Quantitative Risk Analysis in Safety Engineering

Homework 6
Due Monday, February 24, 20120

1. Consider a system component following a Weibull failure distribution with

a shape parameter of 1.4 and a scale parameter, characteristic life, of
550 days.

a) State the behavior range represented by this Weibull distribution

(DFR, CFR, or IFR) and state your reason.

Expected behavior is …

b) If two units of this component are redundant by connection in parallel,

calculate the system of two units Reliability at 100 days, R(100).

2. Write the expression for and find the maximum number of Weibull
components, assumed identical and assumed independent with scale
parameter (Characteristic Life) of 10,000 operating hr and a shape
parameter of β = 1.8 to be placed in series to meet a Reliability criterion
of 0.95 at t = 100 operating hr.

a. Stage what region of the Bathtub curve represents the behavior

of this component, and state why.

Behavior is …

b. Write the expression for the Reliability of n Weibull units in

series. Then find the numerical value of n to result in a reliability
value of 0.95. Also, Calculate the MTTF of this component.

3. IRME 4.12: A component has a Weibull failure distribution with β = 0.86,

and its characteristic life is 2450 days. By how many days will the design
life for a 0.90 reliability specification be extended as a result of a 30-day
burn-in period?

1
 Burn-in period to increase the Reliability specification. Begin with an
expression for the Conditional Reliability to work an additional time t
beyond t0 = 30 day the burn in period given the probability of working at
t0 = 30, and solve for t. Then compare this t with the design life without a
burn in period. When you have calculated the amount of extension in the
design life, comment on the expected cost effectiveness of the burn-in
period to extend the 0.90 Reliability specification.

4. Five assumed (or approximated identical road graders are used for a highway
project. Company data show that the operational life T of each grader follows
Lognormal behavior with a mean life of 1500 hr, σ =450 hr, and δ or cov
(Coefficient of Variation) = σ/μ = 0.30.

Note that we can use the Bernoulli sequence to represent the two states pf
machine failure or success, because the 3 conditions are approximated: binary
(success or fail), equal Pr for each trial, and independent trials. The probability
of one Bernoulli sequence, p = P(fail) for one machine.

a) Given the 5 machines, write an expression for a Bernoulli sequence to

represent the probability of 2 machines fail, 3 machines succeed.

The Bernoulli sequence Pr =

b) Assuming independence among the machines, calculate the Pr that 2

of the 5 graders will fail within 900 hr. Note that the 2 failed machines
could be any 2 of the 5 machines, so the Binomial coefficient is
needed to include all Bernoulli sequences that contribute to the
probability without regard to the order of failures and order of non-
failures, which are removed by the Binomial coefficient. The Bernoulli
sequence together with the Binomial coefficient results in the Binomial
distribution as discussed in class. State the Binomial Coefficient
expression and calculate its value for this case for the number of ways
2 out of 5 machines can fail.

c) Also for this problem, we need the p probability of failure, which we

can calculate from the Lognormal mean life and cov (coefficient of
variation) information. Assuming Lognormal failure behavior of the
machines, write the expression for and calculate the median life value
from the data mean life value. The median is needed as a location
parameter for the Lognormal distribution. State why the median that

2
 you calculate is expected to be different in value from the mean.

So tmed =

d) Then write the expression for and calculate the Lognormal shape
parameter, s. Recall that s is a function of the cov (Coefficient of
Variation) or δ = σ/μ or Std Dev normalized by the mean value, which
is an important metric for dispersion from the mean and for comparing
the widths of distributions with different mean values. Recall the
approximation with generally acceptable error: s2 ~ δ2 when cov ≤ 0.3

So, the Lognormal shape parameter s ~

e) Finally, write the logic expression for and calculate F(T = 900) = p.
Include in your solution a sketch of the Lognormal f(t), pdf, showing
(by highlighting) the area under the curve that represents the
cumulative probability of failure up to 900 hr, F(900) = P(T ≤ 900).
This failure probability is the p needed in the Binomial calculation for
P(x = 2).

3
f) Now write the expression for the Binomial distribution and calculate
the probability of two failures, P(x = 2), among the 5 available
machines. Use the
F(t = 900) as the Binomial p value for failure. Also, write the
expressions for and calculate the Mean value, Variance, Std. Dev.,
and cov for the Binomial pmf.

⎛ 5⎞ 2
P(x) = ⎜ ⎟ p (1− p)3
⎝ 2⎠

p=

P(x = 2) =

Binomial Mean =

Binomial Variance = np(1-p) =

Binomial Std. Dev. =

Binomial Coefficient of Variation, cov, =

g) Write the Binomial expression for and calculate the Reliability of the
system of Five Machines based on the probability of working or 0
failures at 900 hr = R(t = 900).

The system Reliability is the probability of no failures at t = 900 hr:

h) In AgenaRisk, create a node variable with Node Name “Five

Machines”, Node Type = Integer Interval, select Simulation Node,
NPT, Binomial, n = 5, p = value calculated above. For the Binomial
pmf image, run the cursor over the distribution image and find the pmf
Mean, Std Dev, and compare these values with the values calculated
above. Include a screen capture of the Binomial pmf distribution in
your HW solutions.

4
 From the AgenaRisk Binomial pmf distribution: Mean = …, and Std. Dev. = …,
which agree with the calculations above.

i) From the pmf image, report also the Binomial pmf (probability mass
function) median value and explain why, based on the Binomial pmf
character, the median is smaller than the mean value.

5. Operational Risk Case Study, Computer Controlled Train System

This computer-controlled system has been designed with significant resilience such
that if the computer function fails or the computer power supply fails, a human driver
can take control of the train. A top-level failure of the system occurs when both the
computer function AND the manual or human function fail. The computer fails if its
CPU unit fails or its I/O unit fails.

The computer power subsystem also was designed to be resilient by including three
relatively independent and approximated as independent for this case, identical
power supply units, so that the power supply failure occurs only if at least 2 out of 3 of
the power units fail. So 2 out of 3 power units are needed for sufficient and the third
unit provides redundancy and increases resilience of the system, so that if one unit
fails the third unit increases the probability of 2 out of 3 minimum working units for
success. In AgenaRisk, open the Fault Tree model (included with the Homework) for
RDBN Figure 11.6, p. 351.

a) Verify that the initial state of the fault tree model BN for the computer-controlled
train system is the same as shown in Figure 11.6, p.351. Write a verification
statement in your homework paper.

5
Verification is based on the same failure probabilities shown in the provided
FT.cmp file as shown in RDBN Fig. 13.6, p. 433, and shown below:

Note that each of the variables in the BN represents the failure probability of
each modeled system component. This BN is based on the Fault Tree for
the same computer-controlled train system shown in Figure 13.5, p. 432,
shown below:

6
b) Note that AC power failure (Power Failure = True) for the train system
requires failure of at least 2 of the 3 power supplies (power success
requires at least 2 working supplies). The probability of success of 2
out of 3 independent units (or k units needed out of n total) is easily
calculated using the Binomial distribution, discussed in Unit 4, where n
is the number of units, x is the number of successes, and n-x is the
number of failures. Recall how to calculate the Binomial coefficient,
which is the number of contributions to the total probability for a given
n, x. First, the probability of x out of n units working is:

⎛ n⎞ x n -x
P{X = x} = P(x) = ⎜ ⎟ R (1- R)
⎝ x⎠
Work through the example below and check the calculated result
below to review the Binomial distribution and practice its use:

A space vehicle requires 3 working engines out of its total of 4 main

engines to achieve orbit. If each engine has a reliability of 0.97, write

7
 the expression for and calculate the system reliability of achieving
orbit using the Binomial distribution. Comment on the difference
between the reliability of each engine and the system Reliability of the
4 engines working together. Also state the primary approximation or
assumption in using this method to calculate the system Reliability.

c) Show how the BN model of the computer-controlled train system can

be used for Diagnostic Reasoning to identify not only the causes of
failure but also the most likely cause or explanation (MLE). When an
observed failure of the Computer System occurs, the Manual Override
provides access for a human operator to take control followed by the
success of the Manual Operator in taking control and Manual Backup
Failure, now observed to be False.

In setting up or modifying Bayesian networks, we can add one or

more arcs if the evidence shows dependency with one or more
additional causes or network nodes, as discussed with examples in
class.

Note that the Top Event System Failure probability node is 0.0125 or
1.25 %. Right click on the Computer System node and Enter
Observation Failure = True. Recalculate the BN and state in your
answer the change in the System Failure probability value for True.
Include a screen capture of your BN with this entered observation in
your homework paper. Comment on other changes in component
probabilities following the update due to the Observation. Did the
Manual Backup node variable change? State why or why not.

8
Given the observation of Computer System failure, the System Failure
probability has increased from …..

d) Now, right click on the Manual Backup node and Enter Observation
Failure = False. Recalculate the BN and confirm that the System
Failure probability reduces to 0, because the probability that the
Manual Backup succeeded = 1. State in your homework the resulting
nominal probability that the Manual Backup succeeds (does not fail).

e) Your team uses Diagnostic Reasoning to investigate the causes of the

observed Computer System failure and discovers that Power Supply 3
is working. So, enter this observation, Power Supply 3 = False with
the other 2 previous observations still entered (Computer System
failed, Manual Backup working), as in Figure 11.7, p. 351.

Recalculate the BN and state the most likely cause (most likely
explanation, MLE) of System Failure based on the 3 entered
observations, Computer System failure = True, Manual Backup =

9
 False, and Power Supply 3 = False, and the failure probabilities of the
components. This diagnosis of the highest component failure
probability is to identify the maximum a posteriori (MAP) probability by
updated prior probabilities as we have performed using a Bayes
model and we have now performed by updating the BN.

Include a screen capture of your BN with these 3 entered

observations for your homework paper.

f) Now your team tests the CPU and observes that it is working. So
based on this 4th observation with the other entered observations
intact, enter CPU = False, and recalculate the BN. Identify and state
the MAP, most likely explanation, MLE, of Computer System failure,
based on the 3 observations to update the BN probabilities.

Include a screen capture of your BN with these 4 entered

observations for your homework paper.

10