooling seawater flow in the heat rejection stages (kg/s)
Wsteam - Steam flow rate (kg/s)
XBin - Salt concentration in the brine entering a stage (ppm)
XB0 - Salt concentration in the brine leaving brine heater (ppm)
XBout - Salt concentration in the brine leaving a stage (ppm)
XR - Salt concentration in the cooling brine in the recovery section (ppm)
XS - Salt concentration in the cooling brine in the rejection section (ppm)
Greek letters
22
𝜌𝐷 - Distillate density (kg/m3)
References
[1] AL-FULAIJ, H., CIPOLLINA, A., ETTOUNEY, H. & BOGLE, D. 2011. Simulation
of stability and dynamics of multistage flash desalination. Desalination, 281, 404-412.
[2] AL-GOBAISI, D. M. K., HASSAN, A., RAO, G. P., SATTAR, A., WOLDAI, A. &
BORSANI, R. 1994. Towards improved automation for desalination processes, Part I:
Advanced control. Desalination, 97, 469-506.
[3] AL-SHAYJI, K. A., AL-WADYEI, S. & ELKAMEL, A. 2005. Modelling and
optimization of a multistage flash desalination process. Engineering optimization, 37,
591-607.
[4] ALASFOUR, F. N. & ABDULRAHIM, H. K. 2009. Rigorous steady state modeling
of MSF-BR desalination system. Desalination and Water Treatment, 1, 259-276.
[5] ALATIQI, I., ETTOUNEY, H. & EL-DESSOUKY, H. 1999. Process control in water
desalination industry: an overview. Desalination, 126, 15-32.
[6] ALGOBAISI, D. M. K., BARAKZAI, A. S. & ELNASHAR, A. M. 1991. AN
OVERVIEW OF MODERN CONTROL STRATEGIES FOR OPTIMIZING
THERMAL DESALINATION PLANTS. Desalination, 84, 3-43.
[7] ALI, E., ALHUMAIZI, K. & AJBAR, A. 1999. Model reduction and robust control of
multi-stage flash (MSF) desalination plants. Desalination, 121, 65-85.
[8] ALI, J. M., HOANG, N. H., HUSSAIN, M. A. & DOCHAIN, D. 2015. Review and
classification of recent observers applied in chemical process systems. Computers &
Chemical Engineering, 76, 27-41.
[9] ALSADAIE, S. M. & MUJTABA, I. M. 2014. Modelling and Simulation of MSF
Desalination Plant: The Effect of Venting System Design for Non-Condensable
Gases. CHEMICAL ENGINEERING, 39.
[10] ARPORNWICHANOP, A., KITTISUPAKORN, P. & MUJTABA, I. M. 2005. On-
line dynamic optimization and control strategy for improving the performance of
batch reactors. Chemical Engineering and Processing: Process Intensification, 44,
101-114.
[11] AZIZ, N., HUSSAIN, M. A. & MUJTABA, I. M. 2000. Performance of different
types of controllers in tracking optimal temperature profiles in batch reactors.
Computers & Chemical Engineering, 24, 1069-1075.
[12] COTT, B. J. & MACCHIETTO, S. 1989. TEMPERATURE CONTROL OF
EXOTHERMIC BATCH REACTORS USING GENERIC MODEL CONTROL.
Industrial & Engineering Chemistry Research, 28, 1177-1184.
23
[13] DARWISH, M. A. 2013. Thermal desalination in GCC and possible development.
Desalination and Water Treatment, 52, 27-47.
[14] DU, X., YU, P. & LUO, D. 2013. Generic Model Control of Dissolved Oxygen
Concentration in Sewage System. Journal of Computational Information Systems, 9,
6.
[15] EKPO, E. E. & MUJTABA, I. M. 2008. Evaluation of neural networks-based
controllers in batch polymerisation of methyl methacrylate. Neurocomputing, 71,
1401-1412.
[16] EL-DESSOUKY, H. T. & ETTOUNEY, H. M. 2002. Fundamentals of salt water
desalination, Elsevier.
[17] FU, Y. & LIU, X. 2015. Nonlinear dynamic behaviors and control based on
simulation of high-purity heat integrated air separation column. ISA transactions, 55,
145-153.
[18] GHASEM, N. M., HUSSAIN, M. A. & MUJTABA, I. M. 2003. Control of the
Industrial Gas Phase Polyethylene Fluidized Bed Reactor Using Generic Model
Control (GMC). Advances in Petrochemicals and Polymers in the New Millennium.
Thailand.
[19] HAWAIDI, E. A. M. & MUJTABA, I. M. 2010. Simulation and optimization of MSF
desalination process for fixed freshwater demand: Impact of brine heater fouling.
Chemical Engineering Journal, 165, 545-553.
[20] HAWAIDI, E. A. M. & MUJTABA, I. M. 2011. Meeting Variable Freshwater
Demand by Flexible Design and Operation of the Multistage Flash Desalination
Process. Industrial & Engineering Chemistry Research, 50, 10604-10614.
[21] HELAL, A. M. 2003. Uprating of Umm Al Nar East 4–6 MSF desalination plants.
Desalination, 159, 43-60.
[22] HUSAIN, A., WOLDAI, A., ALRADIF, A., KESOU, A., BORSANI, R., SULTAN,
H. & DESHPANDEY, P. B. 1994. MODELING AND SIMULATION OF A
MULTISTAGE FLASH (MSF) DESALINATION PLANT. Desalination, 97, 555-
586.
[23] ISMAIL, A. 1998. Fuzzy model reference learning control of multi-stage flash
desalination plants. Desalination, 116, 157-164.
[24] ISMAIL, A., ABUKHOUSA, E. & IEEE 1996. Fuzzy TBT control of multi-stage
flash desalination plants, New York, I E E E.
[25] JAMSHIDI, M., AKBARZADEHT, M. R. & KUMBLA, K. 1996. Design and
implementation of fuzzy controllers for complex systems - Case study: A water
desalination plant, 64849 Monterrey N L, Instituto Tecnologico Y Estudios
Superiores Monterry.
[26] KAMESH, R., REDDY, P. S. & RANI, K. Y. 2014. Comparative Study of Different
Cascade Control Configurations for a Multiproduct Semibatch Polymerization
Reactor. Industrial & Engineering Chemistry Research, 53, 14735-14754.
[27] KARACAN, S., HAPOĞLU, H. & ALPBAZ, M. 2007. Multivariable system
identification and generic model control of a laboratory scale packed distillation
column. Applied Thermal Engineering, 27, 1017-1028.
24
[28] KATHEL, P. & JANA, A. K. 2010. Dynamic simulation and nonlinear control of a
rigorous batch reactive distillation. ISA Trans, 49, 130-7.
[29] LEE, P. L. & SULLIVAN, G. R. 1988. GENERIC MODEL CONTROL (GMC).
Computers & Chemical Engineering, 12, 573-580.
[30] LI, D., YANG, N., NIU, R., QIU, H. & XI, Y. 2012. FPGA based QDMC control for
reverse-osmosis water desalination system. Desalination, 285, 83-90.
[31] LIOR, N., EL‐NASHAR, A. & SOMMARIVA, C. 2012. Advanced Instrumentation,
Measurement, Control, and Automation (IMCA) in Multistage Flash (MSF) and
Reverse‐Osmosis (RO) Water Desalination. Advances in Water Desalination, 453-
658.
[32] MANIAR, V. M. & DESHPANDE, P. B. 1996. Advanced controls for multi-stage
flash (MSF) desalination plant optimization. Journal of Process Control, 6, 49-66.
[33] MILLER, S., SHEMER, H. & SEMIAT, R. 2014. Energy and environmental issues in
desalination. Desalination.
[34] MUJTABA, I. M., AZIZ, N. & HUSSAIN, M. A. 2006. Neural Network Based
Modelling and Control in Batch Reactor. Chemical Engineering Research and
Design, 84, 635-644.
[35] OLAFSSON, T. F., JAMSHIDI, M., TITLI, A. & AL-GOBAISI, D. 1999. Fuzzy
Control of Brine Heater Process in Water Desalination Plants. Intelligent Automation
& Soft Computing, 5, 111-128.
[36] PORRAZZO, R., CIPOLLINA, A., GALLUZZO, M. & MICALE, G. 2013. A neural
network-based optimizing control system for a seawater-desalination solar-powered
membrane distillation unit. Computers & Chemical Engineering, 54, 79-96.
[37] REDDY, K. V., HUSAIN, A., WOLDAI, A. & AL-GOPAISI, D. M. K. Dynamic
modelling of the MSF desalination process. 1995. 227-242.
[38] SOLIMAN, M. A. 1981. A MATHEMATICAL-MODEL FOR MULTISTAGE
FLASH DESALINATION PLANTS. JOURNAL OF ENGINEERING SCIENCES, 7,
143-150.
[39] SOUSA, R., LOPES, G. P., PINTO, G. A., ALMEIDA, P. I. F. & GIORDANO, R. C.
2004. GMC-fuzzy control of pH during enzymatic hydrolysis of cheese whey
proteins. Computers & Chemical Engineering, 28, 1661-1672.
[40] TANVIR, M. S. & MUJTABA, I. M. 2008. Optimisation of design and operation of
MSF desalination process using MINLP technique in gPROMS. Desalination, 222,
419-430.
[41] TAYYEBI, S. & ALISHIRI, M. 2014. The control of MSF desalination plants based
on inverse model control by neural network. Desalination, 333, 92-100.
[42] VEGA, A., DIEZ, F. & ALVAREZ, J. M. 1995. Programmed cooling control of a
batch crystallizer. Computers & chemical engineering, 19, 471-476.
25
26