Chapter 1 · Lesson 1.1
Safety barriers: roles, PFD and confidence level
A safety barrier reduces the probability or the severity of a scenario. Its performance is measured by its probability of failure on demand (PFD) or by its confidence level (CL).
Four possible roles
Elimination
A design change makes the scenario disappear (substitution of a product, inventory reduction).
Prevention
The barrier reduces the probability of occurrence of the initiating event.
Control
The barrier limits deviations so as not to lose control of the process.
Protection
The barrier limits the extent of damage by protecting the target from hazardous phenomena.
On the timeline of an accident: normal conditions, then deviation, loss of control, hazardous phenomenon, impact on targets and damage. Elimination and prevention act upstream of the deviation, control before the loss of control, protection after the hazardous phenomenon.
Implementation of a barrier
- domino effect physical object (guard, containment bund, relief valve): technical barrier (TB).
- A human resource (checking a device, operator intervention): human barrier (HB).
- A specific organisation (working in pairs, lock-out, hot work permit): organisational barrier (OB).
- It can also be expressed as a fonction (temperature limit, staff training).
Classification
| Family | Sub-family | Examples |
|---|---|---|
| Barriers passives (static material barriers) | Walls, bunds, slabs | Containment bund, earth berm, fire wall |
| Immaterial, symbolic barriers | Instructions, marking, signage | |
| Mixed barriers | ||
| Barriers actives | Safety instrumented systems (SIS) | Sensor, safety PLC, shut-off valve |
| Active material barriers, non-instrumented | Relief valve, rupture disc, check valve | |
| Manually operated systems, human barriers | Emergency stop, operator intervention on alarm |
The term “barrier” comes from the source-flow-target model: a barrier prevents the flow generated by the source from reaching its target or reduces it. On the bow-tie, prevention barriers are placed to the left of the top event (TE) and protection barriers to its right.
Equivalence with the fault tree
A barrier in a bow-tie is equivalent, in a fault tree, to an AND gate between the initiating event (or the effect) and the failure of the barrier: the scenario takes place only if the initiating event occurs et and the barrier fails.
Ps = Pe × PFD = Pe × 10−CL
- Pe : probability (or frequency) of the event at the barrier input;
- Ps : probability (or frequency) at the output, barrier failed;
- PFD: Probability of Failure on Demand, probability of failure on demand;
- CL: confidence level, exponent of the PFD.
Link between PFD, confidence level and risk reduction
| CL | Average PFD | Risk reduction (RR) |
|---|---|---|
| 4 | 10-5 ≤ PFDavg < 10-4 | 10,000 < RR ≤ 100,000 |
| 3 | 10-4 ≤ PFDavg < 10-3 | 1,000 < RR ≤ 10,000 |
| 2 | 10-3 ≤ PFDavg < 10-2 | 100 < RR ≤ 1,000 |
| 1 | 10-2 ≤ PFDavg < 10-1 | 10 < RR ≤ 100 |
| 0 | 10-1 ≤ PFDavg < 1 | 1 < RR ≤ 10 |
Worked examples
- A top event has a frequency of 10-2 per year. A protective barrier of CL 2 (PFD = 10-2) brings the scenario frequency down to 10-2 × 10-2 = 10-4 per year.
- Two independent barriers of CL 1 and CL 2 in series (both must fail): 10-2 × 10-1 × 10-2 = 10-5 per year. The confidence levels add up (1 + 2 = 3, hence 10-3 reduction).
Assessing the relevance of a barrier
- Suitability and independence with respect to the other barriers and to the initiator.
- Robustness for safety use.
- Effectiveness in the context of use and for a given operating time.
- For an active device: response time appropriate to the kinetics of the scenario.
- Confidence level (CL) or probability of failure on demand (PFD, Probability of Failure on Demand).
This reasoning requires barriers that are independent of one another and of the initiating event (no common cause), and an average PFD that is valid over the period considered (periodic tests). A barrier that depends on the same power supply or the same operator as the cause cannot be credited as independent.
PFD = 10−CL. Each confidence level divides the probability of the scenario by 10. A barrier counts in the calculation provided it is independent, effective and tested.
To decide quelles barriers to add and with what confidence level, the LOPA method (lesson 1-2) generalises this calculation to all the protection layers of a scenario.