Design of Electrical Rotating Machines using Interval Branch and Bound based Algorithms Fr´ ed´ eric Messine Inverse Problem of Design IBBA Algorithms Design of Machines Realizations & Conclusion Design of Electrical Rotating Machines using Interval Branch and Bound based Algorithms Fr´ ed´ eric Messine ENSEEIHT-IRIT, Team APO collaboration with GREM 3 -LAPLACE, Toulouse October 2007
67
Embed
Design of Electrical Rotating Machines using Interval ...liberti/cal07/presentations/messine-cal... · Numerical Examples of Electrical Machines Design with a Black-Box Constraint
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
1/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
Realizations &Conclusion
Design of Electrical Rotating Machines usingInterval Branch and Bound based Algorithms
Frederic Messine
ENSEEIHT-IRIT, Team APO collaboration with GREM3-LAPLACE,Toulouse
October 2007
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
2/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
Realizations &Conclusion
Collaborations with the GREM3 Team(LAPLACE-ENSEEIHT)
I Bertrand Nogarede, Professor : Director of GREM3,collaboration since 1995 (2 PhD-thesis).
I Yvan Lefevre, CR, HDR : collaboration since 2002, (1PhD student).
I Francois Pigache, MC : collaboration since 2005.
I Jean-Francois Rouchon, MC : specialist in mechanic,
I Carole Henaux, MC : specialist in electrotechnic,
I Eric Duhayon, MC : specialist of materials,
I Dominique Harribey, IR.
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
3/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
Realizations &Conclusion
Outline
Design of Actuators: an Inverse ProblemDirect and Inverse Problem of Design and FormulationsMathematical Formulations =⇒ Optimization ProblemsClassification and Principle of Optimization Methods
Interval Branch and Bound AlgorithmsPrinciple of Branch and Bound AlgorithmsInterval AnalysisMixed and Constrained Problems
Some Designs of Electrical Rotating MachinesRotating Machines with Magnetic EffectsCombinatorial Models for Electrical MachinesNumerical Examples of Electrical MachinesDesign with a Black-Box Constraint
Some Realizations and Conclusion
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
3/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
Realizations &Conclusion
Outline
Design of Actuators: an Inverse ProblemDirect and Inverse Problem of Design and FormulationsMathematical Formulations =⇒ Optimization ProblemsClassification and Principle of Optimization Methods
Interval Branch and Bound AlgorithmsPrinciple of Branch and Bound AlgorithmsInterval AnalysisMixed and Constrained Problems
Some Designs of Electrical Rotating MachinesRotating Machines with Magnetic EffectsCombinatorial Models for Electrical MachinesNumerical Examples of Electrical MachinesDesign with a Black-Box Constraint
Some Realizations and Conclusion
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
3/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
Realizations &Conclusion
Outline
Design of Actuators: an Inverse ProblemDirect and Inverse Problem of Design and FormulationsMathematical Formulations =⇒ Optimization ProblemsClassification and Principle of Optimization Methods
Interval Branch and Bound AlgorithmsPrinciple of Branch and Bound AlgorithmsInterval AnalysisMixed and Constrained Problems
Some Designs of Electrical Rotating MachinesRotating Machines with Magnetic EffectsCombinatorial Models for Electrical MachinesNumerical Examples of Electrical MachinesDesign with a Black-Box Constraint
Some Realizations and Conclusion
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
3/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
Realizations &Conclusion
Outline
Design of Actuators: an Inverse ProblemDirect and Inverse Problem of Design and FormulationsMathematical Formulations =⇒ Optimization ProblemsClassification and Principle of Optimization Methods
Interval Branch and Bound AlgorithmsPrinciple of Branch and Bound AlgorithmsInterval AnalysisMixed and Constrained Problems
Some Designs of Electrical Rotating MachinesRotating Machines with Magnetic EffectsCombinatorial Models for Electrical MachinesNumerical Examples of Electrical MachinesDesign with a Black-Box Constraint
c(x) ∈ [a, b] is a contraint =⇒ implicit (or explicit) relationsbetween the variables of the problem.
Idea: use some deduction steps for reducing the box X .
Linear case: if c(x) =n∑
i=1
aixi then:
Xk :=
[a, b]−
n∑i=1,i 6=k
aiXi
ak
∩ Xk , if ak 6= 0. (1)
where k is in {1, · · · , n} and Xi is the ith component of X .
Non-linear case: Idea (E. Hansen): one linearizes using T1 (orT2). Then one solves a linear system with interval coefficients.Other Idea: construction of the calculus tree and propagation.
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
15/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Branch and BoundPrinciple
Interval Analysis
Mixed andConstrainedProblems
Design ofMachines
Realizations &Conclusion
Propagation Techniques
c(x) ∈ [a, b] is a contraint =⇒ implicit (or explicit) relationsbetween the variables of the problem.
Idea: use some deduction steps for reducing the box X .
Linear case: if c(x) =n∑
i=1
aixi then:
Xk :=
[a, b]−
n∑i=1,i 6=k
aiXi
ak
∩ Xk , if ak 6= 0. (1)
where k is in {1, · · · , n} and Xi is the ith component of X .
Non-linear case: Idea (E. Hansen): one linearizes using T1 (orT2). Then one solves a linear system with interval coefficients.Other Idea: construction of the calculus tree and propagation.
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
15/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Branch and BoundPrinciple
Interval Analysis
Mixed andConstrainedProblems
Design ofMachines
Realizations &Conclusion
Propagation Techniques
c(x) ∈ [a, b] is a contraint =⇒ implicit (or explicit) relationsbetween the variables of the problem.
Idea: use some deduction steps for reducing the box X .
Linear case: if c(x) =n∑
i=1
aixi then:
Xk :=
[a, b]−
n∑i=1,i 6=k
aiXi
ak
∩ Xk , if ak 6= 0. (1)
where k is in {1, · · · , n} and Xi is the ith component of X .
Non-linear case: Idea (E. Hansen): one linearizes using T1 (orT2). Then one solves a linear system with interval coefficients.Other Idea: construction of the calculus tree and propagation.
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
16/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Branch and BoundPrinciple
Interval Analysis
Mixed andConstrainedProblems
Design ofMachines
Realizations &Conclusion
Example of Propagation Technique based on theCalculus Tree
Let c(x) = 2x3x2 + x1 and
c(x) = 3
where xi ∈ [1, 3] for all i ∈ {1, 2, 3}.
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
16/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Branch and BoundPrinciple
Interval Analysis
Mixed andConstrainedProblems
Design ofMachines
Realizations &Conclusion
Example of Propagation Technique based on theCalculus Tree
Let c(x) = 2x3x2 + x1 and
c(x) = 3
where xi ∈ [1, 3] for all i ∈ {1, 2, 3}.The propagation is:
X
X
X3
*
+
*
2
1
2
=[1,3]
[2,6]
[2,18]
[3,21]
=[1,3]
=[1,3]
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
16/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Branch and BoundPrinciple
Interval Analysis
Mixed andConstrainedProblems
Design ofMachines
Realizations &Conclusion
Example of Propagation Technique based on theCalculus Tree
Let c(x) = 2x3x2 + x1 and
c(x) = 3
where xi ∈ [1, 3] for all i ∈ {1, 2, 3}.The propagation is:
X
X
X3
*
+
*
2
1
2
=[1,3]
[2,6]
[2,18]
[3,21]
=[1,3]
=[1,3] X
X
X3
*
+
*1
2
2
[1,1]
3
3-[ 1,3] =[0,2]
[0,2] =[0,2] [1,3]
3-[2,18] =[-16,1]
[0,2] = [0,1] [2,6]
[0,2] =[0,1] 2
[1,1]
[1,1]
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
17/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Branch and BoundPrinciple
Interval Analysis
Mixed andConstrainedProblems
Design ofMachines
Realizations &Conclusion
Algorithms for Mixed Problems
Continuous variables: real variables (dimensions of anelectrical machines such as diameter).
Discrete variables: integer (the number of pair of poles of amachine), boolean (machine with or without slot), categoricalvariable (which kind of magnet is used).
For integer and boolean variables =⇒ relaxation forcomputing bounds + particular bisection technique andpropagation.
For categorical variables =⇒ we introduce 4 particularalgorithms with propagation and retro-propagation +properties about the bisection techniques.
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
17/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Branch and BoundPrinciple
Interval Analysis
Mixed andConstrainedProblems
Design ofMachines
Realizations &Conclusion
Algorithms for Mixed Problems
Continuous variables: real variables (dimensions of anelectrical machines such as diameter).
Discrete variables: integer (the number of pair of poles of amachine), boolean (machine with or without slot), categoricalvariable (which kind of magnet is used).
For integer and boolean variables =⇒ relaxation forcomputing bounds + particular bisection technique andpropagation.
For categorical variables =⇒ we introduce 4 particularalgorithms with propagation and retro-propagation +properties about the bisection techniques.
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
17/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Branch and BoundPrinciple
Interval Analysis
Mixed andConstrainedProblems
Design ofMachines
Realizations &Conclusion
Algorithms for Mixed Problems
Continuous variables: real variables (dimensions of anelectrical machines such as diameter).
Discrete variables: integer (the number of pair of poles of amachine), boolean (machine with or without slot), categoricalvariable (which kind of magnet is used).
For integer and boolean variables =⇒ relaxation forcomputing bounds + particular bisection technique andpropagation.
For categorical variables =⇒ we introduce 4 particularalgorithms with propagation and retro-propagation +properties about the bisection techniques.
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
18/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
Rotating Machines with Magnetic Effects
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
19/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
Example for the Dimensioning of an ElectricalMotor
Electrical Slotless Rotating Machine with Permanent Magnet:
2
E
pβπ
laeC
C
D
I IBBA standard (defined by Ratschek and Rokne 1988)−→ 1h35,
I IBBA + propagation due to E. Hansen −→ 41.5s,
I IBBA + propagation with the calculus tree −→ 0.5s.
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
19/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
Example for the Dimensioning of an ElectricalMotor
Electrical Slotless Rotating Machine with Permanent Magnet:
2
E
pβπ
laeC
C
D
I IBBA standard (defined by Ratschek and Rokne 1988)−→ 1h35,
I IBBA + propagation due to E. Hansen −→ 41.5s,
I IBBA + propagation with the calculus tree −→ 0.5s.
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
19/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
Example for the Dimensioning of an ElectricalMotor
Electrical Slotless Rotating Machine with Permanent Magnet:
2
E
pβπ
laeC
C
D
I IBBA standard (defined by Ratschek and Rokne 1988)−→ 1h35,
I IBBA + propagation due to E. Hansen −→ 41.5s,
I IBBA + propagation with the calculus tree −→ 0.5s.
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
20/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
Combination of Different Rotating ElectricalMachines
b
b b
Slot
With
Slot
d
C
C D__2
D__2
laE
a
g
βπ__p
re
f
s
Internal RotorInverse Rotor
Rectangular or Sinusoidal Waveform
Without
Figure: 4 structures possible machines ×2 modes (rectangular orsinusoidal waveform).
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
21/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
Discrete Variables for Modeling ElectricalMachines
1. br = 1 for machines with an internal rotoric configurationand br = 0 for an external one,
2. be = 1 for machines with slots or be = 0 slotlessmachines,
3. bf = 1 represents rectangular waveform or bf = 0 for asinusoıdale one.
3 boolean variables to represent 8 possible structures + 2categorical variables.
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
22/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
Combinatorial Models for Electrical Machines
Γem = kΓD [D + (1− be)(2br − 1)E ] LBeKS,
KS = krEj
(be
a
a + d+ (1− be)
),
kΓ =π
2
[bf [1−Kf ]
√β + (1− bf )
√2
2sin(β
π
2)
],
Kf = 1.5pβ
[E + g
D
](1− be).bf ,
Be =2J(σm)la
(2br − 1)D ln[
D+2E(2br−1)(1−be)D−2(2br−1)[la+g ]
] 1
kc,
kc =1
1− be
[Nea2
5πD.g+πD.a
] ,
. . .
Generally, the torque Γem is fixed =⇒ a strong equalityconstraint.
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
23/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
Examples of 4 optimal machines with magneticaleffects
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
24/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
Multicriteria Optimization
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
25/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
Analytical versus Numerical Models
Schedule of conditions =⇒ a fixed torque.
Optimal solutions satisfy the equality constraint about 1%.
Computation of the torque using finite elements methods (bya numerical model); software EFCAD (from the LEEI) orANSYS.
EFCAD and ANSYS are too general =⇒ NUMT
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
25/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
Analytical versus Numerical Models
Schedule of conditions =⇒ a fixed torque.
Optimal solutions satisfy the equality constraint about 1%.
Computation of the torque using finite elements methods (bya numerical model); software EFCAD (from the LEEI) orANSYS.
EFCAD and ANSYS are too general =⇒ NUMT
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
25/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
Analytical versus Numerical Models
Schedule of conditions =⇒ a fixed torque.
Optimal solutions satisfy the equality constraint about 1%.
Computation of the torque using finite elements methods (bya numerical model); software EFCAD (from the LEEI) orANSYS.
EFCAD and ANSYS are too general =⇒ NUMT
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
25/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
Analytical versus Numerical Models
Schedule of conditions =⇒ a fixed torque.
Optimal solutions satisfy the equality constraint about 1%.
Computation of the torque using finite elements methods (bya numerical model); software EFCAD (from the LEEI) orANSYS.
EFCAD and ANSYS are too general =⇒ NUMT
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
26/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
Direct and Inverse Problem of Design
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
27/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
Numerical Validations: NUMT Algorithm
Figure: Draw 2 optimal solutions (min mass and min multicriteria).
Figure: Mesh 2 above machines
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
28/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
Numerical Validations: NUMT Algorithm
−pi −3*pi/4 −pi/2 −pi/4 0 pi/4 pi/2 3*pi/4 pi−10
−8
−6
−4
−2
0
2
4
6
8
10
Angle de calage : −π ≤ Ψ ≤ π
Cou
ple
(N.m
)Min MasseMin VolMin Multi
Figure: Torque of 3 solutions and design of teeth of the slot.
Using Triangle and EFCAD.Name of the Software: NUMT.
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
29/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
Discussions
Analytical Value 6= Numerical Value
Generally about 10% !
I If the gap is less than 3% our solution is validated.
I Else =⇒ modifications of the solution until the numericalvalue is correct.
Optimize with a black-box constraint:min
x∈Rnr ,z∈Nne ,
σ∈Qnc
i=1Ki ,b∈Bnb
f (x , z , σ, b)
gi (x , z , σ, b) ≤ 0 ∀i ∈ {1, . . . , ng}hj(x , z , σ, b) = 0 ∀j ∈ {1, . . . , nh − 1}NUMT (x , z , σ, b) = Γem
Extension of IBBA ?
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
29/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
Discussions
Analytical Value 6= Numerical Value
Generally about 10% !
I If the gap is less than 3% our solution is validated.
I Else =⇒ modifications of the solution until the numericalvalue is correct.
Optimize with a black-box constraint:min
x∈Rnr ,z∈Nne ,
σ∈Qnc
i=1Ki ,b∈Bnb
f (x , z , σ, b)
gi (x , z , σ, b) ≤ 0 ∀i ∈ {1, . . . , ng}hj(x , z , σ, b) = 0 ∀j ∈ {1, . . . , nh − 1}NUMT (x , z , σ, b) = Γem
Extension of IBBA ?
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
29/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
Discussions
Analytical Value 6= Numerical Value
Generally about 10% !
I If the gap is less than 3% our solution is validated.
I Else =⇒ modifications of the solution until the numericalvalue is correct.
Optimize with a black-box constraint:min
x∈Rnr ,z∈Nne ,
σ∈Qnc
i=1Ki ,b∈Bnb
f (x , z , σ, b)
gi (x , z , σ, b) ≤ 0 ∀i ∈ {1, . . . , ng}hj(x , z , σ, b) = 0 ∀j ∈ {1, . . . , nh − 1}NUMT (x , z , σ, b) = Γem
Extension of IBBA ?
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
29/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
Discussions
Analytical Value 6= Numerical Value
Generally about 10% !
I If the gap is less than 3% our solution is validated.
I Else =⇒ modifications of the solution until the numericalvalue is correct.
Optimize with a black-box constraint:min
x∈Rnr ,z∈Nne ,
σ∈Qnc
i=1Ki ,b∈Bnb
f (x , z , σ, b)
gi (x , z , σ, b) ≤ 0 ∀i ∈ {1, . . . , ng}hj(x , z , σ, b) = 0 ∀j ∈ {1, . . . , nh − 1}NUMT (x , z , σ, b) = Γem
Extension of IBBA ?
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
29/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
Discussions
Analytical Value 6= Numerical Value
Generally about 10% !
I If the gap is less than 3% our solution is validated.
I Else =⇒ modifications of the solution until the numericalvalue is correct.
Optimize with a black-box constraint:min
x∈Rnr ,z∈Nne ,
σ∈Qnc
i=1Ki ,b∈Bnb
f (x , z , σ, b)
gi (x , z , σ, b) ≤ 0 ∀i ∈ {1, . . . , ng}hj(x , z , σ, b) = 0 ∀j ∈ {1, . . . , nh − 1}NUMT (x , z , σ, b) = Γem
Extension of IBBA ?
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
30/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
IBBA+NUMT
Idea: Define a zone where the numerical solution is searched.Analytical model + tolerance about 10% =⇒ zone.In this zone, all the solutions satisfies thatNUMT (x∗, z∗, σ∗, b∗) = Γem about 1%.
minx∈Rnr ,z∈Nne ,
σ∈Qnc
i=1Ki ,b∈Bnb
f (x , z , σ, b)
gi (x , z , σ, b) ≤ 0 ∀i ∈ {1, . . . , ng}hj(x , z , σ, b) = 0 ∀j ∈ {1, . . . , nh − 1}0.9× Γem ≤ Γ(x , z , σ, b) ≤ 1.1× Γem
NUMT (x , z , σ, b) = Γem
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
30/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
IBBA+NUMT
Idea: Define a zone where the numerical solution is searched.Analytical model + tolerance about 10% =⇒ zone.In this zone, all the solutions satisfies thatNUMT (x∗, z∗, σ∗, b∗) = Γem about 1%.
minx∈Rnr ,z∈Nne ,
σ∈Qnc
i=1Ki ,b∈Bnb
f (x , z , σ, b)
gi (x , z , σ, b) ≤ 0 ∀i ∈ {1, . . . , ng}hj(x , z , σ, b) = 0 ∀j ∈ {1, . . . , nh − 1}0.9× Γem ≤ Γ(x , z , σ, b) ≤ 1.1× Γem
NUMT (x , z , σ, b) = Γem
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
30/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
IBBA+NUMT
Idea: Define a zone where the numerical solution is searched.Analytical model + tolerance about 10% =⇒ zone.In this zone, all the solutions satisfies thatNUMT (x∗, z∗, σ∗, b∗) = Γem about 1%.
minx∈Rnr ,z∈Nne ,
σ∈Qnc
i=1Ki ,b∈Bnb
f (x , z , σ, b)
gi (x , z , σ, b) ≤ 0 ∀i ∈ {1, . . . , ng}hj(x , z , σ, b) = 0 ∀j ∈ {1, . . . , nh − 1}0.9× Γem ≤ Γ(x , z , σ, b) ≤ 1.1× Γem
NUMT (x , z , σ, b) = Γem
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
30/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
IBBA+NUMT
Idea: Define a zone where the numerical solution is searched.Analytical model + tolerance about 10% =⇒ zone.In this zone, all the solutions satisfies thatNUMT (x∗, z∗, σ∗, b∗) = Γem about 1%.
minx∈Rnr ,z∈Nne ,
σ∈Qnc
i=1Ki ,b∈Bnb
f (x , z , σ, b)
gi (x , z , σ, b) ≤ 0 ∀i ∈ {1, . . . , ng}hj(x , z , σ, b) = 0 ∀j ∈ {1, . . . , nh − 1}0.9× Γem ≤ Γ(x , z , σ, b) ≤ 1.1× Γem
NUMT (x , z , σ, b) = Γem
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
31/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
Algorithm IBBA+NUMT
1. Set X := the initial hypercube.
2. Set f := +∞ and set L := (+∞, X ).3. Extract from L the lowest lower bound.
4. Bisect the considered box chosen by its midpoint: V1, V2.
5. For j:=1 to 2 do
5.1 Compute vj := lb(f ,Vj).5.2 Compute all the lower and upper bounds of all
the analytical constraints on Vj + somededuction steps.
5.3 if f ≥ vj and no analytical constraint isunsatisfied then
I insert (vj , Vj) in L.I set m the midpoint of Vj
I if m satisfies all the analytical constraints
and then if the numerical constraint
NUMT(x , z , σ, b) = Γ is also satisfied then
f := min(f, f (m)).I if f is changed then remove from L all (z , Z)
where z > f and set y := m.6. If f− min
(z,Z)∈Lz < ε (where z = lb(f , Z)) then STOP.
Else GoTo Step 4.
Design ofElectricalRotating
Machines usingInterval Branch
and Boundbased
Algorithms
32/ 37
Frederic Messine
Inverse Problemof Design
IBBA Algorithms
Design ofMachines
MAPSE
CombinatorialModels
NumericalExemples
Extension
Realizations &Conclusion
Numerical Example 1 with IBBA+NUMT
Parameters Min Vg
Name Bounds Unit IBBA IBBA+NUMT
D [0.01, 0.3] m 0.1331 0.1310L [0.01, 0.3] m 0.0474 0.0497la [0.003, 0.01] m 0.0047 0.0047E [0.005, 0.03] m 0.0074 0.0074C [0.003, 0.02] m 0.0049 0.0049β [0.7, 0.9] 0.89 0.89kd [0.4, 0.6] 0.5043 0.5043p [[3, 10]] 8 8m {1, 2} 1 1br {0, 1} 0 0