Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Post on 14-Jan-2016
27 Views
Preview:
DESCRIPTION
Transcript
1Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
The Development of an Advanced Systems Synthesis Environment: Integration of MI(NL)P
Methods and Tools for Sustainable Applications
Zdravko Kravanja
University of Maribor, Faculty of Chemistry and Chemical Engineering,
Smetanova 17, 2000 Maribor, Slovenia
2
Slovenia in pictures Area: 20,273 km2
Population: 2.0 million Capital city: LjubljanaLanguage: Slovenian; also Italian and Hungarian in nationally mixed areas Currency: EURO, €Member of EU - 1 May 2004
EU Presidency for 2008
3Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
http://epi.yale.edu/CountryScores
Environmental Performance Index (EPI)
Slovenia has rank 15
4
Outline
• Introduction
• Process Synthesis and Sustainability, Challenges
• Capabilities of an EO Modular MINLP Process Synthesizer MIPSYN
• Aplications
• Conclusion
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
5Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
But the creative principle resides in mathematics. In a certain sense, therefore, I hold true that pure thought can grasp reality, as the ancients dreamed.
Albert Einstein
6Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Key idea for today and tomorrow
In (bio)chemical supplay chain the traditional use of optimization techniques and tools is
not sufficient
unless its efficiency and applications are consistently upgraded with
sustainable principles
7
Creative Principles of Mathematical Programming
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Optimality Competitive advantage
Feasibility Constraints satisfied
Integrality Simultaneous considerations
Creative principles of MP enables:• Creation of new knowledge and• New innovative solutions
Study of solutions enables one to get new insights,e.g. simultaneous
HI also reduces raw material usage (Lang, Biegler, Grossmann, 1988).
8
Introduction
Incentives for sustainable development
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
• Main problems that have to be circumvented:– Population growth– Limited resources– Environmental and society destruction
• How prevent the worming for 2oC in the next 2 decades?!
• Answer: Sustainable development
• New role of PSE: Sustainable PSE of paramount importance
9
3 X 3 Sustainability Matrix
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Nature Sustainability
Eco-centric 3
Strategies Expanded- 2 anthropozentric 3 Sufficiency 2 Consistency Narrow 1 1 Efficiency anthropozentric
1 2 3 Principle of Justice, Etics Just Reward for Work
Respect for Private Property Fair Distribution of Goods
(M. F. Jischa, Chem. Eng. Technol. 21, 1998)
1
8
27
Figure 1: Diagonal as a measure of sustainability
10
Environmental Aspects
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Environmentally friendly innovation
In addition:
Material brought into the environment < Carrying capacity of
the ecosystem ->min emission of pollutant
Consummation rates of renewables
Their regeneration rates ->
max renewables
Non-renewable resources only if future generation would not be compromised ->
min non-renewables
->Multiobjective
approach
<
(Voss, 1994)
Environmental constraints Opt. Criteria
11
MINLP Model Formulation for Different Levels of Innovations:
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
a) max z = cTy + f(x) – e(x) b) s.t hi(x) = 0 c) gi(x) 0 d) Biy + Cix bi x X = x Rn: xLO x xUP y Y = 0,1m
a) Objective function as a real-world economic function (cost benefit approach):
Max Profit = Production income - Raw material cost - Utility cost
- Investment cost – Environmental loss
b) Equality constraints: mass and energy balances, design equations
c) and d) Inequality constraints: product specifications, operational, environmental and feasibility constraints, logical disjunctive constraints for selection of sustainable alternatives
i Levels }
12
Sustainable and Integrated (Bio)chemical Supply Chain Synthesis
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Fig.??(Marquardt Wolfgang, Lars Von Wedel, and Birget Bayer.AspenWorld 2000, Orlando, FL, 2000)
18
r
Sustainability27
Figure 2: Diagonal as a measure of sustainability
13
Sustainable Product-Process Synthesis
“Synthesis is the automatic generation of design alternatives and the selection of the better ones based on incomplete information”
A. W. Westerberg (1991)
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Sustainable product-process synthesis is the automatic generation of design candidates and the multiobjective
selection of the better ones based on the creative postulation of sustainable alternatives integraly
accross the whole chemical supply chain.
Extension:
14
Challenges Related to the Manifolds Nature of the Synthesis Problems
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Many complex interactions SimultaneousSimultaneous
Discrete and continuous decisions MINLPMINLP
Uncertainty FlexibilityFlexibility
Dynamic systems MIDNLP, MIDNLP, multiperiodmultiperiod
Rule-based decisions Logic-based Logic-based
Multicriterial MultiobjectiveMultiobjective
Features Approach
15
Simultaneous Synthesis and Heat Integration - Methanol Example Problem
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Process synthesis and:• sequential HEN synthesis: - 1,192,000 $/yr (loss!)• simultaneous HI by Duran-Grossmann’s model: - 292,000$ $/yr (loss!)• simultaneous HEN synthesis by Yee’s model:
• Yee, Grossmann, Kravanja (1990) 1,845,000 $/yr (profit!).• Kravanja and Grossmann (1994) 2,613,000 $/yr (profit!)
Figure 3: Methanol process and HEN superstructure
Figure 4: Optimal process scheme with HI HEN
16
Table 1: Types of optimization problems and models
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Equations Linear Nonlinear Difference Differential Model Steady state Multiperiod Dynamic Example Continuous process Life cycle Batch
process Certainty variables
Nominal
Continuous, x LP NLP e.g. e.g. discrete, y 0-1 ILP INLP logical Y DisLP DisNLP x, y MILP MINLP Mul. MINLP Dyn. MINLP x, Y MDisLP MDisNLP Uncertain par. Flexible
Kravanja Z., 2003, Chem. Biochem. Eng. Q. 17 (1), 1-3.
Different Modeling Complexities
17Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Incentives for the development of MP-based tools for process synthesis:
• Several general MINLP solvers www.gamsworld.org/minlp/solvers.html
• Logic-based solver LOGMIP (Vecchietti and Grossmann, 1997)
• Global MINLP Optimizer BARON (Sahinidis, 2000)
• Almost no tool specialized in MINLP synthesis
18
Capabilities of Mixed-Integer Process SYNthesizer MIPSYN
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Extension of PROSYN-MINLP Kravanja, Z. and I.E. Grossmann, Computers chem. Engng.,1990 Kravanja, Z. and I.E. Grossmann, 1994
• Robustnes: – Interactive vs. Automated mode of execution– NLP initialization by a simple flowsheet simulation – Different NLP and MILP optimizers
• Efficient handling of process superstructures– M/D strategy with alternative decomposition schemes of the superstructure– Multilevel MINLP strategies
• Efficient handling of models:– Data- and topology independent modeling– Convex-hull and alternative convex-hull modeling formulation – Model generation from modules of process units and interconnection nodes– Simultaneous heat integration
• Algorithmic power: – Different extensions of the OA algorithm– Different convexifications to prevent poor local solutions– Integer-infeasible path optimization
• Higher-level capabilities:– Multiobjective synthesis– Multiperiod synthesis– Flexible synthesis in the presence of uncertain parameters
19
MIPSYN and Logic Based OA
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Or when NLP is not imroving
20
MIPSYN flowchart
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Topology
P_STRUCT.DAT
Components
P_ COMPON.DAT
User’s modules
MY_MODEL.DAT
Data
P_DATA.DAT
Model generator
MIPSYN Libraries:
AP/OA/ER - Process modules
M/D - Components properties
NLP initializer
Simple simulator
Solution
P_OPTIMUM.RES
Procedure overview
P_B.RES
GAMS
NLP solvers: CONOPT, MINOS, SQP
MILP solver: CPLEX, OSL,
21
Applications
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Chemical Engineering(MIPSYN)
NLP optimization • Process sybsystems • Flowsheets
MINLP synthesis:• Reactor networks• Separator networks• Heat exchanger networks• Overall HI process flowsheets
Mechanical Engineering(TOP)
NLP optimization • Timbes trases • Composite floor systems
MINLP synthesis of mechanical structures:
• Gates for hydropower dams• Steel frames• Steel buildings
Different levels of problem abstraction and application• More general MINLP solver
• Process synthesizer
• Synthesizer shell for different domains
22
PROSYN-MINLP verion
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
23
MipSyn β Version
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
24
Multilevel-hierarchical MINLP Synthesis
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Combination of the hierarchical strategy and MINLP superstrucutre approach
(Kravanja and Grossmann;1997)
MINLP 1: RCT network:- Detailed RCT network model- Simple SEP model - Simultaneous heat integration
MINLP 3: HEN synthesis- Fixed RCT/SEP structure- Detailed RCT and SEP modules- Staged HEN synthesis model
Tagret HI
Identify SEP tastks
Tagret HI
Identify process streams
HI
Identify SEP tastks
Profit UB
LB
STOP if
UP≈LB
LoopMINLP 2: SEP/RCT network:- Detailed RCT models- Detailed SEP models- Targeted heat integration
25
MINLP 1: Initial Reactor Network and Simplified Separation Superstructure
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
HDA example
26
MINLP 1 – Optimal Solution
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Identified separations
Targeted HI
Upper Bound 6.505 M$/yr
27
MINLP 2: Detailed RCT and Identified SEP
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
28
MINLP2: Optimal Solution
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Identified hot and cold
streams
Targeted HI
Upper Bound 5.892 M$/yr
29
MINLP 3: HEN Synthesis within Fixed Flowsheet
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Lower Bound 5.201 M$/yr
30
MINLP II resolved: UB = 5.240 M$/yr
MINLP III: LB = 5.201 M$/yr
STOP
OPTIMAL SOLUTION:
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
MINLP 2 Resolved
Since UP≈LB →
31
Multilevel Synthesis of Mechanical Structure
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Superstructure : • 2 main gate element • 4 to 6 horizontal girders • 5 to 9 vertical girders
MINLP1: topology optimization
• relaxed standard dimensions
• OAs accumulated for MINLP2
MINLP2: simultaneous topology and standard dimension optimization
• discrete standard dimension
• OAs accumulated for MINLP3
MINLP3: simultaneous topology, standard and rounded dimension
• optimization and pre-screening
• 10 discrete dimensions on each side from the optimal solution of MINLP2
LINKED MULTILEVEL HIERARCHICAL STRATEGY (LMHS)
SYNTHESIS OF ROLLER HYDRAULIC STEEL GATEHydroelectric Project Blanda, Iceland
(S. Kravanja, A. Soršak, Z. Kravanja; 2003)
32
Optimal Structures
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
4000 mm
4600 mm
4120 mm
30 30 10 10 10 30 30
100 100 100 100 100 100 100
28245 1079.5 1079.5 1079.5 1079.5 282 45
4972
414
25
479
40
Optimal solution: 8804 €Self-manufacturing costs of the erected gate: 13498 €35% net profit
19622 y !
33
Optimal Synthesis Under Uncertainty
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
• Statement: Engineering problems have in the practice much larger numbers of
uncertain parameters than we can handle rigorously
• Consequences:• Flexible but suboptimal (safety factors)• Optimal at nominal conditions but may be inoperable
• Motivation: The synthesis and design of flexible and optimal engineering
structure
• Goal: An automated and robust strategy for problems with up to 100 of
uncertain parameters.
34
MINLP Synthesis Under Uncertainty
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
• Integration over space of Θ – stochastic optimization: EC or EP• 2NP feasibility constraints + 5NP Gaussian quadrature points
Total: 2NP+ 5NP
max P(y,x,d,) max wi Pi (y, xi, d, i) y,x,d i
s.t. h(y, x, d, ) = 0 s.t. hi (y, xi, d, i) = 0
g(y, x, d, ) 0 gi (y, xi, d, i) 0 i QP
xX, dD, TH xi X, d D, i TH
y0,1m y 0,1m discretization
- problem multiperiod problem
Answer: Simplified approach
35
Minimal Set of Feasibility Constraints
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Definition: Critical points are those the worst combinations of uncertain parameters that determine optimal oversizing of design variables
needed to achieve desired flexibility
• Extreme vertex points when the problem is convex No 2NP
• A priory determination of Critical Points (Novak Pintarič and Kravnja, 2008)
• Sequential scanning of all vertex points• Without sequential scanning of all vertex points
– KKT based method (rigorous)– Iterative method– Approximate non-iterative method
No = ND• Combination of Critical Points by using set covering problem
No ≤ ND (less than ND/5)
36
Apriory Identification of Critical Points and Minimal Set of Feasibility Constraints
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
fx
, , ,
fx
fx
dLO UP
fx
min ( , , , , )
s.t. ( , , , , ) 0
( , , , , ) 0( , , )
, , , , 0,1
ix z d
m
C y x z d M d
h y x z d
g y x z dd g x z
x z d R y
Drawback: approximative
Advantages:
• Model size depend on the number of design variables
• Robust
• Can be applied to complex large-size process models
Maximization of di
NLPi
No ≤ ND (less than ND/5)
37
Approximate Stochastic Optimization
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Approximate expected objective
function in CBP
Assure flexibility of design in min
No CP
Enforce approximate trade-offs
CBP
,
CBP c
CBP cC
CBP cd d,
LO
min ( , , )
s.t. ( , , ) 0 ( , , ) 0
( , , ) 0 ( , , ) 0 1,...,
( , ) ( , )
; , ,
x d
k k k
k k k
k k k
k
C x d
h x d h x d
g x d g x d k n
d g x d g x
d d x x d R
38
Three-level MINLP Strategy for Flexible MINLP Synthesis
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
MINLP level 1: Deterministic non-flexible synthesis at the nominal conditions
MINLP levels 2 and 3: Flexible stochastic MINLP synthesis
Level 2 Level 3
Significant reduction of problem's
size!
Flexibility analysis ot the final optimal solution
39
Synthesis of Flexible Heat Integrated Methanol Process
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Structure alternatives:• Two feeds• One- or double stage compression of the feed• Two reactors • One- or double stage compression of the
recycle stream• 8 y
HEN:
• One-stage MINLP model
• 4 hot and 2 cold process streams partitioned into several segments
• 38 y for the selection of the matches
From Kravanja, Z., Grossmann, I. E. (1990).Updated prices
40
Level 1: Deterministic Non-flexible Synthesis at the Nominal Conditions
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
– Profit of 37.37 MUSD/yr
– Not feasible if small deviations in the uncertain parameters
from the nominal values
MINLP I
HEN: 2 HEs and 2 coolers
41
Flexible MINLP Synthesis
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
27 uncertain parameters: Gauss distribution, 6 σ interval
• Raw material prices (2)• Temperature of the feeds (2)• Pressure of the feeds (2)• Conversion parameters for reactors (2)• Compression efficiency (1)
• Product demand (1)• Heat transfer coefficients (9)• Price for methanol (1)• Composition of the feeds for H2
and CO (4)• Utility prices (3)
• Only 4 critical vertices !!!• Profit reduced from 37.37 to 33.04 MUSD/a• The same optimal structure as deterministic one
MINLP Level 2: Flexible MINLP synthesis at nominal condition
MINLP Level 3: Flexible MINLP synthesis at CBP
• Profit reduced from 33.04 to 32.72 MUSD/a• The same optimal structure
Flexibility analysis: Flexibility index 1.000
42
Comparison
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Mode DeterministicMINLP I
Flexible – nominalMINLP II
Flexible at CBP(Appr.Stohastic)
MINLPIII
Power COMP-2 (MW) 18.49 29.57 29.57
Power COMP-3 (MW) 15.56 27.97 27.98
Power COMP-4 (MW) 3.34 3.34 3.00
Volumen RCT-1 (m3) 72.78 77.42 77.87
A HE1 (m2) 558.56 529.59 529.33
A HE2 (m2) 208.53 402.82 401.01
A Cooler 1 (m2) 518.46 946.48 967.38
A Cooler 2 (m2) 2436.24 2396.71 2368.37
No of simultaneous points 1 5 5
Continuous variables 572 2656 2656
Discret variables 46 46 46
(In)equalities 580 2892 2892
CPU per NLP (s) 0.1 2.5 1.7
CPU per MILP (s) 0.1 0.85 0.6
43
Multiobjective Sustainable Process Synthesis
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Strength:• Simultaneous approach • Numerous interactions exploited
Drawback:• Richness of the solution depends on
the abundance of alternatives
Two-step superstructural MINLP approach • 1st economic-based MINLP step for basic process
superstructure that comprises technological end economical alternatives
Base case solution
• 2nd multiobjective MINLP step for sustainable superstructure, augmented by additional environmental and other alternatives
Sustainable solution
Novak Pintarič and Kravanja, 2005
44
Solution of the Multiobjective Multilevel MINLP Problem
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
econ econ econ env
econ
max (1 )
s.t.
Design or Synthesis Model
0 1
w RSI w RSI
w
a) Weighted sum method:
b) -constraint method
econ
env
max
s.t.
Design or Synthesis Model
RSI
RSI
where:
Relative environmental index:
m, m, m, ,env ,0 0 0 0
m, m, m, ,
mass usage energy usage water usage polution indicators
1
j
k n c olj c
k IS l EC n WC j PIM c IC o OSk l n c o
q q qRSI PF
N q q q
Relative economic index:
econ 0
PBRSI
PB
45
Solution of the Multiobjective MINLP HDA Case Study
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Fig. 8: Basic process superstructure
1st economic-based MINLP step
46
HDA Case Study 1st Economic-based MINLP Step
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Fig. 2: Economically optimal process flowsheet – base case
PW W1 W2Profitk$/yr
EkJ/kg
Mkg/kg
Wkg/kg
GWkg CO2/kg
Hkg/kg
Xtot
HIQC = 4.203
QH = 05579 0 1.2451 0.3370 0.0078 1.0011 0.9995
47Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Fig. 9: Superstructure, enlarged by sustainable alternatives
Recycling of diphenyle
Heat integration
HDA Case Study 2st Multiobjective Sustainable MINLP Step
48Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Scalar parametric optimization:
0,70
0,80
0,90
1,00
1,10
1,20
0,60 0,70 0,80 0,90 1,00 1,10 1,20 1,30 1,40 1,50 1,60
GEI
Rel
ativ
e pr
ofit
Fig. 10: “Pareto curve” obtained by scalar parametric optimization
Size of NLPs:1400 variables1300 constraints
Size of MILPs:55 binary, 2004 c. variables up to 2040 constraints
1/4h CPU on 1.8 GHzIntel Pentium M processor 1G RAM
HDA Case Study (Cont.) 2st Multiobjective Sustainable MINLP
Step
Very good solutions !
Relative environmental index
R
elat
ive
prof
it
49Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
• Alternatives with synergistic effects on economic and environmental criteria.
• More profitable and less environmentally harmful solution can be obtained
• Most of alternatives do not show clear trends in their impacts on economic and environmental indicators.
• Interactions can be very complex and unpredictable.
• Importance of the simultaneous approach to the sustainable synthesis of process schemes.
Multiobjective Sustainable Process Synthesis
50Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
xLO∙y ≤ xs ≤ xUP∙y:
Declared: 0 ≤ xs ≤ xUP
xf + (xLO – xf)y ≤ xa ≤ xf + (xUP – xf)y
Declared: xLO ≤ xa ≤ xUP
y=1xUP
y=0 xLO
xS,LO=0 xS,UP= xUP
y=0,1
0
xUPxLO
Xa,LO=xLO xa,UP= xUP
Fig1.a: In conventional discrete/continuous formulation
Fig.1b: In alternative discrete/continuous formulation
Translation of variables(Ropotar and Kravanja; 2008, 2009)
Efficient MINLP model formulations
xs = xa – xf(1 – y)
y = 0 → xa = xf
y =1 → xa = xs
51
Alternative logic-based OA algorithm
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
NLP subproblem:
g gs.t. ( ) 0h x
a a g g
,
min ( )k
lik
lik ik
i D k SD forY true
Z c f f
x x
g g g( )A bx
r g a r( , )A bx x
LO a UP
a
a
( ) 0
( ) 0
for
0
ik
ik
lik ik k ik
ik
h
A
c i D ,k SD Y true
c
x
x
x
x x
[Y: xs = xa]
• NLP are solved only for currently selected alternatives• No singularities -> robustnes significantly improved
52Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Alternative Logic-based OA Algorithm:Translation of OA MILP Master Problem
a g
Tg g
T g
g g g
g g
r g s r
min
s.t.
( ), 1,...,
( ) 0
( )
( )
( , )
ik ik iki k
l l lx
l l lx
Z c y
f fl L
h h
A b
E e
A b
x x x x
x x x x
x
y
x x
ss
s
s
Ta s a
Ta a
T s
T
g s
g a
LO g UP
, 1,...,
, , 0,1
0 , ,
ikik
ik
ikik ik
lx ik ik
l l lx ik ik ik
lx ik
l l lx ik ik ik
mn
ik k
xLOy xx
x xUPy
A b y
f
f f y
h
h h y l L
R
i D k SD
X
x
x x
x x x
x x
x x x
x x x y
x x x
xs = xa – xf(1 – y)
s
a g
Tg g
T g
g g g
g g
r g a r
aa
f f
a f
min
s.t.
( ), 1,...,
( ) 0
( )
( )
( ( , )
///////////////////////////////
)
)
1
/
,
(
ik ik iki k
l l lx
l l lx
ikik
ikik ik ik
x
Z c y
f fl L
h h
A b
E e
A b
xx xUP x y
A y b y
x
x
x x x x
x x x x
x
y
x
X
x
x y
x
Ta f
f
Ta a
Ta a
T
T
g
g a
L
a
a
a
LO a UPU
f
O P
T
g
f
( )
( ) , 1,...,
, , 0,1
0 , ,
,
lik ik
l l lx ik ik ik
lx ik
l l lx ik ik ik
mn
ik k
lx ik
lx ik
f
f f y
h
h h y l L
R
i D S
f
h
k D
x
x
x x
x
x x
x
x x
x
x x x
x x
x
x
x
x x x
y
xx x
(CCH-MILP) (ACH-MILP)
xf =xLO
53
Comparision
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
0
1
2
3
4
5
6
small moderate large
Eff
icie
ncy
(CPU
CC
H/C
PUA
CH
)
Problem size
Reactor network
HEN
Allyl chlorideM
ILP
NLP
MIL
PN
LPM
ILP
NLP
MIL
PN
LPM
ILP
NLP
MIL
PN
LP
MIL
PN
LPM
ILP
NLP
MIL
PN
LP
40 ys 32 ys 172 ys
184 ys
249 ys100 ys
600 ys
371 ys
400 ys
Figure 5: Efficiency in solving MILP and NLP master problems vs. problem size
54Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
mn
y
RXXx
yxg
yxh
xfycZ
1,0
,
0),(
0),(
s.t.
)(min T
EO ext
TEO ext
EO EO ext
ext EO ext
EO EO ext
ext EO ext
EO ext
EO ext
EO EO ext ext
min ( , )
s.t.
( , , ) 0
( , , ) 0
( , , ) 0
( , , ) 0
,
,
0,1
n
n n
Z c y f x x
h x x y
h x x y
g x x y
g x x y
x x x X R
X X X
x X R x X R
y
m
extEO nn
n
EO EO ext
ext EO ext
( , , ) 0
( , , ) 0
h x x y
h x x y
0),( yxh
ext EO
TEO EO
EO EO EO
EO EO EO
EO EO
min ,
s.t.
( , , , ) 0
( , , , ) 0
0,1
n n n
m
Z c y f x Φ x
h x Φ x y y
g x Φ x y y
x X R R
y
),( EOext yxΦx
What if models are too large and compex to be solved in EO environment?
Answer: Hybrid models
Hybrid Modeling and Solution Environment for Disjunctive Models
55
Reactive-Distillation Superstructure (ETBE)
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Feed 2
Cond.
Reb.
Dist.
prod.
Feed 1
Feed 2
Cond.
Reb.
Dist.
prod.
Feed 1
Feed 2
Cond.
Reb.
Dist.
prod.
Feed 1
Feed 2
Cond.
Reb.
Dist.
prod.
Feed 1
Feed 2
Cond.
Reb.
Dist.
prod.
Feed 1
Cond.
Reb.
Dist.
prod.
Feed 1
Cond.
Reb.
Dist.
prod.
Cond.
Reb.
Dist.
prod.
Cond.
Reb.
Cond.Cond.
Reb.Reb.
Dist.
prod.
Feed 1Feed 1
• Superstructure consists of – Three sections of alternative trays
– Fixed feeds, condenser and reboiler
– Each tray can be • Selected for separation• Selected for reaction or• By-passed
Figure 11: Tray superstructure Figure 12: Column superstructure
Ropotar, Novak Pintarič, Reneaume and Kravanja, 2009
56
Hybride MINLP model in MIPSYN
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
External FORTRAN:• Liquid and vapor enthalpies • Reaction rate • Equilibrium constant • Mass of catalyst• Tray dimension
EO environment in GAMS:• Objective function• MESH equations for separation trays• MESH equations for reaction trays• By-pass• Logical constraints
MIPSYN enables:• Execution of NLP subproblem and external sub-model only for existing trays to
reduce the size and prevents numerical problems to occur. Challenge: how to handle different hybrid model sizes within
MINLP iterations? • Initialization of each NLP which increases the model robustness.• Several strategies to handle nonconvexities• Miltilevel MINLPs: the next level starts from the optimal solution of
the current level
Hybrid Modeling and Solution Environment for Disjunctive Models
57
Solution for the Hybride System
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Process parameters 1-level MINLP with multiple restarts
Multiple level MINLP (2ndlevel)
Multiple level MINLP with constrained integer-cuts (2ndlevel)
Position of the feeds 8, 36 8, 37 10, 37
Position of reaction trays 3, 5, 7, 9, 11,
13, 15, 18, 36
2, 4, 6, 10, 14, 23, 25, 32, 38, 40
3, 5, 7, 10, 12, 14, 16, 21, 34, 37, 39, 41
Number of separation trays 37 36 35
Flow of distillate, mol/s 0.0648 0.0646 0.0642
Flow of product, mol/s 0.0281 0.0282 0.0284
Reboiler duty, W 4 024 3 687 3 377
Condenser duty, W 4 230 3 895 3 586
Isobutylene conversion, % 99.36 99.44 99.71
Annual cost, k$/year 8.926 8.809 8.571
Table 2: Solution for three different strategies.
For up to 10 reaction and 50 separation trays:
• 3000 constraints
• 1500 variables
• 150 binary variables
External
• 500 constraints
• almost all variables
58Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Extending Process Synthesizer MIPSYN for the Synthesis of Bioprocesses
• MIPSYN Library extended for modules:MIPSYN Library extended for modules:
• Substrate preparation
• Bioconversion
• Product purification
• Solids drying
• Objective function - maximizing revenue:Objective function - maximizing revenue:
• Without investment
• With investment
59Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Process description from Ramkumar Karuppiah et al., 2008
Optimization of the Corn-based Ethanol
FEED: FEED: Corn Kernels (18 kg/s) PRODUCTS:PRODUCTS: Ethanol (5.81 kg/s) Distillers Dried Grains with Solubes (4.15 kg/s) Biogas (1.047 kg/s)
Substructures: • Feed preparation (washing, grinding, cooking)• Enzymatic hydrolysis (liquefaction, saccharification) and fermentation • Ethanol purification (distillation, adsorption)• Solids drying (centrifugation, floatatition, drying)
Alternatives - Different routes for separation solid – liquid:• Separation before the beer column• Separation after the bottom of the beer column
60Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Corn Washing water
Water
glucoamylase
Saccaromycescerevisiae,urea, water
,,
,CO2, O2
DDGS
VOC
Biogas
water
Corn grits
REC-1SPL-2
FEED-9
ADS-1
SPL-4
SPL-3
CDES-1
CADS-1
HEH-5
y1
Superheatedsteam
a-amylase
WASH-1FEED-1 FEED-2
GRIND-1
FEED-3
MXR-1
HEH-1PREMIX-1
FEED-5
FEED-6
FEED-7
HEC-2
MXR-2
LTANK-1
SAC-1
HEC-3
HEH-2
MXR-3
STOR-1
FER-1
STOR-2SPL1-1
MXR1-1
MXR-10
FLOT-1
MECP-1
MXR-8
PRD-1
HEH-3
HEC-4
HEC-7
MXR1-2
MXR1-3
SPL1-2
MECP-2
FLOT-2
HEC-10
BC-1
DRY-1
SPL-5
HEC-8
PRD-9
PRD-6
PRD-8
PRD-7
MXR-9
WWT-1MXR-4
PRD-2
HEH-4
PRD-3
SPL-1
MXR-5
MXR-6
MXR-7
HEC-5
HEC-6
PRD-5
HEH-6 Dry air
FEED-8
PRD-4
Bioethanol
y2
Figure 13: Superstructure of a corn-based ethanol plant
Non heat integrated process:
21.018 M$/yr bioethanol: 5,837 kg/sbiogas: 1,015 kg/sDDGS: 4,174 kg/s.
Heat integrated process:
31.952 M$/yr bioethanol: 5,107 kg/sbiogas: 1,047 kg/s inDDGS: 4,150 kg/s.
Solution with MIPSYN
Sythesis of Bioethanol
61
MINLP Synthesis Biogas Process from Organic and Animal Waste
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Figure 14: Superstructure for selecting the optimal processing system for an industrial case study
62Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
Figure 15: Optimal solution for the industrial case study of biogas production with NPW of 7.730 MEUR
MINLP Synthesis Biogas Process from Organic and Animal Waste
63
Conclusion
Vision:
In order to prevent global worming and achieve efficiency and suficiency in production and consumption:
redesign or fundamentaly innovate chemical and process industries based on sustainability principles appliead to the whole (bio)chemical supply chain.
The greatest challenge for the PSE community:
Based on the systems approach, to provide engineers and scientists with powerful concepts, methods and tools so that they will be able to shape this sustainable development.
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
64
THANK YOU
Plenary Lecture, ESCAPE 19, Krakow, Poland, 14 – 17 June 2009
top related