You are on page 1of 62

Pressure Relief Valve

Pop Test Data


Statistical
Replacement Interval Calculation
by
Frederick T. Elder
IRC Research and Technology Forum
February 9, 2007
(c) Frederick T. Elder
When to Replace per IIAR 110
„ After a known relief, and within a reasonable time,
spring-loaded relief valves shall be replaced by new
or remanufactured certified valves. If re-seating is
not complete, replacement shall be immediate.
„ When a component reliability program is in place to
verify relief valve functionality and longevity by
history, testing, disassembly and inspection, and
periodic statistical review of these activities, relief
valves may be replaced at any interval justified by
the findings of such a program. In the absence of
such a program, each relief valve shall be replaced
at the frequency recommended by the relief valve
manufacturer. In the absence of both a component
reliability program and manufacturers’
recommendations, relief valves shall be replaced
every five years if not indicated earlier at annual
inspection.
Why Test?
„ Properly assess health of NH3
refrigeration safety system
„ OSHA has required it in prior
settlement agreements
„ It may save $$$
„ It may answer a PHA question
Maintenance Guidance

„ http://www.valve-world.net/srv/ShowPage.aspx?pageID=640
Outline
„ Background and
Advantages of Weibull
Analysis
„ Failure Criteria
„ Generating Weibull Plot
„ Weibayes Analysis
„ Examples
Background
„ Invented by Waloddi Weibull
in 1937 – he used it for
fatigue life estimation

„ Dr. Robert Abernethy the


modern Weibull Analysis
expert

„ Weibull Analysis first used


extensively in aerospace
applications
Waloddi Weibull 1887-1979
Advantages of Weibull Analysis
„ Main advantage: Small sample size
• Samples may be expensive
• Reduces time/cost of testing
• May not have many recorded failures

„ Weibull Analysis is displayed by an


easy to read graphical plot
Pop Test Failure Criteria

Example – 250 psig valve


Opens at pressures < 242.5 psig – failure
Opens at pressures > 262.5 psig -- failure
Alternate Failure Criteria

„ Reduce the set pressure of relief


valves when possible – then expand
failure definition
„ Do not consider low pressure
opening a failure for those valves
where that does not create a hazard
Weibull Analysis Plot
„ Most Weibull Analysis done from plot
„ To Plot, you need:
• Failure criteria
• Number of failures and times
• Number of suspensions and times
„ From Plot, you get:
• Predicted failure rate
• Failure mechanism
Plotting Data

„ Plot scales
• X axis: Age parameter (Units of Hours in Figure)
• Y axis: Cumulative Distribution Function (CDF)
„ Defines percentage of units that will fail up to an age.
Weibayes Analysis
„ Weibayes is used when there
are no or very few failures:
• Finding the MTTF of a unit after
initial testing lead to no failures
• Redesigned component, several
units tested without failure, is
testing sufficient?
• Smaller sample sizes needed
with Weibayes since previous
failure history is known
Weibayes Analysis
1/ β
„ Weibayes Analysis equation, ⎡ ti ⎤
N β
uses an assumed β η = ⎢∑ ⎥
⎣ i=1 r ⎦
„ Can be used when no failures
have occurred Where:
N = total number of
suspensions and failures
„ Need to have back ground r = number of failed units
failure info β = assumed slope
• Company Weibull library t = time or cycles

• Other Weibull libraries


Weibayes Analysis
„ Relief valve failure data
shows typical β value of 1
• http://www.barringer1.com/
wdbase.htm
„ Weibayes can be used to
determine replacement
interval time
„ Can input data into
Weibull program or
calculate by hand using
equation
Weibayes Analysis
Determine New Replacement Interval
for Test with Zero Failures
1. Gather suspension data
2. Find η (as described in next slides)
3. Find k1-value from One-Failure Plan
table for your assumed β and
number of samples being tested
4. Replacement Interval =η(k1)
Weibayes: Finding η With No Failures
Hand Calculation
„ Confidence Limit Equation for Zero Failures:
• Use:

⎧ ⎫
η = ⎨2∑ Ti β χ 2 ( C ; 2r + 2 ) ⎬ for r ≥ 0
⎩ ⎭

where r=# of failures


Ti=Time of each replacement

{χ ( C; 2 ⋅ f + 2 )}
2
: look up this value from Chi-squared
table for C confidence and 2r+2 degrees of freedom
Weibayes: Finding η With No Failures
WinSMITH Calculation
„ Can Select Specific
Confidence
• Enter number of units, all
as suspensions
• Select Weibayes method
• Choose specific
confidence, 63.2%
confidence equivalent to
assuming 1 failure is
imminent
• Find η from Weibayes plot
Weibayes Example: No Failures
During Testing

„ Parameters: 30 relief valves used for


5 years, 0 failures, want to increase
Replacement Interval
„ Question: How many years can the
valves be used and have at most one
failure with a 90% confidence?
Weibayes Example: No Failures
During Testing
„ Data entered in WinSMITH
• 30 suspensions, 5 year time
• Weibayes method, β=1, 90% Confidence

η=65.14
Weibayes Example: No Failures
During Testing
„ Table of K1-values For One-Failure Test Plans, β=1

„ Read N=30, K1=0.132


„ Complete table and equation to derive K1-values
included in Appendix B
Weibayes Example: No Failures
During Testing
„ Replacement Interval: 65.14(0.132)= 8.6
years
„ So with a 90% confidence, you can replace
the relief valves every 8.6 years and have
at most one failure during that period
„ Reasonable approach: 8.6 years minus 5
years = 3.6 years/2=1.8 years, so add 1.8
years to 5 year zero failure plan to have
reasonable probability of no failures
Weibayes Analysis
Determine New Replacement Interval for
One or More Failures During Testing

„ Most Common
„ Typically, there will be failures
Weibayes Analysis
Determine New Replacement Interval for
One or More Failures During Testing

1. Gather failure and suspension


data
2. Find η (as described in next slides)
3. Find k0-value from Zero-
Failure Plan table for your
assumed β and number of
samples being tested
4. Replacement Interval =η(k0)
Weibayes: Finding η With Failures
Hand Calculation
„ Use Weibayes equation to find η
„ Use: ⎡ ⎤
β 1/
2⋅ f
ηc = η ⎢ ⎥
⎢⎣ { χ ( C ; 2 ⋅ f + 2 )} ⎥⎦
2

to get a specific confidence, where f=# of


failures
„ {χ ( C ; 2 ⋅ f + 2 )} : look up this value from
2

Chi-squared table for C confidence and


2f+2 degrees of freedom
Weibayes: Finding η With Failures
Hand Calculation
„ Chi-Squared Table, C: 90% Confidence
Weibayes: Finding η With Failures
WinSMITH Calculation

„ Enter number of failures, all


with the assumed time of
half the usage time
„ Enter number of suspensions
„ Choose the specific
confidence
„ Find η from the Weibayes
plot
Weibayes Example: One or More
Failures During Testing

„ Parameters: 30 relief valves used for


5 years, 2 failures, don’t know when
failures occurred
„ Question: How many years can the
valves be used and have zero
failures with a 90% confidence?
Weibayes Example: One or More
Failures During Testing
„ Data entered in WinSMITH
• 28 suspensions, 5 year time
• 2 failures, assumed half of 5 years, or 2.5 years
• Weibayes method, β=1, 90% Confidence

η = 27.26
Weibayes Example: One or More
Failures During Testing
„ Table of K0-values For Zero-Failure Test Plans, β=1

„ Read N=30, K0=0.077


„ Complete table and equation to derive K0-values
included in Appendix B
Weibayes Example: One or More
Failures During Testing
„ Replacement Interval:

27.26(0.077)= 2.1 years

„ So with a 90% confidence, you can


replace the relief valves every 2.1 years
and have no failures during the interval
Remember

„ MI of pipes and vessels is also of high priority


„ Relief Valves not to be placed back in service
after testing
„ Need judgment to extend the replacement/test
interval
„ Failed relief valve may never be needed
Where to Buy Weibull Material
„ The New Weibull Handbook and the
WinSMITH software packages can be
purchased at:
• http://www.weibullnews.com/contents.h
tm#Prices
Sources
„ Engineering Safety Relief Systems, March 2006.
by Reindl, D.T, Jekel, T.B., Available from the
Industrial Refrigeration Consortium
„ The New Weibull Handbook, Fourth Edition, 2000,
by Robert Abernethy
„ The New Weibull Handbook, Fifth Edition, 2006,
by Robert Abernethy
„ Fitness for Service of Pressure Relieving Systems,
by W. E. Short II, presented at The 2003 ASME
Pressure Vessels and Piping Conference
„ Reliability Testing of Relief Valves, by Robert E.
Gross, presented at The 2004 ASME Pressure
Vessels and Piping Conference
Sources
„ Plant Guidelines for Technical Management of
Chemical Process Safety, pp 169-172, by Center
for Chemical Process Safety, 1992
„ Armor Swift Eckrich – OSHA settlement
agreement of October 9, 1997
„ IBP – OSHA settlement agreement of 2001
„ Code Requirements for Safety Relief Systems,
Todd Jekel, 2005 Research and Technology
Forum, January 20, 2005
Sources
„ Center for Chemical Process Safety (1998).
Guidelines for Pressure Relief and Effluent
Handling Systems. (pp. 104-107). Center for
Chemical Process Safety/AIChE. Online version
available at: http://www.knovel.com/knovel2/
Toc.jsp?BookID=831&VerticalID=0
„ Center for Chemical Process Safety (1989).
Process Equipment Reliability Data with Data
Tables. P 212
Appendix A:
Weibull Analysis Background
Advantages of Weibull Analysis
„ Weibull Analysis
can be used for:
• Failure Distribution

• Failure Forecasts
and Predictions

• Maintenance
Planning

• Effectiveness of a
Redesign
Weibull Analysis Software
„ WinSMITH Weibull from Fulton
Findings
• http://www.barringer1.com/wins.htm

„ Created by Wes Fulton and Dr. Bob


Abernethey
Weibull Analysis Software
„ Enter age data, suspensions and failures
„ Software will:
• Plot Data
• Calculate Eta, Beta, and PVE numbers
• Run a distribution analysis
• Generate a results report
Plotting Data
„ Age must be known for data
• Standard Life Data: exact age of parts
known
• Interval Data: Age of parts not exactly
know, so parts are grouped
„ Could be from weekly, monthly, etc
inspections
„ Age may be operating time,
starts/stops, etc.
Plotting Data
„ Failures
• Establish failure mode
• Every part displaying
this mode constitutes a
failure
„ Suspensions
• Parts that failed via a
different mode
• Parts that have not yet
failed
„ Early Suspension: Age
below age of first
failure
„ Late Suspension: Age
above age of last
failure
Plotting Data
„ Plot scales
• X axis: Age parameter (Units of Hours in Figure)
• Y axis: Cumulative Distribution Function (CDF)
„ Defines proportion of units that will fail up to an age as a
percentage
Plotting Data

WinSMITH Data Plot


Plotting Data
„ Two-parameter most widely used Weibull
distribution
„ CDF (Cumulative Distribution Function):
β
F(t) = 1- e - (t/η)

• F(t) = fraction failing up-to-time t


• t= failure time
• η = characteristic life
• e = 2.718281, the base for natural logarithms
• β = slope parameter
Plotting Data on Weibull Paper
„ Arrange failures and suspensions in time ascending
order
„ Set up the following table:

„ Fill in Rank and Reverse Rank, and in Time column,


include whether it was a Suspension or Failure
„ If two data points have the same time to failure, they
are both presented in the column, and they will both
get median rank values
Plotting Data on Weibull Paper
„ Use equation to get Adjusted Rank (A.R.):
A.R.=[Reverse Rank X Previous Rank + (N+1)] / [Reverse Rank + 1]

„ Use Benard’s Median Rank formula to get


new Median Rank (since adjusted rank is
not an integer):
Benard’s M.R.=(i-0.3) X 100 / (N+0.4)

„ Fill out previous table, and plot:


• Benard’s M. R. on the y-axis
• Time on x-axis
„ Draw a best fit line through the points,
make sure it is 1:1 Weibull paper
Weibull Example: Preparing
Weibull Plot by Hand
„ Parameters: You are given the
following data, 8 total parts, 5
failures at 49,82,96,30, and 90 hours
and, 3 suspensions at 45,10, and
100 hours
„ Question: At how many hours can
you expect approximately 50% of
the parts to fail?
Weibull Example: Preparing
Weibull Plot by Hand
„ Set up and fill in table:

„ Plot the points on 1:1 Weibull Paper


„ Draw a best fit line through the points and
draw a line across from the 50% mark and
down to the time axis
Weibull Example: Preparing
Weibull Plot by Hand

50% fail by 77 hours


Interpreting the Plot

Eta, Beta, PVE,


and N/S

„ PVE %: Goodness of fit indicator for


Weibull line
• 10% is acceptable, 50% is average
„ N/S: Total number of data points/ Number
of Suspensions
Interpreting the Plot
„ Eta - Characteristic life: Age at which
63.2% of parts will fail
• Parameter most effected by suspensions
Interpreting the Plot

„ Beta – Slope of Weibull line: Failure Mode


• Beta < 1.0 indicates infant mortality
• Beta = 1.0 indicates random failures that are
independent of age
• Beta > 1.0 indicates wear out failures
Interpreting the Plot
„ Use PVE number to evaluate fit of
line

„ Use Beta to evaluate failure method

„ Look for “Bad” Weibull characteristics


Bad Weibull
„ Curved Weibull data
• Origin not at t=0, must
use three-parameter
Weibull
„ Outlying data points
• Look at engineering
aspects of data
recording, test records,
calibrations, etc.
„ Two different slopes of
Weibull data
• More than one failure
mode represented by
data, try to separate
data
Bad Weibull
„ Close Serial Numbers
• Batch problem

„ If PVE number is unacceptable


• Look to different distributions, Log
normal, Three-parameter Weibull

„ Careful, few data points leads to high


PVE number
Failure Forecasting
„ Expected number of
failures that may
occur in a specific
period of time

„ Predicts:
• Future failures when
failed units are replaced
• Future failures when
failed units are not
replaced
Failure Forecasting
„ Additional input needed:
• Age of components in service
• Usage rate
• Introduction rate of new units
• Failed parts replacement info
Appendix B: Weibayes
Extras
Weibayes: Finding η With No Failures
Hand Calculation
„ Assume at Least One Failure is Imminent:
• Use Weibayes equation to find η
• Assume 1 failure (r=1) since a failure is
imminent (yields 63% confidence)
• Use the following table to achieve different
confidences:
Zero Failure Plan Table, β = 1

„ K=[(-1/N)*ln(0.1)](1/β)
Chi Squared Table for Use With
Weibayes Hand Calculations

„ Use 0.10 column for 90% Lower Bound,


0.05 for 95% Lower Bound, etc.
One-Failure Test Plan Table, β = 1

„ (1-Confidence)=(e-(k)β)N+N(e-(k)β)N-1(1- e-(k)β)

You might also like