You are on page 1of 11

NPTEL

Course On

STRUCTURAL
RELIABILITY
Module # 05
Lecture 7

Course Format: Web

Instructor:
Dr. Arunasis Chakraborty
Department of Civil Engineering
Indian Institute of Technology Guwahati
7. Lecture 07: Stochastic Response Surface
Method (Continue...)

In this lecture example is shown related to practical field. This will help a clear inside view to the
method.

Ex#09:

In this example, reliability analysis of retaining wall shown in Figure 5.8.8 is performed for
overturning failure. To simplify the analysis, the ground water table is assumed to be below the

Figure 5.7.8 Cantilever retaining wall

base of the retaining wall. The parameters 𝐻, 𝑡𝑠 , 𝐵, 𝑡𝑏 , 𝐵𝑠 are height of the wall from ground
surface, thickness of stem, base width, thickness of base slab and the distance of wall centerline
from heel of wall respectively. Performance function for overturning failure is given by

𝑔 𝑋 = 𝑀𝑟 − 𝑀𝑜

Course Instructor: Dr. Arunasis Chakraborty


1
Lecture 07: Stochastic Response Surface Method (Continue...)
In the above equation, 𝑀𝑟 and 𝑀𝑜 are the stabilizing and the overturning moments, details of
these moments are given below.

The wall-backfill system is divided into four parts as marked in the figure above. Vertical load
due to each of the four components are given by
𝑡𝑠
𝑊1 = 𝑤𝑠 𝐵𝑠 − 𝐻 − 𝑡𝑏
2
𝑊2 = 𝑤𝑐 𝑡𝑠 𝐻 − 𝑡𝑏
𝑊3 = 𝑤𝑐 𝑡𝑏 𝐵
𝑡𝑠
𝑊4 = 𝑞 𝐵𝑠 −
2

In the above equation, 𝑤𝑠 , 𝑤𝑐 and 𝑞 are unit weight of soil, unit weight of concrete and surcharge
load respectively. 𝑊1 is the weight due to the mass of soil, weight due to the stem is given by
𝑊2 , weight of the base slab is 𝑊3 and 𝑊4 is the weight due to surcharge load on the soil. Total
vertical load due to the wall (𝑊𝑇 ) is the sum of weights of the four individual parts, and is given
by

𝑊𝑇 = 𝑊1 + 𝑊2 + 𝑊3 + 𝑊4

The resisting moments about the toe due to each of the four segments are

𝑡 2
𝐵𝑠 − 2𝑠
𝑀1 = 𝑤𝑠 𝐻 − 𝑡𝑏
2
𝑀2 = 𝑤𝑐 𝑡𝑠 𝐻 − 𝑡𝑏 𝐵𝑠
𝐵2
𝑀3 = 𝑤𝑐 𝑡𝑏
2
𝑞 𝑡𝑠 2
𝑀4 = 𝐵𝑠 −
2 2

The moments 𝑀1 , 𝑀2 , 𝑀3 and 𝑀4 are due to loads 𝑊1 , 𝑊2 , 𝑊3 and 𝑊4 respectively. Total


moment 𝑀𝑤 due to the weight of the different components of the wall is

𝑀𝑤 = 𝑀1 + 𝑀2 + 𝑀3 + 𝑀4

The distance between the heel and the point of action of the total weight 𝑊𝑇 is 𝑥𝑤 which is given
as

𝑀𝑤
𝑥𝑤 =
𝑊

Therefore, the total resisting moment about the toe of the wall is

𝑀𝑟 = 𝑊 𝐵 − 𝑥𝑤
Course Instructor: Dr. Arunasis Chakraborty
2
Lecture 07: Stochastic Response Surface Method (Continue...)
The active earth pressure that cause overturning moments is evaluated using Coulomb’s earth
pressure theory. The active earth pressure acting on the retaining wall is given by

1
𝑃𝑎 = 𝐾 𝑤 𝐻 2 + 𝐾𝑎 𝑞𝐻
2 𝑎 𝑠

Where, the active earth pressure coefficient 𝑘𝑎 is defined as

𝑐𝑜𝑠 2 𝜑 − 𝛼
𝑘𝑎 = 2
sin⁡𝜑 + 𝛿 𝑠𝑖𝑛 𝜑 − 𝛾
𝑐𝑜𝑠 2 𝛼 𝑐𝑜𝑠 𝛼 + 𝛿 1+
𝑐𝑜𝑠 𝛼 + 𝛿 𝑐𝑜𝑠 𝛾 − 𝛼

In the above equation, 𝛼 and 𝛾 are the angle of the wall and the inclination of the backfill
respectively. The wall is assumed to be vertical, the resultant of the pressure acts on the vertical
𝐻
face at an angle equal to the wall friction angle and at 3 height from base. Thus, the total
overturning moment about the toe is obtained by evaluating the total moment caused by the earth
pressure and multiplying it with a factor to account for the friction between the wall and the soil.
Therefore, the total overturning moment is given as

1 𝐻3 𝐻2
𝑀𝑜 = 𝐾𝑎 𝑤𝑠 + 𝐾𝑎 𝑞 𝑐𝑜𝑠 𝛿
2 3 2

The details of random variables are given below

Table 5.7.4 Random Variables

Distribution Statistics
Variable
Case 1 Case 2 μ Cov (%)
𝑞 N LN 20 (kN) 10
𝛾𝑐 N N 24 (kN/m3) 7
𝛾 N N 18 (kN/m3) 7
𝜙 N LN 30° 7
𝛿 N LN 10° 7

Solution. CDF of the performance function obtained by Monte Carlo simulations and 4th order
Polynomial chaos expansion have been plotted in Figure 5.8.9. It is observed that CDF from
polynomial chaos matches exactly with the CDF obtained from Monte Carlo Simulations. Figure
5.7.10 shows the variation in reliability with change in number of regression points, where n =
No. points required/No of unknowns. From 𝑛 = 1.5 convergence is obtained, so for required

Course Instructor: Dr. Arunasis Chakraborty


3
Lecture 07: Stochastic Response Surface Method (Continue...)
degree of accuracy, the number of regression points has been chosen as 1.5 times the number of
unknowns.

Figure 5.7.9 Comparison of CDF from MCS and PCE

Figure 5.7.10 Number of regression points vs Reliability index

Course Instructor: Dr. Arunasis Chakraborty


4
Lecture 07: Stochastic Response Surface Method (Continue...)
Figure 5.7.11 and Figure 5.7.12 show the variation of 𝛽 and 𝑃𝑓 with change in the height of the
wall keeping the length of base slab constant at 1.8m.

Figure 5.7.11 Variation of 𝜷 with change in height of wall

Figure 5.7.12 Variation of probability of failure with change in height of wall

Course Instructor: Dr. Arunasis Chakraborty


5
Lecture 07: Stochastic Response Surface Method (Continue...)
Similarly, Figure 5.7.13 and Figure 5.7.14 show the variation of 𝛽 and 𝑃𝑓 with change in the
length of the base slab keeping the height of the wall constant at 4.0m. It can be seen that the
values of 𝛽 and 𝑃𝑓 obtained from SRSM and Monte Carlo simulations (with sample size 6
million) have a close match.

Figure 5.7.13 Variation of 𝜷 with change in length of base slab

Figure 5.7.14 Variation of probability of failure with change in length of base slab

Course Instructor: Dr. Arunasis Chakraborty


6
Lecture 07: Stochastic Response Surface Method (Continue...)
Figure 5.7.15 and Figure 5.7.16 provide the design curves, herein the values of height of the wall
and the length of base slab have been varied and the corresponding values of 𝛽 and 𝑃𝑓 has been
plotted.

Figure 5.7.15 Variation of 𝜷 with change in length of base slab and height of wall

Figure 5.7.16 Variation of 𝑷𝒇 with change in length of base slab and height of wall

Course Instructor: Dr. Arunasis Chakraborty


7
Lecture 07: Stochastic Response Surface Method (Continue...)
Ex#010: As per IS: 456-2000, shear resistance of reinforced concrete beam is expressed by

0.87𝑓𝑦 𝐴𝑠𝑣 𝑑
𝑉𝑅 = + 𝜏𝑐 𝑏𝑑
𝑠𝑉

where, 𝑓𝑦 is yield strength of steel, 𝐴𝑠𝑣 is total area of stirrup bar, 𝑠𝑉 is spacing of stirrup, 𝑏 is
width of the section, 𝑑 is the depth of the section and 𝜏𝑐 is shear strength of concrete which can
be expressed by

0.85 0.8𝑓𝑐𝑘 1 + 5𝜂 − 1
𝜏𝑐 =
6𝜂

here, 𝜂 is constant. It depends on percentage of reinforcement of the section and compressive


strength of the concrete, as expressed below

0.8𝑓𝑐𝑘
𝜂=
6.89𝑝𝑡

where, 𝑓𝑐𝑘 and 𝑝𝑡 are concrete compressive strength and percentage of steel (𝐴𝑠𝑡 /𝑏𝑑)
respectively. Here limit state function is

𝑔 𝑋 = 𝑉𝑅 − 𝑉𝑆

where, 𝑉𝑆 is external shear force at the section. Details of random variables are given below

Table 5.6.5 Random Variables

Variable Type 𝝁 𝜹
𝑏 N 300 𝑚𝑚 0.01
𝑓𝑐𝑘 LN 25 𝑁/𝑚𝑚2 0.15
𝑑 N 450 𝑚𝑚 0.01
𝑠𝑣 N 150 𝑚𝑚 0.10
𝑓𝑦 N 415 𝑁/𝑚𝑚2 0.10
𝑉𝑆 LN 100 𝑘𝑁 0.35

Solution. In this problem percentage of steel is taken 1%. Reliability is evaluated by Rackwitz-
Fiessler algorithm using the PCE approximation of the performance function. Reliability indexes
obtained from different orders of polynomial approximation are compared with Monte Carlo
simulations, shown in Figure 5.7.17. It easily observed that the 2nd order approximation is not

Course Instructor: Dr. Arunasis Chakraborty


8
Lecture 07: Stochastic Response Surface Method (Continue...)
accurate for higher values of 𝛽, however it converges to exact value for lower 𝛽 values. On
increasing the order of the polynomial, accuracy of the approximation of the limit surface
increases and consequently 4th order PCE approximation matches closely with Monte Carlo
simulations. For the corresponding values of reliability index probabilities of failure are shown in
Figure 5.7.18. For 4th order PCE, regression/collocation points are generated using the roots of
the one-dimensional hermite polynomial of 5th order (as PCE is of 4th order). Six dimensional, 4th
order PCE has 210 unknown coefficients, in order to robust estimation of the coefficients
2.5*210 i.e. 525 collocation points are used for regression. Therefore, only 315 function
evaluations are required in SRSM for accurate reliability estimation whereas 6 million function
evaluations are required for Monte Carlo Simulations. Though RSM may require even less
function evaluations, but it may not give accurate results.

Figure 5.7.17 Variation of Beta with increase in Shear

Course Instructor: Dr. Arunasis Chakraborty


9
Lecture 07: Stochastic Response Surface Method (Continue...)

Figure 5.7.18 Variation of Probability of failure with increase in Shear (𝑿𝟓 )

Course Instructor: Dr. Arunasis Chakraborty


10

You might also like