© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
Dynamics and control of incineration processes
Davide Manca
Lesson 2 of “Dynamics and Control of Chemical Processes” – Master Degree in Chemical Engineering
Table of contents
Implementation and validation of a detailed first-principles dynamic model of the combustion section of an incineration plant
Analysis of a multivariable MPC control system Model reduction
Synthesis of a non-linear MPC control system Identification of an ARX linear model
Synthesis of a linear MPC control system and comparison with
the non-linear counterpart
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
Hot section of an incineration plant
Primary kiln
Economizer Superheater
Boiler
Postcombustion chamber
Grate kiln
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
Boiler and turbine
Fabric filter and exhaust gas
washing
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
DeNOx and Catox
Layout of an incineration plant
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
Primary combustion chamber
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
Primary kiln – Modeling scheme
1
2
2.1
2.2
2.3
2.4 2.5
3.1 3.2 3.3 3.4
4
Homogenous combustion zone
Heterogeneous
combustion zone
i , RIF OUT
i , RIF IN
i , RIF i
,
RIF F F R
dt
dM
Solid Phase
Gas Phase
0
k k OUT , i
rif i , RIF OUT
i ,
k x W
PM W R
NG i
NC
k 1 , 1 ,
OUT i ,
F RIF IN
i ,
F RIF
OUT i ,
W k
IN i ,
W k
i-th grate
NG i 1 ,
Primary kiln – Material bal. on the grates
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
O k H
N m x NO y
x pSO
kHCl nCO
N Cl O S H
C n m p q k y
O i2 i i 2 2
2 2
2,
y x i NO , i
1 3150
1 2500
2
2
100
out i , O G
% i , NO G i
, NO i
, NO
W ) T / exp(
T
W
Combustion reaction in the solid phase
(Bowman, 1975)
2 2
CO 1 O CO
2 1 2 1 2
N O NO
2 2
Combution reactions in the gas phase
Kinetics
, ( , , , , , , )
sc i p i i waste i inerts i
A f d M M
1 1 exp( , )
CG i
i
N
i sc i
i
sc A
A * , ,
2
2
, , *
, ,
,
x i O i
waste i sc i waste
O i
k x
R A PM
Corrective factor with adaptive parameters
) , ,
, ,
( , , ,
, f d W x T
k x i p i i aria i k i Solid granular bed
Kinetic determining step: O
2diffusion
,
2,
O isee L2–18 eq. (8)
Primary kiln – Combustion kinetics
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
x
h
b
waste
m x b h
Grate schematization
F w ast e (t) [kg /s]
0
t
O
t
O
CG
O
N
t 3600
Time [s]
2 0
2 2
1 ( ( ))
( ) exp
2 2
waste
t t
F t m
Δ
Waste pulse schematization
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
11 11.5 12 12.5 13 13.5 14 14.5
0 0.5 1 1.5 2
St eam flo wrat e [t /h ]
Time [h]
Average value
NT
NT
k
k
1y y
S MEDIA
T NT t
Valore istantaneo
11.5 12 12.5 13 13.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Ave rage s tea m f low rate [t /h]
Time [h]
1 Min 5 Min 10 Min
• High : delay time
• Low : high oscillations
average
t
Moving average
Manca D., M. Rovaglio, Ind. Eng. Chem. Res. 2005, 44, 3159-3177
Model equations
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
Manca D., M. Rovaglio, Ind. Eng. Chem. Res. 2005, 44, 3159-3177
Model equations
Manca D., M. Rovaglio, Ind. Eng. Chem. Res. 2005, 44, 3159-3177
Model equations
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
Manca D., M. Rovaglio, Ind. Eng. Chem. Res. 2005, 44, 3159-3177
Model equations
Manca D., M. Rovaglio, Ind. Eng. Chem. Res. 2005, 44, 3159-3177
Model equations
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
Manca D., M. Rovaglio, Ind. Eng. Chem. Res. 2005, 44, 3159-3177
Model equations
Manca D., M. Rovaglio, Ind. Eng. Chem. Res. 2005, 44, 3159-3177
Model equations
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
Manca D., M. Rovaglio, Ind. Eng. Chem. Res. 2005, 44, 3159-3177
Model equations
Manca D., M. Rovaglio, Ind. Eng. Chem. Res. 2005, 44, 3159-3177
Model equations
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
Manca D., M. Rovaglio, Ind. Eng. Chem. Res. 2005, 44, 3159-3177
Model equations
Manca D., M. Rovaglio, Ind. Eng. Chem. Res. 2005, 44, 3159-3177
<
Model equations
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
Material balances
Energy balances
Momentum balances
Total
Primary kinl
29 2 86
Postcombustion
chamber 15 1 26
Heat recovery section
55
10
6 9 5 20
Grand total = 132 DAE system
Model dimensions in terms of DAEs
Number of feed-strokes [1/h] 40
Inlet waste flowrate [kg/h] 4,000
Number of strokes to the first grate [1/h] 23 Number of strokes to the second grate [1/h] 20 Number of strokes to the third grate [1/h] 13 Number of strokes to the fourth grate [1/h] 23
Primary air flowrate [Nm3/h] 12,500
Secondary air flowrate [Nm3/h] 6,500
Some significant input process variables
Outlet smokes temperature from the primary kiln [°C] 1,035 Outlet smokes temperature from the postcombustion chamber [°C] 1,115 Outlet oxygen molar fraction from the postcombustion chamber [%] 8.5 Outlet CO content from the postcombustion chamber [mg/Nm3] 11
Some significant output process variables
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
REAL PLANT
6 8 10 12 14 16
0 0.2 0.4 0.6 0.8 1
Steam flo w rate [ t/h ]
Time [h]
MODEL
Model validation
3400 3600 3800 4000 4200 4400 4600 4800
11 12 13 14 15 16 17
Portata di rifiuto [kg/h]
Tempo [h]
3000 3500 4000 4500 5000 5500 6000 6500
11 12 13 14 15 16 17
Portata di aria secondaria [Nm3/h]
Tempo [h]
9600 9900 10200 10500 10800 11100 11400 11700 12000 12300 12600
11 12 13 14 15 16 17
Portata di aria alle griglie [Nm3/h]
Tempo [h]
10200 10400 10600 10800 11000 11200 11400 11600 11800 12000 12200 12400
11 12 13 14 15 16 17
Potere calorifico del rifiuto [kJ/kg]
Tempo [h]
9 10 11 12 13 14 15 16
11 12 13 14 15 16 17
Portata vapore [t/h]
REAL INPUTS
DYNAMIC SIMULATOR
Model validation
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
9 10 11 12 13 14 15 16
11 12 13 14 15 16 17
S tea m f low rate [t /h]
Time [h]
Experimental data Simulated values
Model Real plant
Model validation
Experimental data Simulated values
940 960 980 1000 1020 1040 1060 1080 1100 1120 1140 1160
11 12 13 14 15 16 17
Primary kiln temperature [°C]
Time [h]
5 6 7 8 9 10 11
11 12 13 14 15 16 17
Time [h]
Postcombustion oxygen molar fraction [%]
960 980 1000 1020 1040 1060 1080 1100 1120
Experimental data Simulated values
Postcombustion temperature [°C]
Experimental data Simulated values
On line model validation
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
600 700 800 900 1000 1100 1200 1300 1400 1500 1600
11 12 13 14 15 16 17
HCl
PCI Waste heat of combustion [kJ/kg]
HCl content [mg/Nm3]
Time [h]
10200 10400 10600 10800 11000 11200 11400 11600 11800 12000 12200 12400
13000 14000 15000 16000 17000 18000 19000
11 12 13 14 15 16 17
Total air flowrate [Nm3/h]
Time [h]
Inferred waste heat of combustion
Experimental trend of the total air flowrate
Inference of heat of combustion and HCl
6 8 10 12 14 16
0 0.2 0.4 0.6 0.8 1
Steam flowrate [t/h]
Time [h]
Dynamic response to different forcing actions
8 10 14 16
Steam flowrate [t/h]
12
3 x N
1 x 3N
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
Multivariable model based
control
Implementation and validation of a detailed first-principles dynamic model of the combustion section of an incineration plant
Analysis of a multivariable MPC control system Model reduction
Synthesis of a non-linear MPC control system Identification of an ARX linear model
Synthesis of a linear MPC control system and comparison with the non-linear counterpart
Table of contents
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
Use of a model for predictive pursposes
y
set+
Model Constraints +
Optimizer
Model Process
+
-
+ - -
d
yˆ
y
yˆ u
u
MPC – Implementation scheme
ˆ( ).... (
min
ˆ 1)c
u k u k h
f
obj
2 1
21
ˆ ( ) ( ) ˆ ( ) ( )
p p
k h k h
obj y y y u u
j k i k
f
e j PF j
u i PF i
Δ
2 2
ˆ ) ˆ ˆ ( , 0 ˆ min
) ˆ ˆ ( , 0 max )
(
MIN MIN MAX
MAX
u u
u i
u u
u i
i u PF
) ˆ ( ) ( )
( ) ;
ˆ (
) ˆ (
) ( )
ˆ ( )
ˆ ( k y k y k
j y
j y
k j
j y e
SET
SET
y
) 1 ˆ (
) 1 ˆ (
) ˆ ( ) ˆ (
u i
i u i
i u u
Optimization problem to be solved:
2 2
ˆ ) ˆ ˆ ( , 0 ˆ min
) ˆ ˆ ( , 0 max )
(
MIN MIN MAX
MAX
y y
y j
y y
y j
j y PF
MPC – Objective function
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
CONTROLLED VARIABLES (CV)
Steam flowrate (set point)
Primary kiln temperature (upper bound)
Postcombustion chamber temperature (upper bound) Postcombustion chamber temperature(lower bound)
Volumetric O 2 fraction in the postcombustion chamber (lower bound) CO content in the postcombustion chamber (upper bound)
MPC – Controlled variables (CV)
MANIPULATED VARIABLES (MV)
Strokes number to the feed grate Primary air flowrate (to the grates) Secondary air flowrate to the kiln
Strokes number to the first moving grate Strokes number to the second moving grate
MPC – Manipulated variables (MV)
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
NON-LINEAR MPC LINEAR MPC
Control model
Detailed model
(DAE)
Linearized identified
model (ARX)
MPC – Control model
11 11.5 12 12.5 13 13.5 14 14.5
0 0.5 1 1.5 2
S tea m f low rate [t /h]
Time [ h ]
Continuous model 132 DAE Continuous model 68 DAE Average value of the discontinuous model
3600
2 0 2 2
( ( ))
1 exp( )
2 2
3600
o CG
o
t N
t waste
CG
t t
F m
N
Averaged flowrate Δ
Same simulated average values but significantly reduced CPU times
Continuous model (132 DAE)
10% disturbance on the waste flowrate
Discontinuous model
(132 DAE) 6.02 s
2.26 s CPU time for a 1
min simulation interval
Simplified continuous
model (68 DAE) 0.54 s
NL MPC – Continuous simplified model
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
14.5
Time [h]
11 11.5 12 12.5 13 13.5 14
0 0.2 0.4 0.6 0.8 1
Steam flowrate [t/h] (CV)
Average value Instantaneous value Set point
3900 4000 4100 4200 4300 4400 4500 4600
0 0.2 0.4 0.6 0.8 1
Waste flowrate [kg/h] (MV)
Time [h]
Time [h]
12600 12750 12900 13050 13200 13350 13500
0 0.2 0.4 0.6 0.8 1
Air flowrate to the grates[Nm3/h] (MV)
5700 5800 5900 6000 6100 6200 6300 6400 6500 6600 6700 6800
0 0.2 0.4 0.6 0.8 1
Secondaryair flowrate [Nm3/h] (MV)
Time [h]
NL MPC – Servo problem
11.5 12 12.5 13 13.5 14 14.5
0 0.5 1 1.5 2 2.5 3
Average steam flowrate [t/h] (CV)
Time [h]
NL MPC PI Set point
3600 3800 4000 4200 4400 4600 4800 5000
0 0.5 1 1.5 2 2.5 3
Waste flowrate [kg/h] (MV)
Time [h]
NL MPC PI
MPC Controller
Faster control action
Reduced oscillations of both the controlled and manipulated variables
NL MPC – Comparison with PI controller
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
3900 4000 4100 4200 4300 4400 4500
0 0.2 0.4 0.6 0.8 1
waste flowrate [kg/h] (MV)
Time [h]
11 11.2 11.4 11.6 11.8 12 12.2 12.4 12.6
0 0.2 0.4 0.6 0.8 1
Steam flowrate [t/h] (CV)
Time [h]
Averaged value Intstantaneous value
Set point
Time [h]
5200 5400 5600 5800 6000 6200 6400 6600
0 0.2 0.4 0.6 0.8 1
Secondary air flowrate [Nm3/h] (MV)
12600 12750 12900 13050 13200 13350 13500 13650 13800 13950 14100
0 0.2 0.4 0.6 0.8 1
Air flowrate to the grates [Nm3/h] (MV)
Time [h]
NL MPC – Regulator problem
) (t u
Plant
) ( t y
ARX (Auto Regressive Model with Exogenous Input )
nb k nb k
k na
k na k
k
k
a y
a y
a y
b u
b u
b u
y
1 1 2 2 1 1 2 2Matrix of NY·NY output parameters at time (k-i)
i
y
k NY output vector at time (k-i)
j
u
k NU input vector at time (k-j)
a
ib
jMatrix of NY·NU input parameters at time (k-j)
na Model order respect to the outputs
nb Model order respect to the inputs
Linear MPC – ARX model
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
0 200 400 600 800 1000 1200 1400 1600 1800 2000 -1
-0.5 0 0.5 1 1.5
Time
Output
Input and output signals
0 200 400 600 800 1000 1200 1400 1600 1800 2000 -0.04
-0.02 0 0.02
Time
Input
1000 1100 1200 1300 1400 1500 1600 1700 1800 1900 2000 -1
-0.5 0 0.5 1 1.5
Time
Measured and simulated model output
PRBS Input (Pseudo Random Binary Sequence)
MATLAB SYSTEM IDENTIFICATION
TOOLBOX
LINEAR MODEL ARX
Linear MPC – Identification procedure
12 12.5 13 13.5
0 0.5 1 1.5 2
Time [h]
Continuous model Linear model
Steam flowrate [t/h]
Similar trends within the prediction horizon
CPU times get significantly reduced
h
pLinear model (ARX)
10% disturbance on the waste flowrate
Simplified continuous
model (68 DAE) 0.54s
4.E-5s CPU time for a 1 min
simulation interval
Linear MPC – Models comparison
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano 11
11.5 12 12.5 13 13.5 14
0 0.2 0.4 0.6 0.8 1
Ave rage s tea m f low rate [t /h] ( CV )
Time [h]
Linear MPC Set point Non-linear MPC
Comparison between linear and non-linear MPC
The non-linear MPC reaches the setpoint faster
Linear MPC – Servo problem
11.4 11.6 11.8 12 12.2 12.4
0 0.2 0.4 0.6 0.8 1
Ave rage s tea m f low rate [t /h] ( CV )
Linear MPC Set point Non-linear MPC
Linear MPC – Regulator problem
Comparison between linear and non-linear MPC
The non-linear MPC shows a smaller
deviation from
the setpoint
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
CPU time for a 20 min
simulation of the whole plant
CPU time of the optimization procedure (600 F
objcalls) Non-linear
MPC
0.65 [s] 390. [s]
Linear MPC 7.E-4 [s] 0.45 [s]
11 11.5 12 12.5 13 13.5 14
0 0.2 0.4 0.6 0.8 1
Po rta ta v a p o re m e d ia [t/ h ] (C V)
Tempo [h]
MPClineare Set point MPC non lineare
The control efficiency is comparable The CPU time is shorter than the control time only for the linear MPC Higher robustness for the linear MPC
Optimization CPU time < t c
The implementation of the
linear MPC on the real plant is viable
Conclusions
12 12.5 13 13.5 14
0 0.2 0.4 0.6 0.8 1
Ave rage s tea m f rlow rate [t /h] ( CV )
Time [h]
HP = 20 HP = 10 HP = 40 Set point
Prediction horizon, h p :
short: reduced predictive capability
long: the prediction becomes less realistic
Linear MPC – Effect of varying h p
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano 11
11.5 12 12.5 13 13.5 14
0 0.2 0.4 0.6 0.8 1
Steam flowrate [t/h] (CV)
Time [h]
Average value Instantaneous value Set point
4000 4500 5000 5500 6000 6500 7000
0 0.2 0.4 0.6 0.8 1
Secondary air flowrate [Nm3/h] (MV)
Time [h]
3800 3850 3900 3950 4000 4050 4100 4150 4200 4250 4300
0 0.2 0.4 0.6 0.8 1
Waste flowrate [kg/h] (MV)
Time [h]
12000 12600 13200 13800 14400 15000 15600 16200 16800 17400 18000
0 0.2 0.4 0.6 0.8 1
Grates air flowrate [Nm3/h] (MV)
Time [h]
h c = 3
Control horizon, h c :
short:
Few degrees of freedom.
Bolder actions from manipulated variables long:
Longer CPU time, Smoother control action
Linear MPC – Effect of varying h c
Small weighs ( ) on the
manipulated
variables
1111.2 11.4 11.6 11.8 12 12.2 12.4 12.6 12.8
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Steam flowrate [t/h] (CV)
Time [h]
Average value Istantaneous value
Set point
4000 4200 4400 4600 4800 5000 5200 5400
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Waste flowrate [kg/h] (MV)
Time [h]
12000 12150 12300 12450 12600 12750 12900 13050 13200
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Grates air flowrate [Nm3/h] (MV)
Time [h]
5600 5800 6000 6200 6400 6600
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Secondary air flowrate [Nm3/h] (MV)
Time [h]
18 20 22 24 26 28 30
0 0.2 0.4 0.6 0.8 1 1.2 1.4
First grate strokes [Strokes/h] (MV)
Time [h]
18 20 22 24 26 28 30
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Second grate strokes [Strokes/h] (MV)
Time [h]
u
The controller becomes nervous.
We can observe a RINGING trend
Linear MPC – Ringing
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
k k
h
c 1k h
pCo n tro lled va ria bl e (CV) M an ipu lat ed va ria b le (M V)
Time Time
p
c h
h h p high:
• High predictive capability
• Significant role played by the model error
h c high :
• High number of d.o.f.
• Smoother control actions
This is the effective action that is implemented in the plant
t c
t c Control time:
Shorter than 1/10 of the characteristic time
Linear MPC – Prediction and control horizon
t
d 1111.5 12 12.5 13 13.5 14 14.5
0 0.5 1 1.5 2
Steam flowrate [t/h] (CV)
Time [h]
3800 3900 4000 4100 4200 4300 4400 4500 4600
0 0.5 1 1.5 2
Waste flowrate [kg/h] (MV)
y
x
pmin 40
t d
p
t c
h p
h c
= 1 min
= 20
= 1
MPC – Selection of the controller parameters
© Davide Manca – Dynamics and Control of Chemical Processes – Master Degree in ChemEng – Politecnico di Milano
940 950 960 970 980 990 1000 1010
0 0.2 0.4 0.6 0.8 1
Postcombustion temperature [°C] (CV)
Time [h]
No constraint With constraint
Law constraint
10.4 10.6 10.8 11 11.2 11.4 11.6 11.8 12
0 0.2 0.4 0.6 0.8 1
Steam flowrate [t/h] (CV)
Time [h]
No constraint With constraint Set point
Setpoint step change of –10%
Law constraint on the outlet temperature from the
postcombustion chamber:
T > 950 °C
NL MPC – Constraints
11 11.2 11.4 11.6 11.8 12 12.2 12.4
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Steam flowrate [t/h] (CV)
Time [h]
No switch Set point
Switch
11 11.2 11.4 11.6 11.8 12 12.2 12.4
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Steam flowrate [t/h] (CV)
Time [h]
No prediction Set point Prediction