Lecture 6: First-Order Reliability Method

Lecture #6
Outline

• First-Order Reliability Method (FORM)

• Transformation from X-Space to U-Space
• Reliability Index and Most Probable Point (MPP)
• Reliability Estimation Error by FORM
• Iterative Search of MPP

2
 First-Order Reliability Method

Recall from Lecture #5: Expansion Methods

 The first-order Taylor series expansion of G(X) at the mean value

G ( X) ≈ G (μX ) + 
(
N ∂G μ
X)

i =1 ∂X i
(
X i − µ Xi )
= a1 X 1 + ⋯ + aN X N + b
or G(X) = aTX + b, where a = [a1, a2,…, aN]T contains the first-order
partial derivatives of G with respect to X.
 The first-order approximate mean and variance of G
µG = E [G ] ≈ E aT X + b  = aTμ X + b σ G2 = aT Σ Xa

 Assuming G is normal, the reliability can be estimated as

 µG 
R = Φ − 
 Gσ

3
 First-Order Reliability Method

Recall from Lecture #5: Expansion Methods

 Potential drawbacks of expansion methods:
 Only utilize the first two statistical moments of X while ignoring the
distribution information of these variables;
 Uses the mean point of X as the reference point for building a linear or
quadratic approximation of G(X), which may lead to a large error in
estimating a high reliability (or a low probability of failure).
 As an attempt to overcome these drawbacks, Hasofer and Lind
proposed the FORM.
 The basic idea of FORM is to linearize G(U) at the most probable
(failure) point (MPP) on the limit state function (LSF) G(U) = 0 in
the U-space.
 The U-space is composed of independent standard normal
variables U that are transformed from X.

4
 First-Order Reliability Method

Transformation from X-Space to U-Space

 For a normal random variable Xi with mean μXi and standard
deviation σXi, transformation T can be simply defined as
X − µ Xi
Ui = T ( X i ) = , i = 1,2,..., N
σX i

 In a general case, the transformation uses the CDF mapping

FXi ( X i ) = Φ (Ui )
where FXi and Φ are the CDFs of Xi and Ui, respectively.

5
 First-Order Reliability Method

Transformation from X-Space to U-Space

 Transformation of a standard uniform random variable Xi to a
standard normal random variable Ui
1.0 1.0
0.8 CDF mapping 0.8
0.6 0.6
ΦU i(ui)

FXi(xi)
0.4 0.4
0.2 0.2
0 0
ui Ui xi Xi
0.9
1.0
0.6
fU i(ui)

fX i (xi)

0.3

0 00
0 ui Ui xi 1 Xi

6
 First-Order Reliability Method

Transformation from X-Space to U-Space

 Through the transformation, G(X) in the X-space is mapped to an
equivalent function G(U) = G(T(X)) in the U-space.
G(U) > 0

U2
X2

G(X) > 0 Failure region

Failure region

G(X) ≤ 0 G(X) = 0
Safe region Failure surface
fX(x)
G(U) ≤ 0
Safe region

Mean point

X1 0 U1
0
G(U) = 0
fU(u) Failure surface

7
 First-Order Reliability Method

Most Probable Point (MPP) and Reliability Index

 In the U-space, the MPP u* denotes the point on the failure
surface which has the minimum distance to the origin.
G(U) > 0

U2
X2

G(X) > 0 Failure region

Failure region

G(X) ≤ 0 G(X) = 0
Safe region Failure surface
fX(x)
G(U) ≤ 0
Safe region

MPP u*
Mean point

X1 0 U1
0
G(U) = 0
fU(u) Failure surface

8
 First-Order Reliability Method

Most Probable Point (MPP) and Reliability Index

 This minimum distance is called the reliability index (β). The
reliability R can be computed as (when R ≥ 0.5): Φ .
G(U) > 0

U2
X2

G(X) > 0 Failure region

Failure region

G(X) ≤ 0 G(X) = 0
Safe region Failure surface
fX(x)
G(U) ≤ 0
Safe region
Reliability
index β
MPP u*
Mean point
FORM
X1 0 U1
0
G(U) = 0
fU(u) Failure surface

9
 First-Order Reliability Method

Most Probable Point (MPP) and Reliability Index

 This reliability measure is exact only if the failure surface in the
U-space is of a linear form.
G(U) > 0

U2
X2

G(X) > 0 Failure region

Failure region

G(X) ≤ 0 G(X) = 0 Error

Safe region Failure surface region

fX(x)
G(U) ≤ 0
Safe region
Reliability
index β
MPP u*
Mean point
FORM
X1 0 U1
0
G(U) = 0
fU(u) Failure surface

10
 First-Order Reliability Method

Reliability Estimation Error by FORM

 Reliability is defined as the following multidimensional integral:
R =  f U ( u ) du
ΩS

where Ω denotes the safety region in the U space, and U is

the joint probability density function (PDF) of U.
 FORM approximates the LSF with a linear function. The reliability
estimate by FORM can then be expressed as:
?
R ≈  ˆ S f U ( u ) du Φ( β )
ΩFORM

 Since U is symmetric around the origin, we can rotate the

axes from [ , ] to [ , ] while keeping the same joint PDF.
R ≈  ˆ S f U ' ( u ') du '
ΩFORM

11
 First-Order Reliability Method

Reliability Estimation Error by FORM

 This reliability measure, Φ , is exact only if the failure
surface in the U-space is of a linear form.
?

U2
 ˆS
Ω FORM
f U ( u ) du Φ( β )

Ω
MPP u*

FORM
0 U1
G(U) = 0
fU(u) Failure surface

12
 First-Order Reliability Method

Reliability Estimation Error by FORM

 Note that input variables [ , ] follow independent standard
normal distributions. It then follows that:

R =  ˆ S fU1′ ( u1′ ) fU 2′ ( u2′ ) du1′du2′

ΩFORM

where Ω , , ∈ ∞, ∞ .
 The above integral can be further derived as:
β +∞
R≈ fU1′ ( u1′ ) du1′ ⋅  fU 2′ ( u2′ ) du2′
−∞ −∞

Standard Full area

normal CDF Φ ( β )
1
under PDF

 FORM uses a linear approximation of the LSF for reliability

estimation and the estimation error is due to this approximation.
13
 First-Order Reliability Method

Iterative Search of MPP

 The task of MPP search can be formulated as an optimization
problem with one equality constraint in the U-space
Minimize U
Subject to G ( U) = 0

where the optimum point on the failure surface is the MPP u*.
The MPP search generally requires an iterative optimization
scheme based on the gradient information of G(U).

14
 First-Order Reliability Method

Iterative Search of MPP

 Hasofer-Lind and Rackwitz-Fiessler (HL-RF) method
 Step 1: Set the number of iterations k = 0 and the initial MPP estimate u = u(0)
that corresponds to the mean values of X..
 Step 2: Transform u(k) to x(k) using CDF mapping. Compute the performance
function G(u(k)) = G(x(k)) and its partial derivatives in the U-space as
 ∂G ∂G ∂G 
( )
∇UG u( k ) = 
∂
 1U
,
∂U 2
,…,
∂U

N  U=u( k )

 Step 3: Update the search point at the current iteration as

  Steepest ascent direction

u ( k +1)
=  u( ) • n( ) −
k k
G( )
u(k)
 n( k ) (
∇UG u( k ) )
  (k)
 ( )
∇UG u( )
k

n =
∇ G ( u( ) )
U
k

 Step 4: Repeat Steps 2 and 3 until the convergence of u.

15
 First-Order Reliability Method

Iterative Search of MPP

 Schematic of MPP search with the HL-RF method
U2

G(U) > 0
Failure region
u(0)
u(1)
( ))
G u(
0

G(U) ≤ 0
∇ G ( u( ) )
0
Safe region U

MPP u*
n(0)
G(U) = G(u(0)) > 0

0 U1
G(U) = 0
Failure surface
16
 First-Order Reliability Method

Example on MPP Search

 Consider the following performance function
80
G ( X1 , X 2 ) = 1 −
X12 + 8 X 2 + 5

where X1 and X2 each follow a normal distribution with the mean

4 and the standard deviation 0.6. Find the MPP using HL-RF.
 Solution
 The first iteration in the HL-RF method is detailed as follows:
 Step 1: Set the number of iterations k = 0 and the initial values u = (0,0).
 Step 2: Transform u(0) to x(0). x(0) = (4,4). Compute G(X) at x(0) as
80
G ( X1, X 2 ) = 1 − ≈ −0.5049
X12 + 8 X 2 + 5

and the partial derivatives as:

∂G ∂G ∂X1 160 X1σ X1 ∂G ∂G ∂X 2
= = 2
≈ 0.1367, = ≈ 0.1367
∂U1 ∂X1 ∂U1 ( X 2 + 8 X + 5) ∂U 2 ∂X 2 ∂U 2
1 2
17
 First-Order Reliability Method

Example on MPP Search

 Solution
 Step 3: Update the search point at the current iteration as

( ) ∇UG u( )
0
1

0
( ) ( )
u( ) = u( ) • ∇UG u( ) − G u( ) 
0 0
 2
∇ G ( u( ) )
0
U

= ( 0,0) • ( 0.1367,0.1367 ) − ( −0.5094) 

( 0.1367,0.1367 ) ≈ (1.8633,1.8633)
0.13672 + 0.13672

Iteration history for MPP search

∂G ∂G
Iteration U1 U2 β G( U)
∂U1 ∂U 2
0 0.0000 0.0000 0.0000 -0.5094 0.1367 0.1367
1 1.8633 1.8633 2.6351 -0.1090 0.0944 0.0738
2 2.7774 2.1707 3.5251 -0.0059 0.0860 0.0607
3 2.9225 2.0630 3.5772 0.0001 0.0863 0.0600
4 2.9366 2.0416 3.5766 0.0000 0.0864 0.0600
Optimum 2.9366 2.0416 3.5766 0.0000 0.0864 0.0600
18