You are on page 1of 4

variables can be selected.

Selected variables are shown in


Condition to execute MPCPro algorithm
is that number of controlled and constrained parameters
must be less than or equal to manipulated parameter. [13]

6 PROCESS MODELING EQUATIONS Table 1


MPC application assumes a linear process model. For
single-input, single-output (SISO) process prediction, the SP SP Low High
Variables
equations are in the form: Low High Limit Limit
Bottom Side
Controlled 0 300 - -
X k = A X k−1 +b ∆u k + f W k ----------------------(2) Variable
Temp.
Upper Side Temp. 0 500 - -
y 0=C X k -------------------------- (3) Constraint Reboiler Valve
- - 0 95
Variable Position
T Reflux Valve
- - 0 95
Where X k−1=[ y 0 , y 1 , … y i , … , y p−1 ] is the vector Position
Manipulate Sidestream flow - - 35 80
of process output prediction at a time k – 1, 0, 1, 2, … p – d Variable Reboiler Flow - - 45 90
1 steps ahead. Matrix A is the shift operator defined for a Pressure - - 15 25
self-regulating process as Reflux Flow - - 50 90
Disturbanc Reflux
- - - -
T e Temperature
A X k−1=[ y 1 , y 2 , … y i , … , y p−1 ] -----------
Controlled Variable Constraint Variable
(4)
Time Upper Side Bottom Side Reboiler Reflux
T
b=[b0 , b1 , ….. , bi , … , b p−1 ] is the vector of p step Temp. Temp. Valve Valve
Position Position
response coefficients. ∆ u k =uk −uk−1 is the change in 11.45 135.81 226.34 96.07 85.47
p m 12.00 138.54 231.13 91.01 87.43
the process input/controller output. W k = y − y is the 12.15 138.66 221.11 100 80.75
process output measurement minus the model output (the 12.30 141.58 235.66 100 80.45
difference between the process and the model that results
from the noise, unmeasured disturbances, and model After putting all the relevant values for the Controlled,
inaccuracy). f is the p dimension filter vector with unity Manipulated, Constraints and Disturbances, launch the
default values. Matrix C is the operator for selecting the DeltaV PredictPro application for the commission and
current model output defined as y 0=C X k +1 [ 8] testing. Parameters required for model generation are
Time to Steady State and Step Size. Here values for above
parameters are 240 sec and 5% respectively. Step testing
offers interaction between process variables during MPC
7 MPC IMPLEMENTATION ON THE PART
implementation. DeltaV provides Pseudo Random Binary
OF DISTILLATION COLUMN Signal (PRBS) signal to MV for a test to generate
Here bottom product composition is considered for
corresponding CV response. As shown in Figure 2.
control. Temperature is the indirect measurement
Table 2 and Table 3
parameter for the composition [3]. Bottom temperature is
gives us corresponding values of Controlled and
main control variable for the bottom composition. To
Constraint variables for change in Manipulated variable.
manipulate this control variable, used manipulated
From this we can obtain gain and can verify the
variables are Reboiler flow ,Side Stream flow, Reflux
flow and Pressure. Reflux Valve Position and Re-boiler
Table 2
Valve position are constraint variables. By the following Manipulated Variable
way MPC has been implemented using DeltaV.
Time Reflux Reboile Pressur Sidestream
r e
7.1 MODEL IDENTIFICATION
Model plays an important role in Model Predictive 11.45 57 72 20 70.99
Control (MPC) because it facilitates approximately 12.00 67 62 17.50 75
12.15 47 82 22.50 65
correct time response of change in process output
12.30 47 82 22.50 75
(controlled or constrained variable) for a change in
process input (manipulated variable). Looking at the
Table 3
difficulties of Distillation Process, model identification
assists in the development of plant model that can
adequately characterize temperature and flow control.
model which will be generated in future. By
Multivariable model tries to show relationships between
selecting the AUTO generation button, model is
input-output pair that may have prominent association
generated. The validity of the step response can
with each other. By using MPCPro block of DeltaV
Figure 2: PRBS test for Model Identification
be obtained from the background color of the coefficients of the ARX model; and d is dead time in
individual step response as shown in the Figure scans. As shown in Figure 4, the FIR response provides
3. Red colored step response is the most valuable information on the process gain and response.
significant for the associated Manipulated This alternative method essentially involves comparison
process input.Step test provides Steady State of ARX (Auto Regressive model with External input)
Process Gain, Time Constant and Dead Time. with FIR. The squared error is shown for all Control and
Matrix is obtained in the Transfer Function form Constraint parameters. Here squared error is 0.21. Refer
as below: Figure 5. In addition, for selected parameters, the
Bottom side temperature calculated and actual value can be plotted for the original

[ Reboiler Valve Position


Upper SideTemperature
Reflux Valve Position
=
] time or for the timeframe selected by the green bar in the
overview. This involves comparison of calculated output
and actual output. This is shown in the Figure 6.

0.19 e−22 s 0.4 e−10 s 0.211 e−16 s

[ ]
0 0 Figure 6 : Comparison between actual and predicted value of CV
18 s +1 15.2 s +1 16.61 s+1
0.893 e−2 s Reflux
0

−0.20 e−8 s
19.72 s +1
0.98 e−2 s
4.15 s+1
4.15 s+1
0.078 e−20 s
20.109 s +1
0
0 0

0 0

0 0
0

0
[ ] 9 CONTROLLER GENERATION
Reboil
A dynamic matrix is used for developing an MPC
Pressure
controller. A dynamic matrix is built from step responses
Reflux T
to predict
Sidestream
the changes in the process outputs that result
from moves of the manipulated variables over the control
horizon. Dynamic matrix Su as in Equation below,
calculates prediction vector ∆ X k resulting from c future
moves of MV, defined by the vector∆ u (k ).
Model is verified by choosing the Control and Constraint
parameter. The squared error is shown for all Control and ∆ X k =¿ Su ∆ u(k).=
Constraint parameters. In addition, for selected
parameters, the calculated and actual values can be
plotted for the original time or for the timeframe selected
by the green bar in the overview. Figure 5: Performance of Model

8 MODEL VERIFICATION
Alternatively, the FIR response can be used as a guide in
manually editing the step response. FIR identifies pulse
response coefficients, as in below for a SISO process.

p
∆ y k =∑ hi ∆ uk −i ---------------------- (5)
i=1
b0 0 0 0

[ ][ ]
∆u k
where p is prediction horizon, with a typical default value b1 b0 0 0
∆ uk +1
for MPC model 120; ∆ y k is change in the process b2 b1 b0 0
∆ uk +2
output at the time k; ∆ u k−i is change in the process input ⋮ ⋮ ⋮ ⋮ =
at the time k – i; and hi is the pulse response coefficient of ⋮
bi bi−1 bi−2 bi−c+1
Figure 3: Model overview in DeltaV ⋮
⋮ ⋮ ⋮ ⋮
the model[8]. On other hand ARX has fewer coefficients ∆u k+c−1
b p−1 b p−2 b p −3 b p−c
which are defined with higher confidence, provided the
process dead times are known. ∆ y0

[]
v a
∆ y1
y k =∑ ai y k−i+ ∑ bi uk−d −i -------------- (6) ∆ y2
i=1 i=1

Where a, v are autoregressive and moving average ∆ yi
equation orders of ARX; a = 4, v = 4 satisfy most ⋮
applications;a i , bi are moving average and autoregressive ∆ y p−1

Figure 4: Comparison between ARX and FIR Response


Flow) results in change in most relevant controlled
variable i.e. Bottom Side Temperature from 217.620 F
Once the model is accurate, Controller Generation is the
to 218.910 F . This value is noted down. It also deviates
next stage. Condition to execute Controller Generation is
number of Controlled and Constrained Parameters must other controlled variable, Upper Side Temperature from
be less than or equal to Manipulated Parameter. Penalty their Set Point (SP) and constraint variable from its steady
on Move (POM) and Penalty on Error (POE) are two state value. During this procedure, other manipulated
parameters to adjust the robustness of control and speed variables remained to its original position. Note that,
of response. By using Controller Setup, parameters are constraint variable (Reboiler valve Position) crossed its
selected for controller generation. This gives condition higher limit (95%) i.e. 97.49%.
number. Lower condition number gives better control. A. In Local Mode (Without MPC)
Penalty on Move is a parameter that affects robustness.
To make control less aggressive Penalty on Move of that Table 3
parameter increases. The MPC controller minimizes the
squared error of a controlled variable over prediction Manip Controlled Constrai
horizon and the squared error of controller output over ulated Variable nt
control horizon. Variabl Variable
e
2 2 Reboiler Botto Upper Reb Ref
min
∆ MV (k)
{‖Г y
[CV ( k )−R (k )]‖ +‖Г u ∆ MV (k )‖ } Flow m Side
Tempe
side
tempe
oile
r
lux
Val
rature 0rature 0Val
94.9 ve
95.
----------- (7)
70 KPPH 217.62 127.79
F 2%F 43
0 97.4
0 95.
75 KPPH 218.91 128.77
F 9%F 38
94.9 95.
217.64 0127.89
where CV(k) is the controlled output p-step ahead 0
70 KPPH F 1%F 42
prediction vector; R(k) is the p-step ahead reference
trajectory (set point) vector; ΔMV(k) is the c-step ahead
incremental control moves vector; Гy is a diagonal Now, noted value of Bottom Side Temperature in local
penalty matrix on the controlled output error; Гu is a mode is given as Set Point of Bottom Side Temperature
diagonal penalty matrix on the control moves; p is the in MPC mode and results are checked. It is rightly
prediction horizon (number of scans); and c is the control observed that not only controlled variable tracks set point
horizon (number of scans). but also does not affect other CVs (Controlled and
Constraint Variable). To track set point, MV
(Manipulated Variable) utilization is carried out. As
10 MPC CONTROLLER EQUATION constraint (Reboiler Valve Position) is reached to its
higher limit, Reboiler flow (MV) didn’t change
ΔMV(k) = (SuT Г yT Г ySu + Г uT Г u)-1 SuT Г yT Г y Ep(k) significantly (69.90 KPPH to 69.95 KPPH). Instead of
--------- (8) that, sidestream is changed from 69.71 BPD to 71.13
BPD. Again set point is changed from 218.910 F to
Where Su is the p × c process dynamic matrix built from
the step responses of dimension p ×c for a SISO model
217.620 F . Then, Reboiler flow (mv) is changed
significantly, because now, its related constraint (Reboiler
and pn × cm for a MIMO model with m manipulated
Valve Position) is in limit. Like this way, Set point tracks
inputs and n controlled outputs; and E p (k ) is the error Set point tracks as well as constraints are handled Using
vector over prediction horizon. Presenting MPC control MPC as shown in Figure 7.
equation in the form
B. In MPC Mode (With MPC)

Δ MV ( k )=K mpc E p ( k )----------------(9) Table 4


Controlled Variable Manipulated Variable
S.P. of Reboile
Where K mpc =¿(SuT Г yT Г ySu + Г uT Г u)-1 SuT Г yT Bottom Side
Bottom Side
Temperatur r Valve Reboiler
Sidestrea
m
Reflux
Flow Flow
Г y Temperatur
e
e Position flow
0 0 69.90KPP 76.79
217.62 F 217.62 F 94.40%
H
69.71BPD
Bbl/day
69.95KPP 76.90
11 RESULTS 218.910 F 218.910 F 94.45%
H
71.13 BPD
Bbl/day
65.11KPP 74.75
217.620 F 217.620 F 92.48%
H
70.25 BPD
Bbl/day
11.1 Comparison of variables between with MPC and
without MPC (local)
In Local Mode i.e. without MPC, deviation of 5% in
12 CONCLUSION
steady state value of manipulated variable ( Reboiler
Model generated in MPC is of high accuracy, as it is
rightly confirmed from the methods used in Model of Multi-component Packed Distillation Column’,
Verification. Further MPC gives better results over PID in World Applied Sciences Journal 13, 2011.
the application of Interactive Multivariable Control
System. As seen from above results MPC was able to [ 12] Yucai Zhu, Rohit Patwardhan, Stephen B.
Wagner, Jun Jhao, ‘Toward a low cost and high
handle constraints very effectively as in this case Reflux performance MPC: The role of system
Valve Position and Reboiler Valve Position was identification’, Computers and Chemical
constrained between low limit and high limit. Engineering 51 (2013) 124-135.

[ 13] Books on-line, DeltaV DCS Documentation,


Emerson Process Management, USA.
ACKNOWLEDGMENT:
We are very thankful to Emerson Export
Engineering Centre for giving us opportunity. We would
like to thank Mr. Koustubh Palnitkar and Mr. Archis Labhe
for technical assistance. We are very grateful to Mr. Sachin
Soman for moral inspiration while carrying out this project.

REFERENCES:

[ 1] Willian L. Luyben, ‘Process Modelling, Simulation


and Control for Chemical Engineers’, 1990.

[ 2] Warren L. McCabe, Julian C. Smith, Peter


Harriott, ‘Unit Operations of Chemical engineering’,
Mc GRAW-HILL International Editions, Chemical
and Petroleum Engineering Series, Fifth
Edition,1993.

[ 3] Bѐla G. Lipták, Instrument Engineers’ handbook :


Process Control, Third Edition, 1995.

[ 4] B. Wayne Bequette, ‘Process Control, Modelling


and Simulation’, Prentice Hall of India Pvt.
Ltd.2003.

[ 5] S. Joe Qin, Thomas A. Badgwell, ‘A survey


of industrial model predictive control technology’,
Control Engineering Practice 11 (2003) 733-764.

[ 6] Sudhir Panditrao, Sudhir Agashe, Prashant


Shevgaonkar, ‘Model Predictive Control of Pilot
Spray Dryer Unit Designed and implemented for an
Educational Institute ’.

[ 7] Dale E. Seborg, Thomas. F. Edgar, Duncan A.


Mellichamp, ‘Process Dynamics and
Control’,Second Edition, 2004.

[ 8] W.K. Wojsznis (2005), Model Predictive Control


and Optimization.

[ 9] Tri Chandra S.Wibowo, Nordin Saad, and Mohd


Noh Karsiti, ‘System Identification of an Interacting
Series Process for Real-Time Model Predictive
Control’, 2009 American Control Conference Hyatt
Regency Riverfront, St. Louis, MO, USA June 10-
12, 2009.

[ 10] Vu Trieu Minh, Wan Mansor Wan


Muhamad, ‘Model Predictive Control of a
Condensate Distillation Column’, International
Journal of Systems Control (Vol. 1-2010/ Iss.1).

[ 11] Saniye Ay and Suleyman Karacan,


‘Decoupling Constrained Model Predictive Control

You might also like