The initiating event frequency is usually the largest single number in a LOPA. Most companies keep an approved table; where it does not cover a cause, the engineer builds the value from data and the company's approval process.
From equipment failure history
IEF = number of relevant failures / years of operation of the population. Example: 6 control valve failures causing high pressure across 30 similar loops over 10 years gives 6 / (30 × 10) = 0.02 per loop-year. With few events, use an upper confidence bound instead of the point estimate.
From tasks and human error
IEF = tasks per year × probability of the error per task. Example: 100 tank transfers per year, line-up error probability 1 × 10⁻³ per transfer: IEF = 0.1 per year. The error probability should come from the company's human reliability guidance, considering procedures, checks and stress.
Things to check
- Is the cause specific? "Control valve fails open" and "operator closes outlet valve" are different scenarios.
- Is there a hidden enabling condition (a cause that only matters during start-up)?
- Does the history include near misses? They count as causes even if a layer stopped the consequence.
- Is the population representative (service, maintenance regime)?
In First Out AI
Every LOPA scenario has a "Source of the IEF" field. The SRS draft quotes it, and the AI flags scenarios where it is empty.
Written for First Out AI as engineering guidance. It supports engineering work and does not replace it; check every value against your standards, certificates and company rules with a competent functional safety engineer.