Identification and Control of a Laboratory Distillation Column
23
Identification and Control of a Laboratory Distillation Column Martin Klaučo Martin Jelemenský, Richard Valo, Miroslav Fikar Slovak University of Technology in Bratislava, Slovakia February 7, 2014 I A M SSKI 2014 February 7, 2014 1 / 12
Transcript of Identification and Control of a Laboratory Distillation Column
Identification and Control of a Laboratory Distillation
Column
Martin KlaučoMartin Jelemenský, Richard Valo, Miroslav Fikar
Slovak University of Technology in Bratislava, Slovakia
February 7, 2014
IAM
SSKI 2014 February 7, 2014 1 / 12
Distillation
Separation process based ondifferent volatility of substances
Mixture of methanol and water
Obtain distillate of givenconcentration
Temperature control
SSKI 2014 February 7, 2014 2 / 12
Distillation
Separation process based ondifferent volatility of substances
Mixture of methanol and water
Obtain distillate of givenconcentration
Temperature control
0 0.2 0.4 0.6 0.8 160
65
70
75
80
85
90
95
100
105
x , y [−]
T[◦
C]
xy
SSKI 2014 February 7, 2014 2 / 12
Distillation Column
reboiler
pre-heater
valvemanual
feed
condenser
distillateaccumulator
reflux valve distillate
SSKI 2014 February 7, 2014 3 / 12
Distillation Column
reboiler
pre-heater
valvemanual
feed
condenser
distillateaccumulator
reflux valve distillate
PV: Temperature
MV: Reflux ratio
DV: feed temperature
SSKI 2014 February 7, 2014 3 / 12
Control Design
Identification
State estimator design
Model Predictive Controller design
Controller and estimator tuning
SSKI 2014 February 7, 2014 4 / 12
Identification
Perform step responses
0 2000 4000 6000 8000 10000 12000 14000 16000
0
0.2
0.4
0.6
0.8
1
Time [s]
R[-
]
0 2000 4000 6000 8000 10000 12000 14000 16000
40
60
80
100
120
Time [s]
T[◦
C]
SSKI 2014 February 7, 2014 5 / 12
Identification
Perform step responses
Butterworth low pass filter(ωn = 0.005rad/s)
0 2000 4000 6000 8000 10000 12000 14000 16000
0
0.2
0.4
0.6
0.8
1
Time [s]
R[-
]
0 2000 4000 6000 8000 10000 12000 14000 16000
40
60
80
100
120
Time [s]
T[◦
C]
SSKI 2014 February 7, 2014 5 / 12
Identification
Perform step responses
Butterworth low pass filter(ωn = 0.005rad/s)
Matlab identification toolbox
xk+1 = Axk + B(uk − us)
y = Cxk + y s
0 2000 4000 6000 8000 10000 12000 14000 16000
0
0.2
0.4
0.6
0.8
1
Time [s]
R[-
]
0 2000 4000 6000 8000 10000 12000 14000 16000
40
60
80
100
120
Time [s]
T[◦
C]
SSKI 2014 February 7, 2014 5 / 12
Model Validation
1000 1500 2000 2500 3000 3500 4000 4500 5000
55
65
75
85
95
105
1000 5000
0
0.2
0.4
0.6
0.8
1
Time [s]
T[◦
C]
R[-
]
SSKI 2014 February 7, 2014 6 / 12
Model Validation
1000 1500 2000 2500 3000 3500 4000 4500 5000
55
65
75
85
95
105
1000 5000
0
0.2
0.4
0.6
0.8
1
Time [s]
T[◦
C]
R[-
]
SSKI 2014 February 7, 2014 6 / 12
Model Validation
1000 1500 2000 2500 3000 3500 4000 4500 5000
55
65
75
85
95
105
1000 5000
0
0.2
0.4
0.6
0.8
1
Time [s]
T[◦
C]
R[-
]
SSKI 2014 February 7, 2014 6 / 12
State Estimator and Disturbance Modelling
Stationary Kalman filter:
[
x
d
]
k|k
=
[
x
d
]
k|k−1
+ L(
ym,k − yk|k−1
)
[
x
d
]
k|k+1
=
[
A E
0 I
] [
x
d
]
k|k
+
[
B
0
]
uk|k
yk|k =[
C F]
[
x
d
]
k|k
+ Duk|k
SSKI 2014 February 7, 2014 7 / 12
Model Predictive Control
minN
∑
k=1
||rk − yk ||2Q +N
∑
k=1
||∆uk ||2S
SSKI 2014 February 7, 2014 8 / 12
Model Predictive Control
minN
∑
k=1
||rk − yk ||2Q +N
∑
k=1
||∆uk ||2S
s.t. xk+1 = Axk + Buk + Edk
yk = Cxk + Duk + F dk
SSKI 2014 February 7, 2014 8 / 12
Model Predictive Control
minN
∑
k=1
||rk − yk ||2Q +N
∑
k=1
||∆uk ||2S
s.t. xk+1 = Axk + Buk + Edk
yk = Cxk + Duk + F dk
umin ≤ uk ≤ umax
SSKI 2014 February 7, 2014 8 / 12
Scheme of MPC Closed Loop
MPCDistillationColumn
Disturbances
Estimator
ymur
x , d
SSKI 2014 February 7, 2014 9 / 12
Simulation Results
0 500 1000 1500 20006466687072747678
Time [s]
T[◦
C]
0 500 1000 1500 2000
0
0.2
0.4
0.6
0.8
1
Time [s]
R[-
]
SSKI 2014 February 7, 2014 10 / 12
Simulation Results - Temperature Profile
0 500 1000 1500 20006466687072747678
Time [s]
T[◦
C]
0 500 1000 1500 2000
0
0.2
0.4
0.6
0.8
1
Time [s]
R[-
]
SSKI 2014 February 7, 2014 10 / 12
Simulation Results - Concentration Profile
0 500 1000 1500 20000.7
0.8
0.9
1
Time [s]
y[−
]
0 500 1000 1500 2000
0
0.2
0.4
0.6
0.8
1
Time [s]
R[-
]
SSKI 2014 February 7, 2014 10 / 12
Experimental Results
1000 1500 2000 2500 3000 3500 4000 4500
65
70
75
80
Time [s]
T[◦
C]
1000 1500 2000 2500 3000 3500 4000 4500
0
0.5
1
Time [s]
R[-
]
1000 1500 2000 2500 3000 3500 4000 450020
40
60
80
Time [s]
Tfe
ed[◦
C]
SSKI 2014 February 7, 2014 11 / 12
Conclusions
What has been done:
Identification of laboratory distillation column
Implementation of MPC based on state space model
SSKI 2014 February 7, 2014 12 / 12
Conclusions
What has been done:
Identification of laboratory distillation column
Implementation of MPC based on state space model
What is being done:
Tuning of MPC and estimators for controlling laboratory device
SSKI 2014 February 7, 2014 12 / 12
