Professional Documents
Culture Documents
States Were Selected For Each: Arthur M. (Art) Dowell, III, PE Process Improvement Institute
States Were Selected For Each: Arthur M. (Art) Dowell, III, PE Process Improvement Institute
I/P
S-12 • There are other schemes to share process variables
Vent Vent • Tripped states – light green between the SIF and the BPCS but analysis of their failure
Positioner
modes is beyond the scope of this paper.
SA SB SC
1 2 3
• Provision should be made for the BPCS to calculate the
BPCS Sensors BPCS Valve
SIF Sensors SIF Valve A SIF Valve B
Transitions between states mean value correctly when the sensors are in a degraded
• Based on failure rates (and number of sensors that have state
not yet failed) and repair rate. • 1 sensor with bad PV (detected)
Advantages of each independent • Sensor failure rate, λ = 1.67E-2/year • 1 sensor drifted (detected)
configuration • DD Drift, δδ = 0.8* λ = 1.34E-2/yr.
• DU Drift, δυ = 0.1* λ = 1.67E-3/yr. • It is critical that detected failures be repaired within 72
• Median (or mid-value) • 2oo3 voting for safety hours.
• Bad PV, βπω = 0.1* λ = 1.67E-3/yr.
Select for 3 control sensors sensors • Future work: include more explanation for human factors
• Repair rate, µ = 121.67/yr.
• Reduced trips from • Reduced spurious trips (72 hr. MTTR) that introduce errors during testing [4].
transmitters that drift or that • Lower PFD than 1oo1 • Proof Test Interval = 1 year
have bad PV voting
Acronym Definition
1oo1 1 out of 1 voting
Sensors shared between SIF and BPCS 2oo3 2 out of 3 voting
BPCS Basic Process Control System
βπω Bad PV failure rate
Signals CPU Central Processing Unit (or Controller Card)
from SIF DD Dangerous Detected
Sensors to
BPCS (hard
Input Logic Output
Cards Solver Cards
DU Dangerous Detected
BPCS
Glossary
wired)
SIF δδ Drift dangerous detected failure rate
CPU Output
Drift dangerous undetected failure rate
Input
Cards Cards
2oo3 Voting δυ
I/P Current to pneumatic transducer
λ Failure rate, per year
Median Select
or Mid-Value Select mA Milliampere
MTTR Mean Time To detect and Repair, years. 1/MTTR = µ
A B C µ repair rate, per year
SIF Sensors & PFD Probability of Failure on Demand
BPCS Sensors O = unavailability state Pm = Mean probability of a state PV Process Variable
Mean unavailability states (PFD) = 5.6E-4 Mean availability states = 9.99E-1 SIF Safety Instrumented Function
SIL Safety Integrity Level
How to Estimate Relative Risk of Sensors
Shared?
• Fault Tree Analysis could be used, but it is difficult to model
Conclusions Calculations done by
failure modes and failure sequences in time.
• Markov models can handle different failure modes and
•For 2oo3 sensor voting, the Probability of Failure • The Markov Model Module in Reliability
Workbench 13.0.2.0 provided by
failure sequences in time. on Demand (PFD) is 5.62E-4. This is a reasonable Isograph LTD.
• A Markov model was used [3].
• For simplicity, the model was limited to the sensors value for the sensor part of an SIF.
configuration.
•For median select or mid-value select for control
ART DOWELL, III Principal Process Safety Engineer
sensors, the expression PFD = λT/2 is solved for λ. adowell@piii.com
For a 1 year test interval, the failure rate, λ, for the Art has a BA and BS in Chemical Engineering and has more than 50 years of process safety
experience, including 42 years at Rohm & Haas; now The Dow Chemical Company. Art has
control sensors is 1.12E-3/yr, an order of published more than 50 papers, many on PHA, LOPA, and SIS implementation.
PII has helped more than a hundred companies implement process safety since 2003. This
includes leading hundreds of PHAs for both new projects and existing plants and training
magnitude lower than the single sensor failure more than 4000 PHA leaders and scribes. PII staff has performed more than 1000 LOPA and
more than 1000 SIL Verification.
rate of 1.67E-2/yr.