Anda di halaman 1dari 21

Reliability-based Design Optimization

in OpenSees

Terje Haukaas, UBC, Vancouver

Developer Symposium, Richmond Field Station, August 24, 2005


Vision
Rational decision making under uncertainty (in performance-based engineering)

 “Is
“Isthe
thestructural
structuralperformance
performanceOK?”
OK?”


Prediction
Predictionofofstructural
structuralperformance
performance


Can
Canonly
onlybebedone
doneininaaprobabilistic
probabilisticsense
sense


Reliability
Reliabilityanalysis
analysis



“If
“Ifnot,
not,which
whichparameter
parametertotochange?”
change?”


Response
Responsesensitivities
sensitivities


Importance
Importancemeasures
measures

 Ultimate
Ultimatequestion:
question:“Have
“Havewe
webalanced
balancedcost
costand
andsafety?”
safety?”


Reliability-based
Reliability-baseddesign
designoptimization
optimization(RBDO)
(RBDO)


Rational
Rationaldecision
decisionmaking
making

Developer Symposium, Richmond Field Station, August 24, 2005


Decision Making
 Scientists:
Scientists:Can
Canresearch
researchforever
foreverwithout
withoutmaking
makingaadecision
decision
 Engineers:
Engineers:Must
Mustmake
makedecisions
decisionsunder
underuncertainty
uncertainty
 The
Thedecision
decisioncan
canbe
betotoperform
performfurther
furtherstudies,
studies,or
ortotomake
makeaafinal
finaldecision
decision
 What
Whatisisan
anacceptable
acceptabledesign?
design?/ /How
Howtotodetermine
determinetarget
targetsafety
safetylevel?
level?

1)
1) Calibration
Calibrationtowards
towardsthe
thecurrent
current“accepted”
“accepted”design
designpractice
practice
2)
2) Calibration
Calibrationtowards
towardsthe
thebackground
backgroundrisk
riskininsociety
society
3)
3) Determination
Determinationofofsocial
socialconsensus
consensuslevel
level
4)
4) Reliability-based
Reliability-basedoptimal
optimalstructural
structuraldesign;
design;balance
balancecost
costand
andsafety
safety

Developer Symposium, Richmond Field Station, August 24, 2005


Reliability-based Optimal Design
 Minimize total expected cost, subject to constraints

Total expected cost


Initial cost

Expected cost of failure


Optimum Design variable

{
min c0 + c f p f | f ≤ 0 }
Reliability-based c0 = initial cost
design optimization cf = cost of failure
(RBDO)
pf = probability of failure
f = constraints on design variables

No reliability constraints?

Developer Symposium, Richmond Field Station, August 24, 2005


Determination of Costs and Probabilities

c0 Cost of construction
New structures
Retrofit of old structures

cf Cost of failure
Damage cost
Downtime cost
Value of life, cost of injury
Present value of future events

pf Probability of failure
Finite element reliability analysis
Issues: multiple failure modes / dependence / system reliability analysis

Developer Symposium, Richmond Field Station, August 24, 2005


Objective
 Implement RBDO analysis capabilities in main-stream engineering analysis software

 Open-source
 Object-oriented
 Already extended with reliability and response sensitivity capabilities
 Framework of analysis tools: ReliabilityAnalysis

FindDesignPoint SearchDirection StepSizeRule ProbabilityTransformation

ConvergenceCheck RandomNumberGenerator GFunEvaluator GradGEvaluator

Developer Symposium, Richmond Field Station, August 24, 2005


Problems and Solutions

 {
Problem: min c0 + c f p f | f ≤ 0 , p f ≤ p f target }

 Challenges
 Computational cost
 Potentially non-smooth reliability

 Solution techniques
 Nested bi-level approach
 Mono-level approach
 Genetic algorithms
 Response surface methods and neural networks
 … and …

Developer Symposium, Richmond Field Station, August 24, 2005


The Decoupled Sequential Approach
(Ref: Royset, Polak, Der Kiureghian)

 Reformulation
 Replace the failure probability by an auxiliary variable

{
min c0 + c f ⋅ a | f ≤ 0, p f = a, 0 ≤ a ≤ p f target }
 Replace the constraint pf = a by the requirement that the limit-state function
be zero at a distance –Φ-1(pf) from the origin

 Do not allow the limit-state function to be negative within a ball in the space
of random variables

 Solution
 “Semi-infinite optimization problem”
 Method of Outer Approximations (Polak 1997)

Developer Symposium, Richmond Field Station, August 24, 2005


Implementations in OpenSees
1)
1) Solve
Solvethe
thesemi-infinite
semi-infiniteoptimization
optimizationfor
foraagiven
givenball
ballradius
radius
2)
2) Run
Runreliability
reliabilityanalysis
analysiswith
withmethod-of-choice
method-of-choice
3)
3) Update
Updatethe
theradius
radiusof
ofthe
theball
balland
andrepeat
repeat

Finite element reliability module

x, u
RBDO Algorithm
g, ∇g, f j , ∇f j , F, ∇F

Developer Symposium, Richmond Field Station, August 24, 2005


Objects ReliabilityAnalysis

DSA-MOOAAnalysis DSA-SAnalysis

ReliabilityDomain

NonlinSingleIneqOpt NonlinMultiIneqOpt

designVariable costFunction objectiveFunction


PolakHeNonlinSingleIneqOpt PolakHeNonlinMultiIneqOpt
constraintFunction designVariablePositioner

evaluateFun evaluateGradFun

evaluateFun LinMultiIneqOpt evaluateGradFun

LSSOLLinMultiIneqOpt

Developer Symposium, Richmond Field Station, August 24, 2005


Response Gradients

 Direct differentiation method (DDM)


∂g ∂g ∂d ∂v
=
∂u ∂d ∂v ∂u

∂d n ∂Pnext ∂Pnint
Kn = − |d n fixed
∂v ∂v ∂v

Developer Symposium, Richmond Field Station, August 24, 2005


Gradient Discontinuities
stress
Fy αE
γFy
smoothing

strain
E
− γFy
− Fy
Steel01

Column Fiber Section Beam Fiber Section


unconfined concrete
20 fibers

confined concrete
20 fibers

reinforced steel layer

unconfined concrete
2 fibers

Developer Symposium, Richmond Field Station, August 24, 2005


Example
 Six-story reinforced concrete frame in Vancouver
 Load case: 1.0 × dead load + 0.5 × live load + 1.0 × earthquake load
 Original element dimensions:
H6
width (b) x depth(h)
Interior columns 500 x 500 mm
H5
Exterior columns 450 x 450 mm
First 3 storys beams 400 x 600 mm
H4
Top 3 storys beams 400 x 550 mm

H3  Drift ratio control:


Roof displacement is 297mm by
H2 static pushover analysis.
Drift ratio is
297/23100 = 1.3% < limit of 2%
H1
 Reliability analysis results:
Reliability index β = 3.25
Failure probability pf = 0.0577%

Developer Symposium, Richmond Field Station, August 24, 2005


Example

Column Fiber Section Beam Fiber Section


unconfined concrete
20 fibers

confined concrete
20 fibers

reinforced steel layer

unconfined concrete
2 fibers

Nonlinear material models

Developer Symposium, Richmond Field Station, August 24, 2005


78 random variables
c.o.v c.c.
Variable Mean Type Description
.
H1 28490 kN 0.15 lognormal lateral load on floor 1
H2 48950 kN 0.15 lognormal lateral load on floor 2
H3 70070 kN 0.15 lognormal lateral load on floor 3
0.7
H4 89100 kN 0.15 lognormal lateral load on floor 4
H5 109780 kN 0.15 lognormal lateral load on floor 5
H6 131890 kN 0.15 lognormal lateral load on roof

f cc' 1 L f cc' 8 39 MPa 0.15 0.7 lognormal confined concrete strength


modulus of elasticity of
E cc1 L E cc8 9750 MPa 0.10 0.7 lognormal
confined concrete

f c'1 L f c'14 30 MPa 0.15 0.7 lognormal unconfined concrete strength


modulus of elasticity of
Ec1 L Ec14 15000 MPa 0.10 0.7 lognormal
unconfined concrete
f y1 L f y14 400 MPa 0.15 0.7 lognormal steel bar strength
200000
E1 L E14 MPa
0.05 0.7 lognormal modulus of elasticity of steel

Developer Symposium, Richmond Field Station, August 24, 2005


18 design variables
Variable Initial Value Description
b1× h1 0.45×0.45m width and depth of exterior columns of first three stories
half of the area of reinforced bars of exterior columns of
A1 0.003m2
first three stories
b2× h2 0.45×0.45m width and depth of exterior columns of top three stories
half of the area of reinforced bars of exterior columns of
A2 0.003m2
top three stories
b3× h3 0.50×0.50m width and depth of interior columns of first three stories
half of the area of reinforced bars of interior columns of
A3 0.003m2
first three stories
b4× h4 0.50×0.50m width and depth of interior columns of top three stories
half of the area of reinforced bars of interior columns of
A4 0.003m2
top three stories
b5× h5 0.40×0.60m width and depth of exterior columns of first three stories
A5 0.0024m2 area of reinforced bars of first three stories’ beams
b4× h4 0.40×0.55m width and depth of exterior columns of top three stories
A6 0.0024m2 area of reinforced bars of top three stories’ beams

Developer Symposium, Richmond Field Station, August 24, 2005


Problem Definition
 Limit-state function: g = 23.1m × 2% − d roof

 Cost functions  Objective functions:

F =  ∑ ∑ (bi hi + 100 ⋅ Ai ) ⋅ Li 
4 6
(bi hi + 100 ⋅ 2 ⋅ Ai ) ⋅ Li +
 i =1 i =5 
+ p f ( x ) ⋅ 5 ⋅  ∑ ∑ (bi hi + 100 ⋅ Ai ) ⋅ Li 
4 6
(bi hi + 100 ⋅ 2 ⋅ Ai ) ⋅ Li +
 i =1 i =5 

Cost of failure is 5 times initial cost Cost of steel is 100 times that
of concrete (per volume)

 Structural constraints:
 0 ≤ bi, hi
 0.5 ≤ bi/hi ≤ 2
 0.01 ⋅ bi hi ≤ Ai ≤ 0.02 ⋅ bi hi

Developer Symposium, Richmond Field Station, August 24, 2005


Reliability-based Design Optimization
1.0

0.9
 Initial design:
0.8
 Reliability index: 3.12
0.7

Lateral Load Factor


 Total expected cost: 130.7 0.6

0.5

 Optimized design: 0.4


(1) Original mean point
 Reliability index: 3.0 0.3
(2) Original MPP
 Total expected cost: 85.6 0.2
(3) Optimal mean point
0.1 (4) Optimal MPP
0.0
0.00 0.08 0.16 0.24 0.32 0.40 0.48
Roof Drift (m)

Developer Symposium, Richmond Field Station, August 24, 2005


Concluding Remarks
 Sophisticated structural model

 Implementation is software that is increasingly


employed in the earthquake engineering community

 Response gradients obtained by the DDM

 Valuable tool for performance-based engineering:


guide for design improvement

 Technical issues
 Expansion of reliability constraints
 Computational cost
 Handling of nonlinearities

Developer Symposium, Richmond Field Station, August 24, 2005


Acknowledgements
 Hong Liang, former graduate student at UBC, Vancouver
 Johannes O. Royset, Naval Postgraduate School, Monterey

Developer Symposium, Richmond Field Station, August 24, 2005


Thank you for your attention!
Developer Symposium, Richmond Field Station, August 24, 2005

Anda mungkin juga menyukai