FJCP 2005 14/11/2005
[email protected],
[email protected]
Research and Future Business Division
DASSAULT AVIATION
Alexis Anglada, Laurent Zimmer
Towards a COnstraint and Decision based Environment for Design: the CODED Project.
FJCP 2005 14/11/2005
Our goal An industrial example Constraint Explorer Future Works
Outline
FJCP 2005 14/11/2005
and solving system Decision Aiding system
Modelling
Physical laws Engineering knowledge Components of the shelf Customer requirements Preference elicitation
Our goal : a preliminary design environment
FJCP 2005 14/11/2005
An industrial example
FJCP 2005 14/11/2005
To design an air conditioning system (ACS)
Characteristics:
For ram air and main air Mono or Multi-pass
Type of exchange surface Type of entry section
FJCP 2005 14/11/2005
Main
Cross-flow heat exchanger
FJCP 2005 14/11/2005
Constraint Explorer
FJCP 2005 14/11/2005
Constraint Explorer
matrix
#Afr1:= Lx1*Lz1; #Afrr1:= Ly1*Lz1;
alias
T2 : real = [216.5,600]; Asurface1 : int = [1,6];
real and integer variables
FJCP 2005 14/11/2005
coKc1: mat[3,3]; coKcf1:mat[3,3]=[[-8.4777443e-5, 9.4071329e-5, 6.3169044e-3], [-2.5463817e-3, 2.9576219e-3 ,-1.7313148e-1], [-3.7733783e-1, -3.0743137e-2, 1.5510967]];
Language:
Modelling in CE 1/3
Equations, Inequalities
Conditional Constraints
FJCP 2005 14/11/2005
Asurface1=1 ~> {b1=1.382e-2; rh1=9.589e-4; delta1=2.540e-4; beta1=951; AfsurA1=0.863; facade1=2;}; Facade1 = 1 ~> {coKc1=coKcf1; coKe1=coKef1;};
T1 = T0b*(1 + (1/etaec)*((p1/p0b)^((gamma-1)/gamma) - 1));
Language:
Modelling in CE 2/3
b1 1.382e-2 6.350e-3 9.525e-3 9.525e-3 5.21e-3 12.3e-3
rh1 9.589e-4 7.709e-4 7.557e-4 6.699e-4 3.84e-4 8.525e-4
FJCP 2005 14/11/2005
delta1 2.540e-4 1.524e-4 2.540e-4 2.540e-4 1.02e-4 1.02e-4
beta1 951 1204 1138 1250 2231 1115
AfsurA1 facade1 0.863 2 0.756 1 0.822 1 0.840 1 0.841 1 0.862 1
Component of the shelf expressed by conditional constraints
Asurface1 1 2 3 4 5 6
Modelling in CE 3/3
Component of the shelf expressed by conditional constraints
FJCP 2005 14/11/2005
Asurface1=1 ~> {b1=1.382e-2; rh1=9.589e-4; delta1=2.540e-4; beta1=951; AfsurA1=0.863; facade1=2;}; Asurface1=2 ~> {b1=6.350e-3; rh1=7.709e-4; delta1=1.524e-4; beta1=1204; AfsurA1=0.756; facade1=1;}; Asurface1=3 ~> {b1=9.525e-3; rh1=7.557e-4; delta1=2.540e-4; beta1=1138; AfsurA1=0.822; facade1=1;}; Asurface1=4 ~> {b1=9.525e-3; rh1=6.699e-4; delta1=2.540e-4; beta1=1250; AfsurA1=0.840; facade1=1;}; Asurface1=5 ~> {b1=5.21e-3; rh1=3.84e-4; delta1=1.02e-4; beta1=2231; AfsurA1=0.841; facade1=1;}; Asurface1=6 ~> {b1=12.3e-3; rh1=8.525e-4; delta1=1.02e-4; beta1=1115; AfsurA1=0.862; facade1=1;};
Modelling in CE 3/3
152 variables 148 constraints (56 conditional constraints) 1296 possible configurations
FJCP 2005 14/11/2005
2 drag formula, 2 heat exchanger efficiency, the system entropy
5 performance criteria for decision support
(number of possible entry section*number of possible exchange surface)²
Model characteristics:
The ACS model
FJCP 2005 14/11/2005
chronological, round robin
Backtrack algorithms
HC3,BC3,BC4
Hull and Box Consistency
Interval analysis
Solving in CE
FJCP 2005 14/11/2005
Complete exploration of the 1296 configurations 225 configurations solution for the requirements Time of calculus: 1h30
Results:
The ACS model
0
0,2
0,4
0,6
0,8
1
1,2
0
Aepsilon1
50
NS
FJCP 2005 14/11/2005
Aepsilon2
100
Solutions and criteria
AvdeltaD
150
AvdeltaDra
200
FJCP 2005 14/11/2005
Future works
def (C ) :
i i / Xi ∈ con
∏D
Each constraint has a satisfaction function
→A
FJCP 2005 14/11/2005
Solving by transformation (NSCSP NCSP)
Criteria = a soft constraint The fuzzy framework in order to compute the global preference
Numerical Semiring based CSP (NSCSP)
Modelling with Soft Constraint:
Decision Aiding 1/2
FJCP 2005 14/11/2005
Hierarchical CSP Weighted CSP Composed Soft CSP …
Choice of the preferences model according to the design
One best solution but only one criterion
Decision Aiding 2/2
How to have good propagation and solving?
Catalogue for component of the shelf
FJCP 2005 14/11/2005
Management of the conditional constraints
Global constraint
Design provides hard mixed integer and real non linear problems
Other …
Questions
FJCP 2005 14/11/2005
Thanks