Estimation of An Event Occurrence for LOPA Studies - … · Estimation of An Event Occurrence for...
Transcript of Estimation of An Event Occurrence for LOPA Studies - … · Estimation of An Event Occurrence for...
1
Estimation of An Event Occurrence for LOPA Studies
Randy Freeman S&PP Consulting Houston, TX 77041 713 408 0357 [email protected]
2
Problem
Your LOPA team members tell you that the initiating event of concern has never happened in the history of the process unit. No bad events in 20 years. What do you do?
Ignore the event and move on – It can never happen
What else?
3
Available Methods to Handle Problem
Assume event has happened once, M1 = 1/N
Assume event has almost happened M2 = 1/2N
Bayesian Results M3 = 1/3N M4 = 1/4N
4
Methods (cont)
Poisson Arrivals of Failures At 90% limit
M5 = 0.105/N = 1/10N
At 10% limit M6 = 2.303/N
5
Methods (cont.)
Chi Square Confidence Limit (95%)
M7 = λ = χα;2 / (2 n) = 5.991/(2 n)
Chi Square Confidence Limit (50%)
M8 = λ = χα;2 / (2 n) = 1.3863/(2 n)
8
Methods (cont.)
Hypothesis Test – Normal distribution with 95% confidence
M11 =
Zα = 1.6445
)( 2
2
α
α
znz+
10
1.00E-03
1.00E-02
1.00E-01
1.00E+00
0 10 20 30 40 50 60 70 80 90 100
Nunber of Trials, N
Failu
re R
ate,
fail/
yr
. M1
M2M3M4M8M9M10
Legend
M3
Best Guess Methods
11
1.00E-03
1.00E-02
1.00E-01
1.00E+00
1.00E+01
0 10 20 30 40 50 60 70 80 90 100
Number of Trials, N
Failu
re R
ate,
fail/
Yr
.
M5M6M7M11M12
Legend
M12
M5
M7
M6
M11
Confidence Limit Methods
13
Recommendations for Basic LOPA
Determine if the event of concern is physically possible. If not physically possible, delete from LOPA analysis
Make sure you have a minimum of 10 years of data. If not use values from LOPA book
Use M1 = 1/N for events found to be physically possible
14
Recommendations for Beyond Basic LOPA Reviews
Use estimators M3, M8 or M12.
M3 (Rule of 1/3N) is easier to remember and calculate.
15
Example – Case 1
Data No occurrences in three years What frequency should be assigned? Answer Since the history is less than 10
years, use the frequency presented in LOPA
book.
16
Example – Case 2
Data One occurrence in ten years What frequency should be assigned? Answer Methods presented in this paper do not apply when an event has occurred. Start with a frequency of 1/10 and review value in LOPA book.
17
Example – Case 3
Data No occurrences in ten years What frequency should be assigned? Answer Determine if history is “valid” Start with a frequency of 1/10 If LOPA at greater than order of
magnitude significance, use 1/30
18
Example – Case 4
Data No occurrences in 30 years What frequency should be assigned? Answer Determine if history is “valid” Start with a frequency of 1/10 If LOPA at greater than order of
magnitude significance, use 1/90
19
Example – Case 5
Data No occurrences in 20 years What frequency should be assigned? Answer Determine if history is “valid” No history of pump overpressure of PRV
PRV set P = 150 psig Pump Deadhead P = 100 psig
Not credible scenario – Delete from LOPA Study
Can it really happen?
We pressurize a vessel to 130 psig. The MAWP is 100 psig.
What happens?????
Does the vessel blow up????
20
More Information
Vessel in good condition per API 579 – Fitness for Service
Test Pressure = 1.5 * MAWP Test Pressure = 150 psig Piping is 2 inch schedule 40 with a
catastrophic burst pressure of 7000 psig
ANSI Class 150 Flanges with a MAOP of 285 psig at 100 F
22
ANSWER – NOTHING HAPPENS!
No rupture No weld tear No flange failure No release Nothing happens!
Maybe an ASME code violation
requiring an inspection for fitness for service
23
Lessons Learned
Before starting detailed LOPA analysis verify that the scenario of interest can actually happen.
Physics and Chemistry are powerful tools to sort out imaginary from real safety issues.
24
25
Conclusions
There is no single “correct answer” Use an estimator that is consistent
with the other assumptions in your LOPA study
More Details
“What to Do When Nothing Has Happened?”, pp 204- 211,Process Safety Progress, Vol. 30, No. 2, September 2011
Published online in Wiley Online Library (wileyonlinelibrary.com). DOI 10.1002/prs.10463, 2011
27