SlideShare a Scribd company logo
1 of 19
Adaptive Switching of Variable-Fidelity Models in
Population-based Optimization Algorithms
Ali Mehmani*, Souma Chowdhury#, and Achille Messac#
* Syracuse University, Department of Mechanical and Aerospace Engineering
# Mississippi State University, Department of Aerospace Engineering
The AIAAAviation and Aeronautics Forum and Exposition
June 22-26, 2015
Dallas, TX
Making Heuristic Optimization more Viable?
2
• Population-based heuristic algorithms have been successfully
applied to diverse areas of science and engineering over the past
three decades – a core tool in designing complex systems.
• Challenge: Although proven to be effective in solving highly
non-linear problems, they often demand a high number of
function calls – computationally “too expensive” – for
complex problems.
• Need: Control the computational cost per function call, while
preserving the fidelity of the final optima.
Variable fidelity models + Model management strategies
Variable Fidelity Optimization (VFO)
3
• Variable fidelity models refer to models with different levels
of fidelity (low, medium, and high fidelity models)
• Model management strategies are techniques used to select
different models during the optimization process.
Medium fidelityLow fidelity High fidelity
Park model
(5 sec.)
Dynamic model
(8 mins)
CFD model
(30 hrs)
Model Management/Switching
Optimizer
Presentation Outline
4
• VFO: Motivation and Research Objectives
• VFO with Adaptive Model Switching (AMS)
 Methodology
 Quantifying Model Uncertainties
• Numerical Case Studies
 Aerodynamic shape optimization of 2D airfoil
 Shape Optimization of a Cantilever Composite Beam
• Concluding Remarks
• Trust region strategy Alexandrov et al. (1998) and Rodriguez et al.(2001)
low-fidelity model is used to define a sub-region of global optimum, then the high-
fidelity model is used to improve the low-fidelity model.
• Efficient Global Optimization Jones et. al. (1998) and Viana and Haftka (2013)
The accuracy of the low fidelity model (surrogate) is improved by running high
fidelity model in the regions of (i) global optimum and (ii) high predicted error.
• Individual-based evolution control Jin, et al. (2011)
Part of the individuals in the population are chosen and evaluated using high fidelity
model.
• Generation-based evolution control Jin, et al. (2011)
The whole population in specific iteration are evaluated using high fidelity model
Variable fidelity optimization (VFO)
5
Research Motivation
6
• Several of the existing VFO strategies are found to be defined
for specific types of low fidelity model (e.g., EGO works
primarily for Gaussian process-based surrogate models),
thereby limiting their applicability.
• Existing techniques generally consider the combination of only
two models of different fidelities (e.g., Trust-region methods,
and individual- and generation-based techniques).
Research Objectives
7
• Investigate an adaptive and model-independent strategy for managing
multiple variable-fidelity models (called Adaptive Model Switching or AMS),
with the objective to reduce the computational cost while converging to an
optimum with high fidelity function estimates.
• Implement the new adaptive variable fidelity optimization method in Particle
Swarm Optimization, and test its effectiveness through two practical case
studies.
Attributes:
VFO with Adaptive Model Switching (AMS)
8
• The important question is when and where to use models with different
levels of fidelity and cost.
Computationally expensive while wasting resources
Switching to higher fidelity models too early
Mislead the search process to suboptimal (or infeasible) regions
Switching to a higher fidelity model too late
 The AMS metric is a hypothesis testing that is defined by a comparison between
(I) the distribution of the relative fitness function improvement, and
(II) the distribution of the error associated with the model.
Rejection of the test; Don’t Switch Model Acceptance of the test; Switch Model
Fitness Func. Improvement (KDE)
Distribution of Model Error (LogN)
Adaptive Model Switching (AMS)
9
Adaptive Model Switching (AMS)
10
• represents a quantile function of a distribution;
• The p-quantile, for a given distribution function, , is defined by
• is an Indicator of Conservativeness (IoC).
• The IoC for the low fidelity model with an error distribution is defined as
the probability of the model error to be less than h
• The IoC is based on user preferences, and could be guided by the desired trade-
off between reliability and computational cost in the AMS-based VFO.
• Here, we defined pcr = 0.3.
AMS: Uncertainty Quantification
• The uncertainties associated with surrogate
models are determined using an advanced
surrogate error estimation method, called
Predictive Estimation of Model Fidelity *.
* Published in Struct. Multidiscip. Optim. (2015), Presented in AIAA SDM (2013)
• In the case of physics-based low fidelity (PLF)
models, the uncertainty in their output is
quantified through an inverse assessment, using
the same DOE as used for the surrogate model.
• Kernel Density Estimation (KDE) is adopted to model the distribution of the
relative improvement in the fitness function over consecutive t iterations.
11
Variable Fidelity Optimization: Airfoil Design
14 Mehmani et al.
Fig. 5 Design variables gov-
erning the geometry of the
airfoil
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-0.04
-0.02
0.00
0.02
0.04
0.06
0.08
0.10
X
Y
x1
x2 x3
Table1 Design variables in airfoil optimization problem
Description Notation Lower limit Upper limit
Distance between themiddleof suction sideand horizontal axis x1 -0.01 0.01
Distance between themiddleof pressure sideand horizontal axis x2 -0.01 0.01
Distance between thetrailing edgeand horizontal axis x3 -0.01 0.01
Incidence angle x4 0◦
10◦
3.1.1 Aerodynamic modelswith different level of fidelity
To develop a high fidelity aerodynamic model for determining CL and CD (MA
HF),
the commercial Finite Volume Method package, FLUENT, is used. The Reynolds-
Shape variable angle of incidence
* The Mixed-Discrete Particle Swarm Optimization (MDPSO)[1] is used. [1] Chowdhury and Messac (2012)
Wortmann FX60.126
12
Physics-based low
fidelity model (PLF)
The fluid is steady, incompressible, and irrotational.
Computational time = 30 seconds
Surrogate model
(SM)
Kriging-Gaussian-30 sample points
Computational time = less than 0.1 second
Tuned low fidelity
model (TLF)
This model is constructed using the Multiplicative approach, as given by
onds, respectively (i.e., an order of magnitude apart). The pressure field around
the airfoil for the low and high fidelity aerodynamic models at a baseline design
(x1 = 0, x2 = 0, x3 = 0, andx4 = 5◦
) areillustrated in Fig. 7.
1.0171e+05
1.0054e+0
1.0066e+0
1.0077e+0
1.0089e+0
1.0101e+0
1.0112e+0
1.0124e+0
1.0136e+0
1.0147e+0
1.0159e+0
10.50
X
(a) High fidelity model
−0.5 0 0.5 1 1.5
−1
−0.5
0
0.5
1
x
y
1.0124
1.0126
1.0128
1.013
1.0132
1.0134
1.0136
1.0138
x 10
5
(b) Low fidelity physics-based model
Fig. 7 Pressure field around theairfoil at abaselinedesign
The third model is a surrogate model (MA
SM) constructed using a DoE of high
fidelity evaluation involving 30 sample points. The fourth model is a tuned low
fidelity model (MA
TLF). In this article, the tuned low fidelity model is constructed
using theMultiplicativeapproach, asgivenby
˜F(x,a) = f(x) × C(x) (21)
where ˜F isatunedlow fidelity model; f(x) isalow fidelity model;C(x) isanexplicit
tuning surrogateconstructedusingthehighfidelity samples, asshown below:
C(x) =
CL
CD
|HF
CL
CD
|PLF
(22)
C(x) : explicit tuning surrogate, f(x): HF model
High fidelity
model (HF)
The FLUENT is used for solving RANS eq.
Computational time = 300 seconds
Computational time ~ 30 seconds
Airfoil Design : Results
13
0 0.05 0.1 0.15 0.2 0.25
0
2
4
6
8
10
12
14
16
18
20
Error
PDF
0 0.05 0.1 0.15 0.2 0.25
0
5
10
15
20
25
Error
PDF
0 0.05 0.1 0.15 0.2 0.25
0
5
10
15
20
25
30
35
40
45
50
Error
PDF
Distribution of the fitness function improvements in different iterations of the airfoil optimization
0 0.05 0.1 0.15 0.2 0.25
0
5
10
15
20
25
30
35
40
Error
PDF
5th Iteration 15th Iteration
18th Iteration20th Iteration
F.F. improvement
Surrogate Error
TLF Error
Computational Time Function Evaluation
provides the best optimum value that is
5% better that the next best value
185% reduction in computing time
14
13th 19th
Airfoil Design : Results
Variable Fidelity Optimization: Cantilever beam design
Minimizing the maximum deflection of a
cantilever composite beam.
f
Weight density of epoxy resin, ρm [N/ mm3]
Thebeam optimization problemisdefined as
Minimize:
δmax
δ0
, [δ0] = 12.9
subject to
W/ W0 ≤ 1, [W0] = 2.9E4
σmax/ σ0 ≤ 1, [σ0] = 200
x4
2
1.2E6x1
≤ 1
xmin
i ≤ xi ≤ xmax
i , i = 1,2,3
In the optimization formulation, the inequality constra
arerelatedto theallowableweight, themaximumstress, a
on thebeam design (depth ≤ 10× width). Theweight an
given by
W = AρL =
12I
h2
× (12+ 5.2νf)10− 6
× L =
x1
x2
σmax =
q0L2
h
8I
=
1E6x2
8x1
Themodelsused to estimatethemaximum deflection,
3.2.1 Structural modelswith different levelsof fidelity
geometric restriction on the beam design
(depth ≤ 10 × width)
 Minimizing the maximum deflection of a cantilever composite
beam.*
 The design variables are:
(i) the second moment of area (x1),
(ii) the depth of the beam (x2), and
(iii) the fiber volume fraction (x3).
 optimization problem is
15*Zadeh and Toropov, SMO 2009
Cantilever beam design : Model Choices
Physics-based
low fidelity model (PLF)
Surrogate model (SM)
Tuned low fidelity
model (TLF)
High fidelity model (HF)
The models used to estimate the maximum deflection
The PLF Finite Element model is constructed using 2 beam elements in ANSYS
Computational time = 3.30 [Sec]
The HF Finite Element model is constructed using 1000 beam elements in ANSYS.
Computational time = 5.5 [Sec].
Surrogate model is constructed using Kriging with Gaussian correlation function.
Computational time less than 0.3 [Sec]
.
This model is constructed using the Multiplicative approach, as given by
Computational time less than 3.33 [Sec]
.
1.0054e+0
1.0066e+0
1.0077e+0
1.0089e+0
1.0101e+0
1.0112e+0
1.0124e+0
1.0136e+0
10.50
X
(a) High fidelity model
−0.5 0 0.5 1 1.5
−1
−0.5
0
x
y
(b) Low fidelity physics-based mod
Fig. 7 Pressure field around theairfoil at abaselinedesign
The third model is a surrogate model (MA
SM) constructed using a DoE of
fidelity evaluation involving 30 sample points. The fourth model is a tuned
fidelity model (MA
TLF). In this article, the tuned low fidelity model is constru
using theMultiplicativeapproach, asgivenby
˜F(x,a) = f(x) × C(x)
where ˜F isatunedlow fidelity model; f(x) isalow fidelity model;C(x) isanex
tuning surrogateconstructedusingthehighfidelity samples, asshown below:
C(x) =
CL
CD
|HF
CL
CD
|PLF
whereCL andCD arerespectively thelift and drag coefficients.
The surrogate model (MA
SM) and the surrogate component of the tuned lo
delity model (MA
TLF) arebothconstructedusingKrigingwithaGaussiancorrel
function.14,37 Kriging isan interpolating method that is widely used for repre
ing irregular data. Under theKriging approach, thezero-order polynomial fun
isusedasaregressionmodel. InthisarticletheOptimal LatinHypercubeisado
to determinethelocationsof thesamplepoints. ThePEMF method isthen ap
22
C(x) =
δmax|HF
δmax|PLF
The distribution of the error in the tuned low fidelity
rogate model (SM) are estimated using PEMF (Section
in Figs. 13(a) and 13(c), respectively. The distribution o
based low fidelity model (PLF) is estimated using the in
by leveragingthesame30 high fidelity samplesthat were
and SM; thePLF error distribution isshown in Fig. 13(b)
!
! "%
! "(
! ")
! "*
$
$"%
$"(
$")
$"*
%
&'$!
#
./0
pcr = 0.3
QP = 0.0001
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
x 10
4
PDF
pcr = 0.3
QP = 0.097
0.
0
0.
0
0.
PDF
UQ: Using MC simulationsUQ: Using PEMFUQ: Using PEMF
Tuned LF Model[ Surrogate Model Low-fidelity Model
16
Cantilever beam design : Results
Computational Time Function Evaluation
33% lower computational
expense compared to PSO-TLF
119% lower computational expense
compared to PSO-HF
17
Concluding Remarks
• A novel model management technique, called Adaptive Model Switching,
was developed for variable fidelity optimization using population-based
algorithms.
• The method seeks to provide reliable high-fidelity optimum designs at a
reasonable computational expense, by leveraging multiple models.
• Model switching is guided by the comparison between:
1. Error distribution of the current model
2. Distribution of the recent fitness function improvement over the entire
population of candidate designs
• AMS was observed to substantially superior than purely low-fidelity or
purely high-fidelity optimizations, in terms of the balance between “quality
of the optimum” and “computational efficiency”.
• A more intuitive definition of the Indicator of Conservativeness (IoC) as a
function of desired trade-offs between computational expense and reliability
of optimum would further extend the practical applicability of this algorithm. 18
Thank you
Questions
and
Comments
19

More Related Content

What's hot

COSMOS1_Scitech_2014_Ali
COSMOS1_Scitech_2014_AliCOSMOS1_Scitech_2014_Ali
COSMOS1_Scitech_2014_Ali
MDO_Lab
 
MOWF_WCSMO_2013_Weiyang
MOWF_WCSMO_2013_WeiyangMOWF_WCSMO_2013_Weiyang
MOWF_WCSMO_2013_Weiyang
MDO_Lab
 
Adaptive response surface by kriging using pilot points for structural reliab...
Adaptive response surface by kriging using pilot points for structural reliab...Adaptive response surface by kriging using pilot points for structural reliab...
Adaptive response surface by kriging using pilot points for structural reliab...
IOSR Journals
 
ASS_SDM2012_Ali
ASS_SDM2012_AliASS_SDM2012_Ali
ASS_SDM2012_Ali
MDO_Lab
 
Lecture7 cross validation
Lecture7 cross validationLecture7 cross validation
Lecture7 cross validation
Stéphane Canu
 

What's hot (16)

A parsimonious SVM model selection criterion for classification of real-world ...
A parsimonious SVM model selection criterion for classification of real-world ...A parsimonious SVM model selection criterion for classification of real-world ...
A parsimonious SVM model selection criterion for classification of real-world ...
 
COSMOS1_Scitech_2014_Ali
COSMOS1_Scitech_2014_AliCOSMOS1_Scitech_2014_Ali
COSMOS1_Scitech_2014_Ali
 
DSUS_MAO_2012_Jie
DSUS_MAO_2012_JieDSUS_MAO_2012_Jie
DSUS_MAO_2012_Jie
 
MOWF_WCSMO_2013_Weiyang
MOWF_WCSMO_2013_WeiyangMOWF_WCSMO_2013_Weiyang
MOWF_WCSMO_2013_Weiyang
 
Adaptive response surface by kriging using pilot points for structural reliab...
Adaptive response surface by kriging using pilot points for structural reliab...Adaptive response surface by kriging using pilot points for structural reliab...
Adaptive response surface by kriging using pilot points for structural reliab...
 
multi criteria decision making
multi criteria decision makingmulti criteria decision making
multi criteria decision making
 
Online learning in estimation of distribution algorithms for dynamic environm...
Online learning in estimation of distribution algorithms for dynamic environm...Online learning in estimation of distribution algorithms for dynamic environm...
Online learning in estimation of distribution algorithms for dynamic environm...
 
A COMPARATIVE STUDY OF DIFFERENT INTEGRATED MULTIPLE CRITERIA DECISION MAKING...
A COMPARATIVE STUDY OF DIFFERENT INTEGRATED MULTIPLE CRITERIA DECISION MAKING...A COMPARATIVE STUDY OF DIFFERENT INTEGRATED MULTIPLE CRITERIA DECISION MAKING...
A COMPARATIVE STUDY OF DIFFERENT INTEGRATED MULTIPLE CRITERIA DECISION MAKING...
 
Testing the performance of the power law process model considering the use of...
Testing the performance of the power law process model considering the use of...Testing the performance of the power law process model considering the use of...
Testing the performance of the power law process model considering the use of...
 
ASS_SDM2012_Ali
ASS_SDM2012_AliASS_SDM2012_Ali
ASS_SDM2012_Ali
 
A REVIEW ON OPTIMIZATION OF LEAST SQUARES SUPPORT VECTOR MACHINE FOR TIME SER...
A REVIEW ON OPTIMIZATION OF LEAST SQUARES SUPPORT VECTOR MACHINE FOR TIME SER...A REVIEW ON OPTIMIZATION OF LEAST SQUARES SUPPORT VECTOR MACHINE FOR TIME SER...
A REVIEW ON OPTIMIZATION OF LEAST SQUARES SUPPORT VECTOR MACHINE FOR TIME SER...
 
Parametric estimation of construction cost using combined bootstrap and regre...
Parametric estimation of construction cost using combined bootstrap and regre...Parametric estimation of construction cost using combined bootstrap and regre...
Parametric estimation of construction cost using combined bootstrap and regre...
 
DSUS_SDM2012_Jie
DSUS_SDM2012_JieDSUS_SDM2012_Jie
DSUS_SDM2012_Jie
 
Lecture7 cross validation
Lecture7 cross validationLecture7 cross validation
Lecture7 cross validation
 
Multi criteria decision making
Multi criteria decision makingMulti criteria decision making
Multi criteria decision making
 
Special Double Sampling Plan for truncated life tests based on the Marshall-O...
Special Double Sampling Plan for truncated life tests based on the Marshall-O...Special Double Sampling Plan for truncated life tests based on the Marshall-O...
Special Double Sampling Plan for truncated life tests based on the Marshall-O...
 

Viewers also liked

AIAA-SDM-WFLO-2012
AIAA-SDM-WFLO-2012AIAA-SDM-WFLO-2012
AIAA-SDM-WFLO-2012
OptiModel
 
AIAA-MAO-WFLO-2012
AIAA-MAO-WFLO-2012AIAA-MAO-WFLO-2012
AIAA-MAO-WFLO-2012
OptiModel
 
AIAA-Aviation-VariableFidelity-2014-Mehmani
AIAA-Aviation-VariableFidelity-2014-MehmaniAIAA-Aviation-VariableFidelity-2014-Mehmani
AIAA-Aviation-VariableFidelity-2014-Mehmani
OptiModel
 
AIAA-SDM-PEMF-2013
AIAA-SDM-PEMF-2013AIAA-SDM-PEMF-2013
AIAA-SDM-PEMF-2013
OptiModel
 
ASME-IDETC-Sensitivity-2013
ASME-IDETC-Sensitivity-2013ASME-IDETC-Sensitivity-2013
ASME-IDETC-Sensitivity-2013
OptiModel
 
WCSMO-Wind-2013-Tong
WCSMO-Wind-2013-TongWCSMO-Wind-2013-Tong
WCSMO-Wind-2013-Tong
OptiModel
 
AIAA-Aviation-Vidmap-2014
AIAA-Aviation-Vidmap-2014AIAA-Aviation-Vidmap-2014
AIAA-Aviation-Vidmap-2014
OptiModel
 
WCSMO-ModelSelection-2013
WCSMO-ModelSelection-2013WCSMO-ModelSelection-2013
WCSMO-ModelSelection-2013
OptiModel
 
AIAA-MAO-RegionalError-2012
AIAA-MAO-RegionalError-2012AIAA-MAO-RegionalError-2012
AIAA-MAO-RegionalError-2012
OptiModel
 
WCSMO-Vidmap-2015
WCSMO-Vidmap-2015WCSMO-Vidmap-2015
WCSMO-Vidmap-2015
OptiModel
 
AIAA-SDM-SequentialSampling-2012
AIAA-SDM-SequentialSampling-2012AIAA-SDM-SequentialSampling-2012
AIAA-SDM-SequentialSampling-2012
OptiModel
 

Viewers also liked (12)

AIAA-MAO-DSUS-2012
AIAA-MAO-DSUS-2012AIAA-MAO-DSUS-2012
AIAA-MAO-DSUS-2012
 
AIAA-SDM-WFLO-2012
AIAA-SDM-WFLO-2012AIAA-SDM-WFLO-2012
AIAA-SDM-WFLO-2012
 
AIAA-MAO-WFLO-2012
AIAA-MAO-WFLO-2012AIAA-MAO-WFLO-2012
AIAA-MAO-WFLO-2012
 
AIAA-Aviation-VariableFidelity-2014-Mehmani
AIAA-Aviation-VariableFidelity-2014-MehmaniAIAA-Aviation-VariableFidelity-2014-Mehmani
AIAA-Aviation-VariableFidelity-2014-Mehmani
 
AIAA-SDM-PEMF-2013
AIAA-SDM-PEMF-2013AIAA-SDM-PEMF-2013
AIAA-SDM-PEMF-2013
 
ASME-IDETC-Sensitivity-2013
ASME-IDETC-Sensitivity-2013ASME-IDETC-Sensitivity-2013
ASME-IDETC-Sensitivity-2013
 
WCSMO-Wind-2013-Tong
WCSMO-Wind-2013-TongWCSMO-Wind-2013-Tong
WCSMO-Wind-2013-Tong
 
AIAA-Aviation-Vidmap-2014
AIAA-Aviation-Vidmap-2014AIAA-Aviation-Vidmap-2014
AIAA-Aviation-Vidmap-2014
 
WCSMO-ModelSelection-2013
WCSMO-ModelSelection-2013WCSMO-ModelSelection-2013
WCSMO-ModelSelection-2013
 
AIAA-MAO-RegionalError-2012
AIAA-MAO-RegionalError-2012AIAA-MAO-RegionalError-2012
AIAA-MAO-RegionalError-2012
 
WCSMO-Vidmap-2015
WCSMO-Vidmap-2015WCSMO-Vidmap-2015
WCSMO-Vidmap-2015
 
AIAA-SDM-SequentialSampling-2012
AIAA-SDM-SequentialSampling-2012AIAA-SDM-SequentialSampling-2012
AIAA-SDM-SequentialSampling-2012
 

Similar to AIAA-Aviation-2015-Mehmani

AHF_IDETC_2011_Jie
AHF_IDETC_2011_JieAHF_IDETC_2011_Jie
AHF_IDETC_2011_Jie
MDO_Lab
 
A PC-kriging-HDMR Integrated with an Adaptive Sequential Sampling Strategy fo...
A PC-kriging-HDMR Integrated with an Adaptive Sequential Sampling Strategy fo...A PC-kriging-HDMR Integrated with an Adaptive Sequential Sampling Strategy fo...
A PC-kriging-HDMR Integrated with an Adaptive Sequential Sampling Strategy fo...
AIRCC Publishing Corporation
 
Risk_Management_Final_Report
Risk_Management_Final_ReportRisk_Management_Final_Report
Risk_Management_Final_Report
Rohan Sanas
 

Similar to AIAA-Aviation-2015-Mehmani (20)

SIMULATION-BASED OPTIMIZATION USING SIMULATED ANNEALING FOR OPTIMAL EQUIPMENT...
SIMULATION-BASED OPTIMIZATION USING SIMULATED ANNEALING FOR OPTIMAL EQUIPMENT...SIMULATION-BASED OPTIMIZATION USING SIMULATED ANNEALING FOR OPTIMAL EQUIPMENT...
SIMULATION-BASED OPTIMIZATION USING SIMULATED ANNEALING FOR OPTIMAL EQUIPMENT...
 
AHF_IDETC_2011_Jie
AHF_IDETC_2011_JieAHF_IDETC_2011_Jie
AHF_IDETC_2011_Jie
 
Lecture6 xing
Lecture6 xingLecture6 xing
Lecture6 xing
 
The Treatment of Uncertainty in Models
The Treatment of Uncertainty in ModelsThe Treatment of Uncertainty in Models
The Treatment of Uncertainty in Models
 
Agile performance engineering with cloud 2016
Agile performance engineering with cloud   2016Agile performance engineering with cloud   2016
Agile performance engineering with cloud 2016
 
A PC-kriging-HDMR Integrated with an Adaptive Sequential Sampling Strategy fo...
A PC-kriging-HDMR Integrated with an Adaptive Sequential Sampling Strategy fo...A PC-kriging-HDMR Integrated with an Adaptive Sequential Sampling Strategy fo...
A PC-kriging-HDMR Integrated with an Adaptive Sequential Sampling Strategy fo...
 
Sequential estimation of_discrete_choice_models
Sequential estimation of_discrete_choice_modelsSequential estimation of_discrete_choice_models
Sequential estimation of_discrete_choice_models
 
Modelling the expected loss of bodily injury claims using gradient boosting
Modelling the expected loss of bodily injury claims using gradient boostingModelling the expected loss of bodily injury claims using gradient boosting
Modelling the expected loss of bodily injury claims using gradient boosting
 
Evolving CSP Algorithm in Predicting the Path Loss of Indoor Propagation Models
Evolving CSP Algorithm in Predicting the Path Loss of Indoor Propagation ModelsEvolving CSP Algorithm in Predicting the Path Loss of Indoor Propagation Models
Evolving CSP Algorithm in Predicting the Path Loss of Indoor Propagation Models
 
9 coldengine
9 coldengine9 coldengine
9 coldengine
 
A Simulated Annealing Approach For Buffer Allocation In Reliable Production L...
A Simulated Annealing Approach For Buffer Allocation In Reliable Production L...A Simulated Annealing Approach For Buffer Allocation In Reliable Production L...
A Simulated Annealing Approach For Buffer Allocation In Reliable Production L...
 
Risk_Management_Final_Report
Risk_Management_Final_ReportRisk_Management_Final_Report
Risk_Management_Final_Report
 
Block coordinate descent__in_computer_vision
Block coordinate descent__in_computer_visionBlock coordinate descent__in_computer_vision
Block coordinate descent__in_computer_vision
 
IRJET- Optimization of Fink and Howe Trusses
IRJET-  	  Optimization of Fink and Howe TrussesIRJET-  	  Optimization of Fink and Howe Trusses
IRJET- Optimization of Fink and Howe Trusses
 
11 7986 9062-1-pb
11 7986 9062-1-pb11 7986 9062-1-pb
11 7986 9062-1-pb
 
م.80-مبادرة #تواصل_تطويرم.أحمد سعيد رفاعهى-دورة حياة تقدير التكلفة بمشروعات ا...
م.80-مبادرة #تواصل_تطويرم.أحمد سعيد رفاعهى-دورة حياة تقدير التكلفة بمشروعات ا...م.80-مبادرة #تواصل_تطويرم.أحمد سعيد رفاعهى-دورة حياة تقدير التكلفة بمشروعات ا...
م.80-مبادرة #تواصل_تطويرم.أحمد سعيد رفاعهى-دورة حياة تقدير التكلفة بمشروعات ا...
 
Aerodynamic Drag Reduction for A Generic Sport Utility Vehicle Using Rear Suc...
Aerodynamic Drag Reduction for A Generic Sport Utility Vehicle Using Rear Suc...Aerodynamic Drag Reduction for A Generic Sport Utility Vehicle Using Rear Suc...
Aerodynamic Drag Reduction for A Generic Sport Utility Vehicle Using Rear Suc...
 
Advanced Econometrics L11- 12.pptx
Advanced Econometrics L11- 12.pptxAdvanced Econometrics L11- 12.pptx
Advanced Econometrics L11- 12.pptx
 
DEA-Solver-Pro Version 14d- Newsletter17
DEA-Solver-Pro Version 14d- Newsletter17DEA-Solver-Pro Version 14d- Newsletter17
DEA-Solver-Pro Version 14d- Newsletter17
 
Elements CAE white paper
Elements CAE white paperElements CAE white paper
Elements CAE white paper
 

Recently uploaded

Structural Analysis and Design of Foundations: A Comprehensive Handbook for S...
Structural Analysis and Design of Foundations: A Comprehensive Handbook for S...Structural Analysis and Design of Foundations: A Comprehensive Handbook for S...
Structural Analysis and Design of Foundations: A Comprehensive Handbook for S...
Dr.Costas Sachpazis
 
result management system report for college project
result management system report for college projectresult management system report for college project
result management system report for college project
Tonystark477637
 
Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...
Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...
Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...
Christo Ananth
 
UNIT-V FMM.HYDRAULIC TURBINE - Construction and working
UNIT-V FMM.HYDRAULIC TURBINE - Construction and workingUNIT-V FMM.HYDRAULIC TURBINE - Construction and working
UNIT-V FMM.HYDRAULIC TURBINE - Construction and working
rknatarajan
 
VIP Call Girls Ankleshwar 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Ankleshwar 7001035870 Whatsapp Number, 24/07 BookingVIP Call Girls Ankleshwar 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Ankleshwar 7001035870 Whatsapp Number, 24/07 Booking
dharasingh5698
 
FULL ENJOY Call Girls In Mahipalpur Delhi Contact Us 8377877756
FULL ENJOY Call Girls In Mahipalpur Delhi Contact Us 8377877756FULL ENJOY Call Girls In Mahipalpur Delhi Contact Us 8377877756
FULL ENJOY Call Girls In Mahipalpur Delhi Contact Us 8377877756
dollysharma2066
 
Call Girls in Ramesh Nagar Delhi 💯 Call Us 🔝9953056974 🔝 Escort Service
Call Girls in Ramesh Nagar Delhi 💯 Call Us 🔝9953056974 🔝 Escort ServiceCall Girls in Ramesh Nagar Delhi 💯 Call Us 🔝9953056974 🔝 Escort Service
Call Girls in Ramesh Nagar Delhi 💯 Call Us 🔝9953056974 🔝 Escort Service
9953056974 Low Rate Call Girls In Saket, Delhi NCR
 
Call for Papers - Educational Administration: Theory and Practice, E-ISSN: 21...
Call for Papers - Educational Administration: Theory and Practice, E-ISSN: 21...Call for Papers - Educational Administration: Theory and Practice, E-ISSN: 21...
Call for Papers - Educational Administration: Theory and Practice, E-ISSN: 21...
Christo Ananth
 
AKTU Computer Networks notes --- Unit 3.pdf
AKTU Computer Networks notes ---  Unit 3.pdfAKTU Computer Networks notes ---  Unit 3.pdf
AKTU Computer Networks notes --- Unit 3.pdf
ankushspencer015
 

Recently uploaded (20)

Structural Analysis and Design of Foundations: A Comprehensive Handbook for S...
Structural Analysis and Design of Foundations: A Comprehensive Handbook for S...Structural Analysis and Design of Foundations: A Comprehensive Handbook for S...
Structural Analysis and Design of Foundations: A Comprehensive Handbook for S...
 
UNIT - IV - Air Compressors and its Performance
UNIT - IV - Air Compressors and its PerformanceUNIT - IV - Air Compressors and its Performance
UNIT - IV - Air Compressors and its Performance
 
Thermal Engineering-R & A / C - unit - V
Thermal Engineering-R & A / C - unit - VThermal Engineering-R & A / C - unit - V
Thermal Engineering-R & A / C - unit - V
 
Unit 1 - Soil Classification and Compaction.pdf
Unit 1 - Soil Classification and Compaction.pdfUnit 1 - Soil Classification and Compaction.pdf
Unit 1 - Soil Classification and Compaction.pdf
 
result management system report for college project
result management system report for college projectresult management system report for college project
result management system report for college project
 
Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...
Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...
Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...
 
UNIT-V FMM.HYDRAULIC TURBINE - Construction and working
UNIT-V FMM.HYDRAULIC TURBINE - Construction and workingUNIT-V FMM.HYDRAULIC TURBINE - Construction and working
UNIT-V FMM.HYDRAULIC TURBINE - Construction and working
 
Water Industry Process Automation & Control Monthly - April 2024
Water Industry Process Automation & Control Monthly - April 2024Water Industry Process Automation & Control Monthly - April 2024
Water Industry Process Automation & Control Monthly - April 2024
 
VIP Call Girls Ankleshwar 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Ankleshwar 7001035870 Whatsapp Number, 24/07 BookingVIP Call Girls Ankleshwar 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Ankleshwar 7001035870 Whatsapp Number, 24/07 Booking
 
FULL ENJOY Call Girls In Mahipalpur Delhi Contact Us 8377877756
FULL ENJOY Call Girls In Mahipalpur Delhi Contact Us 8377877756FULL ENJOY Call Girls In Mahipalpur Delhi Contact Us 8377877756
FULL ENJOY Call Girls In Mahipalpur Delhi Contact Us 8377877756
 
Call Girls in Ramesh Nagar Delhi 💯 Call Us 🔝9953056974 🔝 Escort Service
Call Girls in Ramesh Nagar Delhi 💯 Call Us 🔝9953056974 🔝 Escort ServiceCall Girls in Ramesh Nagar Delhi 💯 Call Us 🔝9953056974 🔝 Escort Service
Call Girls in Ramesh Nagar Delhi 💯 Call Us 🔝9953056974 🔝 Escort Service
 
(INDIRA) Call Girl Meerut Call Now 8617697112 Meerut Escorts 24x7
(INDIRA) Call Girl Meerut Call Now 8617697112 Meerut Escorts 24x7(INDIRA) Call Girl Meerut Call Now 8617697112 Meerut Escorts 24x7
(INDIRA) Call Girl Meerut Call Now 8617697112 Meerut Escorts 24x7
 
Intze Overhead Water Tank Design by Working Stress - IS Method.pdf
Intze Overhead Water Tank  Design by Working Stress - IS Method.pdfIntze Overhead Water Tank  Design by Working Stress - IS Method.pdf
Intze Overhead Water Tank Design by Working Stress - IS Method.pdf
 
Call for Papers - Educational Administration: Theory and Practice, E-ISSN: 21...
Call for Papers - Educational Administration: Theory and Practice, E-ISSN: 21...Call for Papers - Educational Administration: Theory and Practice, E-ISSN: 21...
Call for Papers - Educational Administration: Theory and Practice, E-ISSN: 21...
 
Call Girls Walvekar Nagar Call Me 7737669865 Budget Friendly No Advance Booking
Call Girls Walvekar Nagar Call Me 7737669865 Budget Friendly No Advance BookingCall Girls Walvekar Nagar Call Me 7737669865 Budget Friendly No Advance Booking
Call Girls Walvekar Nagar Call Me 7737669865 Budget Friendly No Advance Booking
 
Double rodded leveling 1 pdf activity 01
Double rodded leveling 1 pdf activity 01Double rodded leveling 1 pdf activity 01
Double rodded leveling 1 pdf activity 01
 
AKTU Computer Networks notes --- Unit 3.pdf
AKTU Computer Networks notes ---  Unit 3.pdfAKTU Computer Networks notes ---  Unit 3.pdf
AKTU Computer Networks notes --- Unit 3.pdf
 
Roadmap to Membership of RICS - Pathways and Routes
Roadmap to Membership of RICS - Pathways and RoutesRoadmap to Membership of RICS - Pathways and Routes
Roadmap to Membership of RICS - Pathways and Routes
 
(INDIRA) Call Girl Aurangabad Call Now 8617697112 Aurangabad Escorts 24x7
(INDIRA) Call Girl Aurangabad Call Now 8617697112 Aurangabad Escorts 24x7(INDIRA) Call Girl Aurangabad Call Now 8617697112 Aurangabad Escorts 24x7
(INDIRA) Call Girl Aurangabad Call Now 8617697112 Aurangabad Escorts 24x7
 
Call for Papers - International Journal of Intelligent Systems and Applicatio...
Call for Papers - International Journal of Intelligent Systems and Applicatio...Call for Papers - International Journal of Intelligent Systems and Applicatio...
Call for Papers - International Journal of Intelligent Systems and Applicatio...
 

AIAA-Aviation-2015-Mehmani

  • 1. Adaptive Switching of Variable-Fidelity Models in Population-based Optimization Algorithms Ali Mehmani*, Souma Chowdhury#, and Achille Messac# * Syracuse University, Department of Mechanical and Aerospace Engineering # Mississippi State University, Department of Aerospace Engineering The AIAAAviation and Aeronautics Forum and Exposition June 22-26, 2015 Dallas, TX
  • 2. Making Heuristic Optimization more Viable? 2 • Population-based heuristic algorithms have been successfully applied to diverse areas of science and engineering over the past three decades – a core tool in designing complex systems. • Challenge: Although proven to be effective in solving highly non-linear problems, they often demand a high number of function calls – computationally “too expensive” – for complex problems. • Need: Control the computational cost per function call, while preserving the fidelity of the final optima. Variable fidelity models + Model management strategies
  • 3. Variable Fidelity Optimization (VFO) 3 • Variable fidelity models refer to models with different levels of fidelity (low, medium, and high fidelity models) • Model management strategies are techniques used to select different models during the optimization process. Medium fidelityLow fidelity High fidelity Park model (5 sec.) Dynamic model (8 mins) CFD model (30 hrs) Model Management/Switching Optimizer
  • 4. Presentation Outline 4 • VFO: Motivation and Research Objectives • VFO with Adaptive Model Switching (AMS)  Methodology  Quantifying Model Uncertainties • Numerical Case Studies  Aerodynamic shape optimization of 2D airfoil  Shape Optimization of a Cantilever Composite Beam • Concluding Remarks
  • 5. • Trust region strategy Alexandrov et al. (1998) and Rodriguez et al.(2001) low-fidelity model is used to define a sub-region of global optimum, then the high- fidelity model is used to improve the low-fidelity model. • Efficient Global Optimization Jones et. al. (1998) and Viana and Haftka (2013) The accuracy of the low fidelity model (surrogate) is improved by running high fidelity model in the regions of (i) global optimum and (ii) high predicted error. • Individual-based evolution control Jin, et al. (2011) Part of the individuals in the population are chosen and evaluated using high fidelity model. • Generation-based evolution control Jin, et al. (2011) The whole population in specific iteration are evaluated using high fidelity model Variable fidelity optimization (VFO) 5
  • 6. Research Motivation 6 • Several of the existing VFO strategies are found to be defined for specific types of low fidelity model (e.g., EGO works primarily for Gaussian process-based surrogate models), thereby limiting their applicability. • Existing techniques generally consider the combination of only two models of different fidelities (e.g., Trust-region methods, and individual- and generation-based techniques).
  • 7. Research Objectives 7 • Investigate an adaptive and model-independent strategy for managing multiple variable-fidelity models (called Adaptive Model Switching or AMS), with the objective to reduce the computational cost while converging to an optimum with high fidelity function estimates. • Implement the new adaptive variable fidelity optimization method in Particle Swarm Optimization, and test its effectiveness through two practical case studies. Attributes:
  • 8. VFO with Adaptive Model Switching (AMS) 8 • The important question is when and where to use models with different levels of fidelity and cost. Computationally expensive while wasting resources Switching to higher fidelity models too early Mislead the search process to suboptimal (or infeasible) regions Switching to a higher fidelity model too late
  • 9.  The AMS metric is a hypothesis testing that is defined by a comparison between (I) the distribution of the relative fitness function improvement, and (II) the distribution of the error associated with the model. Rejection of the test; Don’t Switch Model Acceptance of the test; Switch Model Fitness Func. Improvement (KDE) Distribution of Model Error (LogN) Adaptive Model Switching (AMS) 9
  • 10. Adaptive Model Switching (AMS) 10 • represents a quantile function of a distribution; • The p-quantile, for a given distribution function, , is defined by • is an Indicator of Conservativeness (IoC). • The IoC for the low fidelity model with an error distribution is defined as the probability of the model error to be less than h • The IoC is based on user preferences, and could be guided by the desired trade- off between reliability and computational cost in the AMS-based VFO. • Here, we defined pcr = 0.3.
  • 11. AMS: Uncertainty Quantification • The uncertainties associated with surrogate models are determined using an advanced surrogate error estimation method, called Predictive Estimation of Model Fidelity *. * Published in Struct. Multidiscip. Optim. (2015), Presented in AIAA SDM (2013) • In the case of physics-based low fidelity (PLF) models, the uncertainty in their output is quantified through an inverse assessment, using the same DOE as used for the surrogate model. • Kernel Density Estimation (KDE) is adopted to model the distribution of the relative improvement in the fitness function over consecutive t iterations. 11
  • 12. Variable Fidelity Optimization: Airfoil Design 14 Mehmani et al. Fig. 5 Design variables gov- erning the geometry of the airfoil 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 -0.04 -0.02 0.00 0.02 0.04 0.06 0.08 0.10 X Y x1 x2 x3 Table1 Design variables in airfoil optimization problem Description Notation Lower limit Upper limit Distance between themiddleof suction sideand horizontal axis x1 -0.01 0.01 Distance between themiddleof pressure sideand horizontal axis x2 -0.01 0.01 Distance between thetrailing edgeand horizontal axis x3 -0.01 0.01 Incidence angle x4 0◦ 10◦ 3.1.1 Aerodynamic modelswith different level of fidelity To develop a high fidelity aerodynamic model for determining CL and CD (MA HF), the commercial Finite Volume Method package, FLUENT, is used. The Reynolds- Shape variable angle of incidence * The Mixed-Discrete Particle Swarm Optimization (MDPSO)[1] is used. [1] Chowdhury and Messac (2012) Wortmann FX60.126 12 Physics-based low fidelity model (PLF) The fluid is steady, incompressible, and irrotational. Computational time = 30 seconds Surrogate model (SM) Kriging-Gaussian-30 sample points Computational time = less than 0.1 second Tuned low fidelity model (TLF) This model is constructed using the Multiplicative approach, as given by onds, respectively (i.e., an order of magnitude apart). The pressure field around the airfoil for the low and high fidelity aerodynamic models at a baseline design (x1 = 0, x2 = 0, x3 = 0, andx4 = 5◦ ) areillustrated in Fig. 7. 1.0171e+05 1.0054e+0 1.0066e+0 1.0077e+0 1.0089e+0 1.0101e+0 1.0112e+0 1.0124e+0 1.0136e+0 1.0147e+0 1.0159e+0 10.50 X (a) High fidelity model −0.5 0 0.5 1 1.5 −1 −0.5 0 0.5 1 x y 1.0124 1.0126 1.0128 1.013 1.0132 1.0134 1.0136 1.0138 x 10 5 (b) Low fidelity physics-based model Fig. 7 Pressure field around theairfoil at abaselinedesign The third model is a surrogate model (MA SM) constructed using a DoE of high fidelity evaluation involving 30 sample points. The fourth model is a tuned low fidelity model (MA TLF). In this article, the tuned low fidelity model is constructed using theMultiplicativeapproach, asgivenby ˜F(x,a) = f(x) × C(x) (21) where ˜F isatunedlow fidelity model; f(x) isalow fidelity model;C(x) isanexplicit tuning surrogateconstructedusingthehighfidelity samples, asshown below: C(x) = CL CD |HF CL CD |PLF (22) C(x) : explicit tuning surrogate, f(x): HF model High fidelity model (HF) The FLUENT is used for solving RANS eq. Computational time = 300 seconds Computational time ~ 30 seconds
  • 13. Airfoil Design : Results 13 0 0.05 0.1 0.15 0.2 0.25 0 2 4 6 8 10 12 14 16 18 20 Error PDF 0 0.05 0.1 0.15 0.2 0.25 0 5 10 15 20 25 Error PDF 0 0.05 0.1 0.15 0.2 0.25 0 5 10 15 20 25 30 35 40 45 50 Error PDF Distribution of the fitness function improvements in different iterations of the airfoil optimization 0 0.05 0.1 0.15 0.2 0.25 0 5 10 15 20 25 30 35 40 Error PDF 5th Iteration 15th Iteration 18th Iteration20th Iteration F.F. improvement Surrogate Error TLF Error
  • 14. Computational Time Function Evaluation provides the best optimum value that is 5% better that the next best value 185% reduction in computing time 14 13th 19th Airfoil Design : Results
  • 15. Variable Fidelity Optimization: Cantilever beam design Minimizing the maximum deflection of a cantilever composite beam. f Weight density of epoxy resin, ρm [N/ mm3] Thebeam optimization problemisdefined as Minimize: δmax δ0 , [δ0] = 12.9 subject to W/ W0 ≤ 1, [W0] = 2.9E4 σmax/ σ0 ≤ 1, [σ0] = 200 x4 2 1.2E6x1 ≤ 1 xmin i ≤ xi ≤ xmax i , i = 1,2,3 In the optimization formulation, the inequality constra arerelatedto theallowableweight, themaximumstress, a on thebeam design (depth ≤ 10× width). Theweight an given by W = AρL = 12I h2 × (12+ 5.2νf)10− 6 × L = x1 x2 σmax = q0L2 h 8I = 1E6x2 8x1 Themodelsused to estimatethemaximum deflection, 3.2.1 Structural modelswith different levelsof fidelity geometric restriction on the beam design (depth ≤ 10 × width)  Minimizing the maximum deflection of a cantilever composite beam.*  The design variables are: (i) the second moment of area (x1), (ii) the depth of the beam (x2), and (iii) the fiber volume fraction (x3).  optimization problem is 15*Zadeh and Toropov, SMO 2009
  • 16. Cantilever beam design : Model Choices Physics-based low fidelity model (PLF) Surrogate model (SM) Tuned low fidelity model (TLF) High fidelity model (HF) The models used to estimate the maximum deflection The PLF Finite Element model is constructed using 2 beam elements in ANSYS Computational time = 3.30 [Sec] The HF Finite Element model is constructed using 1000 beam elements in ANSYS. Computational time = 5.5 [Sec]. Surrogate model is constructed using Kriging with Gaussian correlation function. Computational time less than 0.3 [Sec] . This model is constructed using the Multiplicative approach, as given by Computational time less than 3.33 [Sec] . 1.0054e+0 1.0066e+0 1.0077e+0 1.0089e+0 1.0101e+0 1.0112e+0 1.0124e+0 1.0136e+0 10.50 X (a) High fidelity model −0.5 0 0.5 1 1.5 −1 −0.5 0 x y (b) Low fidelity physics-based mod Fig. 7 Pressure field around theairfoil at abaselinedesign The third model is a surrogate model (MA SM) constructed using a DoE of fidelity evaluation involving 30 sample points. The fourth model is a tuned fidelity model (MA TLF). In this article, the tuned low fidelity model is constru using theMultiplicativeapproach, asgivenby ˜F(x,a) = f(x) × C(x) where ˜F isatunedlow fidelity model; f(x) isalow fidelity model;C(x) isanex tuning surrogateconstructedusingthehighfidelity samples, asshown below: C(x) = CL CD |HF CL CD |PLF whereCL andCD arerespectively thelift and drag coefficients. The surrogate model (MA SM) and the surrogate component of the tuned lo delity model (MA TLF) arebothconstructedusingKrigingwithaGaussiancorrel function.14,37 Kriging isan interpolating method that is widely used for repre ing irregular data. Under theKriging approach, thezero-order polynomial fun isusedasaregressionmodel. InthisarticletheOptimal LatinHypercubeisado to determinethelocationsof thesamplepoints. ThePEMF method isthen ap 22 C(x) = δmax|HF δmax|PLF The distribution of the error in the tuned low fidelity rogate model (SM) are estimated using PEMF (Section in Figs. 13(a) and 13(c), respectively. The distribution o based low fidelity model (PLF) is estimated using the in by leveragingthesame30 high fidelity samplesthat were and SM; thePLF error distribution isshown in Fig. 13(b) ! ! "% ! "( ! ") ! "* $ $"% $"( $") $"* % &'$! # ./0 pcr = 0.3 QP = 0.0001 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 x 10 4 PDF pcr = 0.3 QP = 0.097 0. 0 0. 0 0. PDF UQ: Using MC simulationsUQ: Using PEMFUQ: Using PEMF Tuned LF Model[ Surrogate Model Low-fidelity Model 16
  • 17. Cantilever beam design : Results Computational Time Function Evaluation 33% lower computational expense compared to PSO-TLF 119% lower computational expense compared to PSO-HF 17
  • 18. Concluding Remarks • A novel model management technique, called Adaptive Model Switching, was developed for variable fidelity optimization using population-based algorithms. • The method seeks to provide reliable high-fidelity optimum designs at a reasonable computational expense, by leveraging multiple models. • Model switching is guided by the comparison between: 1. Error distribution of the current model 2. Distribution of the recent fitness function improvement over the entire population of candidate designs • AMS was observed to substantially superior than purely low-fidelity or purely high-fidelity optimizations, in terms of the balance between “quality of the optimum” and “computational efficiency”. • A more intuitive definition of the Indicator of Conservativeness (IoC) as a function of desired trade-offs between computational expense and reliability of optimum would further extend the practical applicability of this algorithm. 18