Upload
ami
View
37
Download
2
Tags:
Embed Size (px)
DESCRIPTION
Solving Inverse Design Problems Involving Radiant Enclosures through Gradient-Based Optimization. K. J. Daun, J. R. Howell, and D. P. Morton Department of Mechanical Engineering The University of Texas at Austin. Radiant Enclosure Design Problems. - PowerPoint PPT Presentation
Citation preview
Solving Inverse Design Problems Involving Radiant Enclosures through
Gradient-Based Optimization
K. J. Daun, J. R. Howell, and D. P. MortonDepartment of Mechanical Engineering
The University of Texas at Austin
Radiant Enclosure Design Problems
• Radiant enclosure design problems are encountered in diverse industrial settings
• Examples:– Paint drying in the automotive industry– Baking ovens in the food industry– RTP of semiconductor wafers– Optimal geometry of solar collector-concentrators
General Form of the Design Problem
• Enclosure contains a heater surface and a design surface
• Goal is to find the enclosure geometry and heater settings that produces the desired qs and T over the design surface
• In order for the problem to be well-posed, only one BC can be specified on each surface!
• Can be solved using forward, inverse, or optimization design methodologies
Design SurfaceqsDS, TDS are
known
Heater Surface
qs = ?
qs = ?
qs = ? qs = ?
qs = ?
qs = ?
Forward Design Methodology
• Either TDS or qsDS is specified over each surface
• Assume TDS is specified over the design surface
• The designer
1.Guesses heater settings and enclosure geometry
2.Calculates qs over the design surface
3.Adjusts design and repeats analysis if necessary
• Requires many iterations, and the final solution quality is limited
Inverse Design Methodology
• Both TDS and qsDS are specified over the design surface, while the heater settings are unspecified
• The inverse design problem is in its explicit form, which is ill-posed
• The resulting set of ill-conditioned equations, Ax = b, must be solved using regularization methods
• Finds a solution in few iterations, but it is often non-physical• This method is only applicable to certain types of enclosure
design problems
Optimization Design Methodology
• TDS is specified over the design surface
• Define an objective function, F,
and design parameters, , that control enclosure geometry and heater settings
• Optimal design is found by minimizing F, usually through gradient-based minimization
• Much more efficient than forward “trial-and-error” method
• Easy to implement design constraints by restricting the domain of
DSN
j
targetsjDSsjDS
DS
qqN
F1
21
Gradient-Based Minimization
• Objective function local minimum is found iteratively
• At the kth step:
1. Check if k = , usually by checking if Fk < crit
2. Choose search direction, pk
3. Choose step size, k, usually by minimizing fk = Fk+kpk
4. Take a step, k+1 = k + kpk
Gradient-Based Minimization• Search direction is chosen based on 1st and 2nd-order
curvature at Fk:
• Search direction choices:
T
N
kkkk FFF
F
,,,21
112
12
212
122
122
112
2
FF
F
FFF
F
n
k
Steepest Descent pk = Fk
Newton 2Fkpk = Fk
Quasi-Newton Bkpk = Fk
Classes of Enclosure Design Problems
Design SurfaceT = Tspec, q = qspec
Heater Surface
q1 = ?
q2 = ?q3 = ?
q4 = ?q5 = ?
Design SurfaceT = Tspec, qs = qspec
q1 = ?Heater Surface
q2 = ? q3 = ? q4 = ? q5 = ?
Design SurfaceT = Tspec, q = qspec
q1 = ? q2 = ?
q3 = ? q4 = ?
Design SurfaceTt = Tspec, qt = qspec
V
q1t = ?Heater Surface
q2t = ? q3t = ? q4t = ? q5t = ?
Inverse Boundary Condition Design
• Goal is to find heat flux distribution over the heater surface
• Enclosure geometry is fixed
• Solved using both inverse and optimization design methodologies
• Inverse design methodologies by Oguma and Howell (1995), Harutunian et al. (1995)
• Optimization design methodology by Daun et al. (2003)
Design SurfaceT = Tspec, qs = qspec
qs = ?
Heater Surface
qs = ? qs = ? qs = ? qs = ?
Governing Radiosity Equations
,uqo b
a
o duuukuqu '',,' ,4 uTu
Radiosity Emitted Radiation Reflected Incident Radiation
= +
b
a
o duuukuq '',,' ,uqs ,uqo
Outgoing Energy Incoming Energy
= +
Solution of Governing Equations• Calculating F and 2F requires 1st- and 2nd-
order heat flux sensitivity, e.g.
• Radiosity sensitivities are solved by post-processing the radiosity solution
1. Solve A x = b, xi = qoi
2. Solve A x = b, xi = qoip
3. Solve A x = b, xi = 2qoipq
• Heat flux sensitivities are found from radiosity sensitivities
DSN
j p
sjtargetsjsj
DSp
qqq
N
F
1
2
Example Problem: Inverse BC Design
Design Surfaceqs
target = 2 W/m2
Eb = 1 W/m2
= 0.5
qs = 0 W/m2 qs = 0 W/m2
1
2
3
4
5
6
7
8
9 10 11 12 12 11 10 9
1
2
3
4
5
6
7
8
Heater Surface = 0.9
Example Problem: Inverse BC Design
u
0
2
4
6
8
10
12
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6
T4(u) [W/m2]
Example Problem: Inverse BC Design
0.95-3
-2.5
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
0.65 0.7 0.75 0.8 0.85 0.9 0.95
u
T4(u) [W/m2] (specified)
qs(u, ) [W/m2]
Optimization of Transient Problems• Solved using:
– Gradient-based optimization (Fedorov et al., 1998)
– Inverse design methodology (Ertürk et al., 2001)
– Nonlinear controls with regularization (Gwak and Masada, 2002)
Design SurfaceTt = Tspect, qt = qspect
V
q1t = ?
Heater Surface
q2t = ? q3t = ? q4t = ? q5t = ?
Governing Equations
b
a
o duuukuq '',,' ,uqs ,uqo
Outgoing Energy Incoming Energy
= + t
uTuucu
,
Stored Energy
u
• Enclosure surface discretized into N elements, and time domain discretized into Nt time steps
• Results in a non-linear system of equations that must be solved at each time step
,,,4iiiii ttttt bTT BA
Transient Optimization• Objective function is defined as
• Heater settings are controlled by cubic splines,
where = ttmax.
• Objective function sensitivities found analytically, through direct differentiation
t DSN
i
N
ji
targetij
DSt
tTtTNN
F1 1
2,
1
3
32
2
123
313
131,
hh
hhshq
Example Problem: Transient Optimization
T1 = 300 K1= 1
T2 = 1000 K2 = 1
=0
=1
V
Heater surface
Design surface Refractory surface
1 2 3 4 5 6 7 8
Material Thickness [cm]
Design Surface AISI 1010 Steel 0.1 0.8
Heater Surface Refractory Brick 1 0.7
Refractory Surface Refractory Brick 1 0.3
Example Problem: Transient Optimization
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
qs
T
24
1 2 3 4
5
6
7
8
Example Problem: Transient Optimization
0
0.2
0.4
0.6
0.8
1
1.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
T
Ts
x = 1
x = 0
Target
x = 0 x = 1
Design Surface
Geometric Optimization
• Both heater settings and geometry can be designed
• Only optimization methods have been used
• Deterministic optimization for diffuse-gray enclosures (infinitesimal-area method)
• Stochastic optimization for enclosures with specular surfaces (Monte Carlo method)
Design SurfaceT = Tspec, q = qspec
Heater Surface
qs
qsqsqsqs
Enclosures Containing Specular Surfaces
`
Energy leaving by emission
Energy in from surrounding surfaces
Energy in from other sources
= –
ΦΦ isi Aq ΦΦ ibii AE
N
jjijbjj AE
1
ΦΦΦ F=
• The exchange factor, Fji, is estimated using Monte Carlo analysis:
• The uncertainty in Fji induces a random error in qs and F
bj
ibjijij N
N FF ~
Stochastic Optimization• Uses Kiefer-Wolfowitz scheme based on steepest-
descent, used when an unbiased estimate of F is unavailable
• At the kth iteration,
1.Check if k =
2.Set
3.Set k = 0ka, 0 a 1
4. k+1 = k + kpk
• The gradient is estimated by central finite difference,
with
kkk FF ~~ p
k
kp
kkp
kk
p h
hFhFF
2
~~~
ee ΦΦ
Φ
10,0 bk
hh
bk
Example Problem: Geometric Optimization
Specular Surfaceqs3 = 0 W/m2
3 = 1
Specular Surfaceqs2 = 0 W/m2
2 = 1
Design SurfaceEb4 = 0 W/m2, qs
target = -1 W/m2, 4 = 1
1
2
u
Heater Surfaceqs1 = 1 W/m2
1 = 1
Example Problem: Geometric Optimization
-0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1.5
1
2
1
2
Initial Enclosure Geometry
Optimal Enclosure Geometry
Minimization Path
Example Problem: Geometric Optimization
-2
-1.8
-1.6
-1.4
-1.2
-1
-0.8
-0.6
-0.4
-0.2
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
u
qs(u
,
) [W
/m2]
qs(u, 0) (initial)
qs(u, *) (optimal)
Conclusions
• Optimization is used to design many types of radiant enclosures
• Solves the inverse design problem implicitly, through iteration
• More efficient than the forward design method, and usually produces better solutions
• More straightforward than inverse design (regularization) method, and easier to implement design constraints