You are on page 1of 12

ENCE 302: Project Final Report

12/9/2019

Ayman Amdad (aymanamdad1@gmail.com)


1

Executive Summary
We have analyzed a cantilever beam, taking into consideration a set of random variables of a

sample size of 100. These random variables consisted of tip displacement (D), modulus of elasticity (E),

beam length (L), cross section width (w), cross section height (t), vertical external force (Y), and

horizontal external force (X). With these variables and their mean, coefficients of variation(COV), and

distribution in mind we created a large enough sample size for each variable using the Monte Carlo

simulation method to calculate a sufficient margin of safety (Z). This information along with the margin

of safety’s mean and standard deviation were used to compute failure probability and then perform a chi

square test analyzing the null hypothesis in order to find the probability distribution of the margin of

safety. The Monte Carlo simulation executed at the end of the report was used on a sample size of 5000 to

see the effect that the number of simulation cycles has on the coefficient of variation of the margin of

safety.
2

Table of Contents
Executive Summary___________________________________________________________________1

Table of Contents_____________________________________________________________________2

Problem Description___________________________________________________________________3

Objectives___________________________________________________________________________4

Methodology_______________________________________________________________________5-6

Results____________________________________________________________________________7-9

Conclusion__________________________________________________________________________10

References and Appendices____________________________________________________________11


3

Problem Description
A cantilever beam is shown in Figure 1 below. One of the failure modes is the tip displacement (at point

B) exceeding an allowable displacement value, D. The displacement performance function (Z) is defined

as the difference between D and the computed tip displacement due to the loading X and Y. The margin

between the allowable displacement and actual displacement is treated typically as a measure of the

structural performance of the beam. This margin Z (in) is expressed in the form of a performance function

as follows:
4

Objectives
1. Generate an appropriate number of random numbers (100 values) for each of the random variables

provided in Table 1.

2. Plot histograms for each of the random variables of Table 1 using a respective set of random numbers

from step 1, and show respective probability distributions used to generate them. Show the plots

(histogram and probability distribution) on the same graph for each variable. Calculate the mean and

standard deviation for each random variable based on the random values and discuss the differences.

3. Calculate the mean and standard deviation of the performance Z as the margin of safety.

4. Compute the failure probability of Pz = P(Z < 0).

5. Perform a goodness-of-fit test on the outcome (Z) distribution considering both normal and lognormal

distributions.

6. Compute the vertical load mean necessary to obtain a failure probability (Pf) of 0.25 while keeping all

other probabilistic values of Table unchanged.

7. Perform trend analysis of Pf with an incrementally increasing number of simulation cycles (N), and

compute the statistical uncertainty in the estimate expressed as a coefficient of variation (COV) as

follows:
5

Methodology
1. For each of the random variables within the table given we utilized the rand() function in order to

produce a set of 100 random numbers for each random variable given within the table. Then

based on the variable’s distribution we applied lognormal and normal formulas including

individual variable attributes (probability, mean, standard deviation) produced by a Monte Carlo

simulation.

2. Once the set of 100 values of each variable were created we found the bin width, number of bins

(k), mean, standard deviation, and COV of each set. Using this information for each random

variable we were able to create a plot comparing the actual frequency histogram with the

probability distributions of each independent variable.

3. After producing the plots we calculated the margin of safety (Z) using the formula given in the

problem statement. The set of values for each of the random variables was used within the

formula to create a set of 100 values of Z.

4. This set of Z was then used to calculate failure probability by collecting all the values less than

zero and dividing by 100 to generate a number between zero and one.

5. After finding the failure probability we calculated the normal distribution of the margin of safety.

In order to do so we needed the number of bins, bin width, bin start and bin finish, max, min, and

data range of the margin of safety. We didn’t compute the logarithmic distribution because it is

not possible to compute due to the occurence of negative values in the set of values calculated for

margin of safety.
6

6. Once the cumulative distribution function was created we test its goodness of fit using the

chi-squared test. This required us to calculate ((Oi-Ei)^2)/Ei (where Oi is the observed frequency

and Ei is the expected frequency) for each bin and take the sum. This allowed us to see if we

should accept or deny the null hypothesis that the 100 values of Z illustrate a normal distribution

based on whether this value was less than or greater than the Chi squared value found in table A.3

for an alpha of .05 using five degrees of freedom.

7. After making the decision to accept or deny the null hypothesis we used the “goal seek”

capability of excel to find the new mean value of Y associated with a failure probability of .25 for

the margin of safety.

8. Lastly on sheet 2 in excel, we performed a Monte Carlo simulation again. We initially created

another set of values for each random variable similar to the initial steps on sheet one, except this

time we had 5000 simulation cycles. We picked such a large set due to the fact that the failure

probability of the margin of safety for the initial 10-20 cycles was 0 and the COV was undefined

because the formula required division by Pf. In order to counter this problem we had to do a large

data set and used increments so we can have a better representation of the overall trend and thus a

better representation of the data. Once the probability of margin of safety was found we then

calculated COV. Using an increment of 500 simulation cycles the COV values plotted against

them created a curve showing the relationship between the coefficient of variation of the failure

probability of the margin of safety and the effect that increasing the number of simulation cycles

has on it.
7

Results
1) Frequency histograms and probability distributions on each graph for all the individual random

variables.

2) Once we calculated the margin of safety for each of the 100 cycles we found the failure probability of

the margin of safety, the probability of Z being less than zero, P(Z<0) = .09.

(note picture only shows the first two of the 100 simulation cycles.)
8

3) A chi-squared test was performed by comparing the X^2 of the data being 6.8544, which was less than

11.073, the chi-squared value found in table A.3 for 5 degrees of freedom and an alpha of 0.05. X^2 of the

data being less than the value from the table means that we must accept the null hypothesis that Z

illustrates a normal distribution and thus cannot be rejected.


9

4) We found the mean that is necessary for the probability of failure to increase from .09 to .25 by using

the “goal seek” function. The “set cell” was .09, the “to value” was 0.25, and the “by changing cell” was

set to the cell containing the original mean of 1 of the vertical load. The goal seek function then produced

a mean of 1.501538462kips for the vertical load in order to satisfy a probability of the margin of safety of

0.25.

5) We created a set of Monte Carlo simulations as we did on sheet one, this time for 5000 cycles instead

of 100. For each of the 5000 cycles we then calculated the margin of safety and the probability of failure

in order to calculate COV. By plotting the COV against increments of 500 cycles of the set we see a

decrease in the trend of the coefficient of variation of the failure probability for the margin of safety, Z,

with the increase in the number of simulation cycles.


10

Conclusions
We noticed that when attaining our set for the margin of safety there were negative and positive

values found. Due to the negative values we were unable to represent the data using a lognormal

distribution, therefore only a normal distribution was tested.

Pertaining to the chi-squared test performed we found that the X^2 value of the data was smaller

than chi-squared value found from Table A.3 using 5 degrees of freedom with an alpha of 0.05 which

justifies accepting the null hypothesis of Z demonstrating a normal distribution.

From attaining a new failure probability of 0.25 from our original one, which was 0.09, we saw

the mean for the vertical load, Y, increase by 50.15%. This allowed us to adjust the failure probability of

the margin of safety through the manipulation of the vertical load mean alone.

From our simulation we created a graph with the components being COV and number of

simulation cycles, N. In this graph we were able to see a general trend that showed a large number of

simulation cycles causes a decrease in COV for failure probability of Z. We know that with a decrease in

the COV comes an increase in accuracy. By using a large enough number of cycles, such as 5000, we

were able to produce a more accurate representation of the probability of failure for a cantilever beam.

0
11

References and Appendices


Ayyub, Bilal M., and Richard H. McCuen. Probability, Statistics, and Reliability for Engineers

and Scientists, Third Edition. CRC Press, 2016.

You might also like