4.4 Hasofer-Lind Reliability Index (1974) A modified reliability index: it did not exhibit the invariance problem.

The correction is to evaluate the limit state function at a point known as the design point instead of the mean values. The design point is generally not known a priori, an iteration technique must be used (in general). Consider a limit state function g ( X 1 , X 2 ,..., X n ) where the random variables Xi are all uncorrelated. The limit state function is rewritten in terms of the standard form of the variables using
Zi = X i Xi

As before, the Hasofer-Lind Reliability Index is defined as the shortest distance from the origin of the reduced variable space to the limit state function g=0. An iteration is required to find the design
1

point {Z 1* , Z 2* ,..., Z n* } in reduced variable space such that still corresponds to the shortest distance (shown in Figure).

Z2

Z* 1

Z1

Z* 2

The iterative procedure requires us to solve a set of (2n+1) simultaneous equations with (2n+1) unknowns:
* * , 1 , 2 ,..., n , Z1* , Z 2 ,..., Z n ,

where
2

i =

g Z i g Z k

evaluated

ai

design

po int

(
k =1

evaluated

at

design

po int

)2

(a)

g X i g g Xi = = Z i X i Z i X i

(b) (c ) (d) (e)

( )
i =1 i

=1

Z i* = i
* * g ( Z 1* , Z 2 ,..., Z n ) = 0

The simultaneous equation procedure: (1) Formulate the limit state function and appropriate parameters for all random variables involved. (2) Express the limit state function in terms of reduced variables Zi. (3) Use Eq. (d) to express the limit state function in terms of and i . (4) Calculate the n i values. Use Eq.(d)
3

(5)

(6)

(7) (8)

here also to express each i as a function of all i and (note the coupling effect between i). Contact the initial cycle: Assume numerical values of and all i , noting that thei values must satisfy Eq. (c ). Use the numerical values of and i on the right-hand sides of the equations formed in Steps 3 and 4 above. Solve the n+1 simultaneous equations in Step 6 for and i. Go back to Step 6 and repeat. Iterate until the and i values converge.

Example: Calculate the Hassofer-Lind reliability index for the three-span continuous beam shown in following Figure.

W A L B L C L D

The random variables in the problem are distributed load (w), span length (L), modulus of elasticity (E), and moment of inertia (I). The limit state to be considered is deflection, and the allowable deflection is specified as L/360. The maximum deflection is 0.0069wL4/EI, and it occurs at 0,446L from either and (AISC,1986). The limit state function is
L wL4 0.0069 g ( w, L, E , I ) = 360 EI

The means and standard deviations of the random variables are listed in following Table. Variables w Mean 10 kN/m Standard deviation 0.4 kN/m
5

L E I

5m kN/m2 8 10 4 m4
2 10 7

~0 kN/m2 1.5 10 4 m4
0.5 10 7

Solution We follow the Steps in the simultaneous equation procedure. (1) Formulate the limit state function and appropriate parameters for all random variables involved. It has been done. (2) Express g as a function of reduced variables. First, substituting some numbers, g can be expressed as
g =0 5 EI 0.0069(5 4 ) w = 0 EI 310.5w = 0 360
E E w w

Define the reduced variables:

Z2 =

Z3 =

I = I + Z 1 I ;

E = E + Z 2 E ;

w = w + Z 3 w

Substitute into g:
( E + Z 2 E )( I + Z 1 I ) 310.5( w + Z 3 w ) = 0
[2 10 7 + Z 2 (0.5 10 7 )][8 10 4 + Z 1 (1.5 10 4 )] 310.5(10 + Z 3 (0.4)] = 0

(3000) Z1 + (4000) Z 2 + (750) Z1Z 2 (124.2) Z 3 + 12895 = 0

(3) Formulate g in terms of andi :

12895 3000 1 + 4000 2 + 750 1 2 124.2 3

(4) Calculate i values:

(5) The iterations start with a guess for , 1 , 2 , 3 . For example, let us start with 1 = 2 = 0.333 = 0.58; 3 = 0.333 = 0.58 and let = 3.0 (6)~(8) The iterations are summarized in the following Table. Notice that between iterations 5 and 6, the values change very little, so the solution has converged. Faster convergence
7

occurs when the correct signs are used for each i (+ for load effects,- for resistances). Thus the calculated reliability index is approximately 3.17 Table Initial guess 1
1 2 3
3 -0.58 -0.58 +0.58 3.644 -0.532 -0.846 0.039

Iteration number 2 3 4 5
3.429 -0.257 -0.965 0.047 3.213 -0.153 -0.988 0.037 3.175 -0.168 -0.985 0.034 3.173 -0.179 -0.983 0.034

6
3.173 -0.182 -0.983 0.034

The matrix procedure: (1) Formulate the limit state function and appropriate parameters for all random variables X i (i = 1,2,..., n) involved. (2) Obtain an initial design point {X i* } by assuming values for n-1 of the random variables Xi (Mean values are often a reasonable initial choice). Solve the limit
8

state equation g=0 for the remaining random variable. This ensures that the design point is on the failure boundary. (3) Determine the reduced variables {Z i* } corresponding to the design point {X i* } using
Z i* = X i* X i

(f).

(4) Determine the partial derivatives of the limit state function with respect to the reduced variables using Eq.(b). For convenience, define a column vector {G} as the vector whose elements are these partial derivatives multiplied by 1:
G1 G {G} = 2 M Gn

where

Gi =

g Z i

evaluated at design po int

(g)

(5) Calculate an estimate of using the following formula:

{G}T {z * } {G}T {G}

where

{Z }
*

Z 1* * Z = 2 M * Z n

(f)

If the limit state equation is linear, then Eq.(f) reduces to

a0 + ai X i
i =1

(ai X i ) 2
i =1

(6) Calculate a column vector containing the sensitivity factor using

{ } =

{G} {G}T {G}

(g)

(7) Determine a new design point in reduced variables for n-1 of the variables using Eq. (d). (8) Determine the corresponding design point values in original coordinates for the n-1
10

values in Step 7 using Eq. (f) in an alterative form X i* = X + Z i* X . (9) Determine the value of the remaining random variable (i.e., the one not found in Steps 7 and 8) by solving the limit state function g=0. (10) Repeat Steps 3 to 9 until and the design point {X i* } converge.
i i

Example: Repeat above Example using the matrix procedure. Solution (1) For convenience, let X1=I, X2=E, X3=w. The limit state equation is
X 3 L4 L g( X1, X 2 , X 3 ) = 0.0069 360 X 2 X1

(2) For the first iteration, we will assume x1*and x*2 are the mean values of X1 and X2. The value of x3* will be obtained by solving the limit state equation g=0. Thus
* x1 = 8 10 4 ; * x 2 = 2 10 7

11

* * * * x 2 x1 x 2 x1 L x = [ ] = 0.4026 3 = 51.53 360 (0.0069) L4 L * 3

(3) Determine reduced variables Zi* for i = 1,2,3 :

= 0;

z =
* 2

* x2 X 2

= 0; = 103.8

z =
* 3

* x3 X 3

(4) Determine the {G} vector, which involves the partial derivatives of g with respect to the reduced variables Zi* :
g G1 = Z1 g G2 = Z 2 g G3 = Z3 g = X 1

{ zi* }

{ xi* }

* x3 L4 = 0.0069 * * 2 X 1 x2 ( x1 )

g = { z i* } X 2 g * = { zi } X 3

{ xi* }

* x3 L4 = 0.0069 * * 2 X 2 x1 ( x2 )

{ xi* }

31

L4 = 0.0069 * * X 3 x2 x1

12

Plugging in the information for the first iteration, we find the vector {G} to be
2.604 10 3 {G} = 3.472 10 3 1.078 10 4

(5) Calculate :
=
{G}T {z * } {G} {G}
T

= 2.578

(6) Calculate {}:

(7) Determine a new design point in reduced coordinates for n-1 of the random variables. To be consistent with what we did in Step 2, we will find z1* and z2*:
* z1 = 1 = (0.600)(2.578) = 1.546 * z2 = 2 = (0.800)(2.578) = 2.062

(8) For the reduced variables from Step (7), determine the corresponding values of the
13

design point in original coordinates:

(9) Determine the updated value of x3* from the limit state equation g=0. We can use the formula presented in the Step 2 for this. The result is x3* = 17.7 . (10) Repeat Step (3) ~ (9) until convergence is achieved. The subsequent iterations are summarized in the following Table.

14

Table Iteration number 1 2 3 4

5.68 10 4 5.75 10 4 6.70 10 4 7.05 10 4 7.12 10 4 7.14 10 4

x2*

2 10 7

9.69 10 6 5.37 10 6 4.54 10 6 4.42 10 6 4.38 10 6 4.37 10 6

x3*

51.5

17.7

9.95

9.80

10.0

10.0

10.0

2.58 3.29 3.21 3.18 3.18 3.18 3.18

x1*

x2*

9.69 10 6 5.37 10 6 4.54 10 6 4.42 10 6 4.38 10 6 4.37 10 6 4.37 10 6

x3* 17.7

9.95

9.80

10.0

10.0

10.0

10.0

Note that the value of (within numerical precision) obtaining using this procedure is the same value obtained using the simultaneous equation procedure.
15