MVFOSM

Professor Terje Haukaas University of British Columbia, Vancouver terje.civil.ubc.
ca
Mean-value First-order Second-

moment Method (MVFOSM)
Consider a situation in which the only information we have about the random variables is
their second-moments, i.e., means, standard deviations, and correlation coefficients.
Under these circumstances, analysis of functions provides the mean and standard
deviation of the limit-state function, i.e., µg and σg. In MVFOSM these two values are
used to construct a measure of the reliability. The measure is called the reliability index
and serves as a surrogate for the failure probability.
This reliability index is accurate for linear limit-state functions but it suffers from the so-
called invariance problem when the limit-state function is nonlinear. This is because if
the limit-state function is linear then its mean and standard deviation are accurately
computed. Conversely, if the limit-state function is nonlinear then first-order
approximations are utilized.
MVFOSM reliability method is so named for three reasons: The fact that the second
moments of the limit-state function are obtained by a Taylor approximation centred at the
Mean Values of the random variables provides the two first letters. The next two letters
indicate that the Taylor approximation is of the First Order. The last two letters state that
only Second Moment information of the random variables is considered.
Second-moment Reliability Index

In MVFOSM the reliability index is defined as the ratio of the mean to the standard
deviation of the limit-state function:
µg
β= (1)
σg
The validity of β as a proxy for failure probability is understood by first imagining the
PDF of the limit-state function. Unless g is a linear function of normal random variables,
or some other simple case, this PDF cannot be established analytically. Nevertheless, this
is a pedagogically useful thought-construct. Obviously, the PDF of g has mean µg and
standard deviation σg. Furthermore, g=0 separates the failure outcomes from the safe
outcomes. By multiplying Eq. (1) by σg it becomes clear that β is the number of standard
deviations from the mean to the failure region. The more standard deviations the failure
domain is away from the mean the safer. In other words, the higher reliability index the
smaller is the failure probability. Conversely, small values of the reliability index indicate
that the failure domain is closer to the mean, which implies a higher failure probability.
In addition to this explanation, a geometric interpretation of β in the space of random
variables, which will be useful in the derivation of other reliability methods, is presented
later in this document.
Mean-value First-order Second-moment Method (MVFOSM) Updated January 31, 2018 Page 1
Professor Terje Haukaas University of British Columbia, Vancouver terje.civil.ubc.ca
Invariance Problem
Formulas for the computation of µg and σg are known from the analysis of functions. If
the limit-station function is linear then µg and σg are exact. If it is nonlinear then first-
order approximations are available:
µ g ≈ g(M X ) (2)
σ g ≈ ∇g(M X )T Σ XX ∇g(M X ) (3)

It is the potential inaccuracy in these two equations that gives rise to the “invariance
problem” of the MVFOSM method. It is called the invariance problem because the
MVFOSM may provide different results for equivalent limit-state functions. Two limit-
state functions are equivalent if they share the surface g=0 in the space of random
variables. Consider the basic reliability problem as an example. For the limit-state
function g=R-S the MVFOSM reliability index is
µg µ R − µS
β= = (4)
σg σ − 2 ρ RSσ Rσ S + σ
2
R
2
S
For the equivalent limit-state function g=ln(R/S) the solution is

µg ln ( µ R µS )
β= = (5)
σg σ R2 µ R2 − (2 ρ RSσ Rσ S ) ( µ R µS ) + σ S2 µS2
In general, Eqs. (4) and (5) yield different results. For example, for
µR=30, µS=20, σR=5, σS=10, and ρRS=0.5 Eq. (4) yields β=1.15 while Eq. (5) yields
β=0.92. This exemplifies the invariance problem. The root of the problem is that the
linearization of the limit-state function is made at the mean. For a nonlinear limit-state
functions this implies that the limit-state surface, i.e., g=0, is different for the linearized
function and the actual function. It would be better to carry out the linearization
somewhere at the limit-state surface g=0, which is shared by all equivalent limit-state
functions. This solution gives rise to FORM and SORM, which are described in other
documents.
Geometric Interpretation of the Reliability Index

To better understand the reliability index and to assist the development of FORM, a
geometric interpretation of the second-moment reliability index is made. First consider a
linear limit-state function of the form
g(X) = a + bT X (6)
Furthermore, transform the function into the space of standard variables, i.e., transform X
into a vector of variables with zero means and unit covariance matrix:
X = M X + D X LY (7)
Substitution of Eq. (7) into Eq. (6) and naming the limit-state function in the standard
space G yields
G(Y) = a + bT M X + bT D X LY
(8)
= c + dT Y
where c and d have been defined. Now carry out MVFOSM with this limit-state function:
µG c c
β= = = (9)
σG dT d d
Compare this result with the geometry formula for the distance from a point to a plane.
Indeed, the distance from Y=0 (the origin in the Y-space) to the limit-state plane G(Y)=0
is:
G(0) c
Δ= = (10)
∇G d
We conclude that the reliability index β is the distance from the origin to the limit-state
surface in the space of standardized random variables. This shows an important appeal of
the standard space: distances can be measured. Conversely, the original random variables
usually have a variety of units and a distance in that space is not a meaningful concept.
Failure Probability under the Normality Assumption

The MVFOSM method yields a reliability index, but it does not provide a failure
probability. This changes in the special case where the limit-state function is linear and
the random variables are normally distributed. In this case, the probability distribution of
the limit-state function is also normal. Furthermore, the second-moment analysis yields
the precise mean, µg, and standard deviation, σg. Consequently, the standard normal
CDF is employed to obtain the failure probability:
⎛ 0 − µg ⎞
p f = P(g ≤ 0) = Φ ⎜ ⎟ = Φ(− β ) (11)
⎝ σg ⎠
Importance Vector
Ranking of the random variables according to relative importance often provides valuable
insight. The most important random variables should be subjected to particular scrutiny,
while unimportant random variables may be omitted from the analysis. An importance
vector from MVFOSM, i.e., a vector with components that measure the relative
importance the corresponding random variable is derived by considering the linearized
limit-state function at the mean point:
g(x) ≈ g(M x ) + ∇g(M x )T ⋅ ( x − M x ) (12)
The variance of the linearized limit-state function is
Var [ g(x)] ≈ ∇gT ⋅ Σ xx ⋅∇g

n n
= ( ∇g1σ 1 ) + ( ∇g2σ 2 ) ++ ( ∇gnσ n ) + ∑ ∑ ∇gi ∇g jσ iσ j ρij (13)
2 2 2
i=1 j=1
j≠i
where n is the number of random variables. It is observed that the direct contribution of
random variable xi to the total variance is ( ∇gi ⋅ σ i ) . For this reason, the following
2
vector is considered an importance vector in MVFOSM:

ω = −∇g T D (14)
The greater absolute value of ω i = ∇gi ⋅ σ i the greater importance of the corresponding
random variable xi. The sign of ωi also matters:
• Positive ωi means that xi acts like a load variable
• Negative ωi means that xi acts like a resistance variable

MVFOSM

Uploaded by

Copyright:

Available Formats

MVFOSM

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

MVFOSM

Uploaded by

Copyright:

Available Formats

Professor Terje Haukaas University of British Columbia, Vancouver terje.civil.ubc.

Mean-value First-order Second-

Second-moment Reliability Index

σ g ≈ ∇g(M X )T Σ XX ∇g(M X ) (3)

For the equivalent limit-state function g=ln(R/S) the solution is

Geometric Interpretation of the Reliability Index

Failure Probability under the Normality Assumption

Var [ g(x)] ≈ ∇gT ⋅ Σ xx ⋅∇g

vector is considered an importance vector in MVFOSM:

You might also like