Quantitative Risk Analysis in Safety Engineering

Homework 3

Due February 3, 2020

1. Decision Making: You will decide whether or not to install an emergency power source
(EPS) to back up a part of your system in case of normal utility power failure. The cost
of the EPS is $2 K (hardware and maintenance for 1 year), and the cost of a power
outage without the EPS = $10 K and the approximate cost of power outage with the
EPS = 0. Based on past outages, the probability of power failure during a year, P(AC
failure) = 0.1 point value based on relative frequency of occurrence, n/N.

a) State the 2 decision alternatives for this case.

I =
II =

b) Place the information needed to make the decision on the decision tree figure below
or on a similar decision tree that you sketch.

Information needed to make a decision:

What are the alternatives? Shown above for this case
What are the values (or costs) of credible outcomes given a decision?
What are the probabilities of outcome occurrences given a decision?

Probability information: for AC failure or no failure

P(AC failure) = 0.1; P(no AC failure) = 0.9

Outcome value or cost information only for the provided relative costs: On the
decision tree use negative, – for costs or losses; use positive, + for profits or values
above 0. Consider only (relative) costs due to EPS equipment or other costs/losses
as stated.

Relative cost of EPS = $2,000 for each possible outcome of AC failure or

outcome of no AC failure

Relative cost of power outage without the EPS = $10,000

Relative cost of power outage with the EPS = $0.

c) Calculate the expected monetary value (EMV) for each decision alternative and
place EMV values on the decision tree.

EMV = ∑P(i)Cost(i) weighted and summed over all outcomes, where here the value

1
 is in monetary terms.

EMV (I) =

EMV (II) =

d) Based on this information, state the optimum alternative based on the EMV of the
two alternatives, and place the optimum EMV above the decision node on the tree.
Also, state the presumed value of the risk averse parameter β weighting of
consequences for this case. (Recall that β = 1 is risk neutral, β > 1 is risk averse,
and β < 1 is risk seeking.)

Note that the stated probability for AC failure has been estimated for a given
time period of one year. So considering both installation and maintenance of
an EPS (overall maintenance and testing cost estimated to be about the same
as installation cost) and the same probability of AC failure for subsequent
years, the costs calculated represent yearly costs.

Therefore, the optimum alternative based on expected monetary value would

cost $1,000/year based on this information.

e) State whether your team, as the decision maker, would make this decision given
only this information, or your team would consider other information or alternatives
prior to making this decision. State examples of other considerations, such as
consequences of power outage, or information that could be incorporated into the
decision process, such as reaction of customers or company stakeholders to power
outages.

2
2. RERA, 2.5 b: Simplify the following Boolean expression: (A  B)  B) using the
distribution and complementation laws.

(A  B)  B) Distribution Law
Complementation Law
Union with null set

(A  B)

3. IRME, 2.3, p. 52: The time to failure density function, pdf, for a system is f(t) = 0.01,
which is valid within 0 ≤ t ≤ 100 days.

Ebeling refers to λ(t) as a “hazard rate function”, which terminology is ambiguous,

because we must remember to distinguish our functions such that λ(t) is a conditional
failure rate function in contrast to f(t), unconditional failure rate function, and the
parameter λ is designated the rate of occurrence of failure or ROCOF.

4. IRME, 2.10, p. 53, where wear–out of a turbine blade is indicated by an increasing

failure rate.

5. Time to failure of a household refrigerator.

The time to failure of a particular refrigerator type is represented by the following pdf:
f(t) = 0.003(10 − t)2 , which is valid within 0 ≤ t ≤ 10 yr, and f(t) = 0 elsewhere.

a) Write the expression for R(t), integrate over t from t to infinity (which here is 10), and
obtain the cumulative Reliability function, R(t). Then calculate the reliability for the
first year, t = 1. Round your calculated value to 2 sd (significant digits).

b) Assuming that 0.90 reliability is required, calculate the refrigerator Design Life
corresponding to 0.90 reliability. Round your calculated value to 2 sd.

c) Write the expression for and obtain the conditional failure rate (hazard rate) function,
λ(t), and state whether it is increasing or decreasing as t approaches infinity for this
case.

3
 d) To evaluate the reliability of this refrigerator, write the expression for and calculate
the MTTF, mean time to failure to two sd.

e) Obtain the probability and thereby the % units, or relative frequency of occurrence, of
this refrigerator that are expected to survive its MTTF.

f) The refrigerator company has a 1-month warranty program. Write the expression for
and obtain the probability that the refrigerator will fail during the first month.

g) Write the expression for and calculate the probability that the refrigerator will survive
the 1-month warranty program. Then write the expression for and calculate the
conditional probability that the refrigerator will survive the rest of the first year given
the probability that it will survive its warranty period of one month. Round your final
result to 2 sd.

R(1/12) =

R(11/12|1/12) =

6. Consider the heart disease medical example in Unit 5, Slides 16-21. Let the prior
probability of a heart disease, P(H) = 10-3, which is the same as in the original example.
Let E = a positive test, which is 98.6% reliable for those who have the disease. So the
probability of a positive test when you have the disease is P(E|H) = 0.986 and the false
negative test probability is P(not E|H) = 1 – 0.986 = 0.014 by Pr Axiom 2.

Also, the test is 97.7% reliability for those who do not have the disease. So the
probability of a negative test when a patient does not have the disease is
P(not E| not H) = 0.977 and the false positive probability for this test is P(E| not H) =
1 – 0.977 = 0.023 by Pr Axiom 2.

Note that this exercise is basically the same problem as the original in Unit 5, Slides 16-
21 except with adjusted and more realistic false negative and false positive probabilities
of a test for heart disease. All realistic tests have false positive and false negative
uncertainties. Note that because the false positive error P(E| not H) = 0.023 is less than
the previous false positive error of 0.05 (Unit 5, Slide 16), the technology of the new test
in this exercise is more advanced and therefore more reliable for diagnosing disease
than the previous test. Therefore, for a more accurate diagnosis, the fraction of the total
evidence for a positive test, E, P(H|E) = P(H,E)/P(E) =P(E|H)P(H)/P(E) is greater and
the P(H|E) distribution sharper, because P(E) has been reduced by a lower false
positive uncertainty.

4
 a) Use the Bayes model, write the expression (of conditional probabilities and prior
probabilities based on the LTP) for the Prior PPOV, Prior Predictive Probability of the
Observed Variable, E, P(E), which is also the marginal probability (summing over all
values of the uncertain variable H) of a positive test, as in the original example.

P(E) =

b) Use the conditional probability data and prior probability of H described above to
calculate, by the LTP the total probability of E, which is the normalizing factor in the
denominator of the Bayes expression in b (to obtain the fraction of the overlap of the
Prior information, P(H), and the Likelihood observed data, P(E|H). For discrete
Bayes calculations, the denominator is calculated first, because it requires all of the
data using the conditional probabilities and the prior probability.

P(E) =

c) State the Bayes model expression in terms of H, E to calculate the posterior

probability of disease given a positive test, P(H|E), which is the posterior probability
distribution of H given the observed test.

P(H | E) =

d) Using your Bayes expression and your calculated value of P(E) to calculate the
updated probability of heart disease given a positive test, P(H|E), which is the
Posterior probability of H given the observed test.

P(H | E) =

e) Consider and state what you could do beyond this one test at a particular level of
technology to support a heart disease diagnosis with an increased reliability and
therefore lower uncertainty.

7. A new structure will be proof tested. It is assumed that the maximum proof load is
specified at sufficiently high level so the calculated probability of the structure surviving
the maximum proof load test is 0.9. Based on expert opinion, this 0.9 value is
considered to be 70% reliable, but there is a 25% chance that the true probability of
surviving the test is 0.5, and there is a 5% chance that the survival probability is only
0.1.

Let p = probability that the structure survives a proof test, so the 3 p are values of the
discrete likelihood function in the Bayes model, which is the probability of observing the
results of a survived proof load test. Use the 0.7, 0.25, 0.05 probability values that the
structure would pass the load test based on expert opinion for the prior probability of p

5
(prior to doing the test), P’(p). Place the probability values on the table below, where
the first row of values is provided.

Likelihood, p Prior, P’(p)

0.9 0.7

a) Sketch the prior probability pmf values and include the sketch in your paper. Then
use the LTP to calculate the total probability of survival of the proof test or the
PriorPPOV, the Prior Predictive Probability of the Observed Variable given all
information, P(ε). This P(ε) is the total probability of observing ε (all test outcomes of
structure S) and the normalizing factor and denominator of the Bayes model,
calculated from the prior distribution and the likelihood function (likelihood of a test
outcome) as shown below.

P(ε | p j ) P ′(p j ) P(ε | p j ) P ′(p j )

P ′′(p j | ε ) = =
P(ε ) k
∑ P(ε | pi ) P ′(pi )
i =1

Let p = Probability that the structure survives the proof test.

P(survival of the structure) = P(S) = P(ε), where ε the observed quantity, and P(ε) is
the prior predictive probability of the observed quantity, ε.

Prior probability values: P(p=0.9) = 0.70, P(p=0.5) = 0.25, P(p=0.10) =0.05

P(ε) =

b) Because 1 structure unit is proof tested and it survives the maximum proof load, use
this likelihood information given above and the prior probability values in the Bayes
model to calculate the posterior probability of p, P”(p| ε) for each of the 3 values of p.
Sketch the 3 updated values of P”(p| ε) based on this test. Compare the posterior
values, P”(p| ε), based on the prior values and test observation, with the
corresponding prior values, P’(p) and include in your sketch with the prior values.
Are the changes in probability values reasonable or expected based on the test
observations?

Let ε = One structure to survive the proof test. Update the prior values of p (0.9. 0.5,
and 0.1) using the likelihood information to obtain values of the posterior distribution

P’’(p=0.9) = P(p=0.9| ε) = P(ε |p=0.9)´P(p=0.9) / P(ε) from the Bayes model

expression = 0.9´0.7 / 0.76 = 0.83

6
 Similarly,
P’’(p=0.5) =
P’’(p=0.10) =

Note that based on the successful test, the prior probability, P(p = 0.9) was
increased from 0.7 to 0.83 and the other two prior probabilities for p = 0.5 and p =
0.1 were decreased as expected and as shown in the sketch below.

c) Use the posterior probability values from the 1st test to update the Prior PPOV, Prior
Predictive Probability of the Observed Variable, ε, structure surviving the proof test,
P’(ε), to obtain the Posterior PPOV, Posterior Predictive Probability of the Observed
Variable. Compare this posterior predictive probability following the test (using the
posterior, instead of the prior, as the weight function), P”(ε), to the Prior Predictive
Probability of surviving the test (using the prior as the weight function). Is this the
change in value expected based on the test?
3
P"(ε ) = ∑ pi P" (pi | ε ) =
i=1