The document proposes a new process safety KPI (key performance indicator) using two metrics: (1) initiating event demand rate and (2) safeguard unavailability. This aims to better predict major losses compared to typical lagging metrics. The KPI can be automatically collected from process data and presented to management to identify unsafe areas and guide corrective actions. An example shows how a CEO and plant managers drilled down through the KPI data to identify and address frequently failing level control increasing demand rates. The KPI approach allows proactive major loss prevention through improved management oversight of process safety performance.

1. The New Process Safety KPI
Edward M. Marszal, PE, CFSE
President, Kenexis Consulting Corporation

2. Speaker Information
Edward M. Marszal, PE, CFSE
President, Kenexis
13 Years Experience
ISA Author “SIL Selection”
ISA Committees - S84, S91, S18
ISA Safety Division Director
ISA, AIChE, NFPA Member
BSChE, Ohio State University

3. Introduction
Process industry safety performance can be improved
• Major Hazards Still a Problem
Upper management disconnected from operations
• No “feel” for day-to-day operation
• Important information not available (can be hidden)
Actionable metrics allow oversight
Typical safety metrics not effective
• Slips, trips, and falls not well correlated to major losses
Better Metrics Essential
• Predictive information can be developed from process history
• Collection and presentation possible with existing tools

4. Management 101
Improved major loss prevention though improved
management
Management process
• Determine the objective
• Identify parameters affecting the objective (KPI)
• Measure performance against the parameters
• Adjust processes, procedures, and equipment to optimize KPI
values

5. Current Metrics
Process industry safety management is based on metrics
• First Aid Cases
• Reportable Injury Rate
• “Near-Miss” Report Rate (good, but lagging, infrequent, and
reliant on human reports)
• Unsafe Activity Reports
• Unresolved PHA Action Items
Current metrics not well correlated with major losses

6. Metrics Desirable Attributes
New metrics are essential to improvement
New metrics fore-shadowed in recent standards
Must “predict” major loss issues
• Could lead to major loss, but stops short
Must be clearly defined and consistently applied
Must be relatively frequent events
Automatic collection and reporting beneficial

7. Accident Causation Model
Hypothesis: Most major accidents happen because
multiple failures occur; starting with an initiating event
Failure Failure Failure Accident
Initiating Event Propagating Event Propagating Event

8. The New KPI
Major loss prevention distilled to two numbers
Initiating Events (Demand Rate)
• Actual/expected frequency of occurrence
Safeguard Unavailability
• Probability safeguards will operate on demand
Metrics applied at all levels, and allow “drill-down”
• Organization -> Site -> Unit -> Hazard

9. Demand Rate
Scaled Demand Rate – Actual/Expected Frequency
• Target 1.0 – Higher is unsafe
Expected demand rate obtained from existing PHA
• Layer of Protection Analysis (LOPA) requires this data
Can be automatically logged, tracked, and reported
• Demands typically historized
• Critical alarm activations
• Safety instrumented function activations
• Pressure relief alarms

10. Safeguard Effectiveness
Safeguard Scaled Unavailability –
Actual Unavailability/Target Unavailability
• Greater than 1.0 is unsafe
Unavailability – fraction of time a safeguard can not perform
is intended function
• Safeguard equipment is in failed state
• Safeguard is bypassed
Required safeguards listed in good PHA (e.g., LOPA)
Data collection and reporting (combined system)
1. Functional test of equipment (database logged)
2. Time in bypass – historized

11. Example
A CEO receives safety KPI as part of a monthly briefing
Actual Target
Scaled Demand Rate 1.413 1.0
Scaled Unavailability 0.877 1.0

12. Example
The CEO drills down into the data to determine the source
of the problem
Facility Scaled Demand Rate
Chemical City 3.45
Port Process 0.877
Polymer Valley 0.351
New Chemical City 0.798
The CEO places a call to the chemical city plant manager

13. Example
The chemical city plant manager goes online and views his
plants data
Process Unit Scaled Demand Rate
Utilities 0.694
Monomer Preparation 0.887
Gas Preparation 10.40
Polymerization 0.899

14. Example
The chemical city plant manager then drills down into the
data, in conference with the operations supervisor and
unit engineer
Hazard Scaled Demand Rate
Separator High Alarm 0.100
Separator Low Shutoff 0.887
Separator Relief 35.00
Liquid Pump Shutoff 0.899

15. Example BPCS
LIC-101
The chemical city team review
the separator low level
shutoff
V-101
LT-101 LT-001
To Low
Pressure
Separator
LV-001 LV-101

16. Example
Consequence (Risk Matrix )
Tag
Item IPF Description Hazard Prevented Severity
No. Consequence
- Safety
1. LSC- High Pressure Low-Low level in the High pressure in 4
001 Separator Low- separator vessel may result downstream vessels may
Low Level Closes in gas blowby of high result in overpressure and
Separator Liquid pressure gas into rupture of the equipment.
Outlet Valve downstream equipment that This could lead to release of
is not rated to withstand flammable material to the
that higher pressures. atmosphere with subsequent
fire or explosion potentially
resulting in injury or fatality
D e s c ri p ti o n o f re q u i re d ac ti o n
Likelihood
Independent Protection Layers Required
Category Selected
Initiating Event SIL (All
IPL IPL SIL
L RR IPLs Causes)
Category Credit
1. Failure of level 4 3 1. Operator Operator 1 SIL 2 SIL 2
controller LIC-101 intervention Intervention
such that valve based on
LV-101 fails to low level
the open position. alarm.

17. Example
The team determine that the expected demand rate of once
in ten years was being significantly exceeded because
level control was very frequently failing.
The level control scheme and equipment was replaced to
reduce the initiating event

18. Conclusions
Current typical process safety metrics are not adequate
New KPI should predict major losses
Typical accidents can be predicted by process demands
and safeguard effectiveness
The required KPI can mostly be collected and reported
automatically
The major limitation to this approach is the inability to
measure accidents that are the direct result of the
initiating event