Every PFDavg is only as good as its failure rates. A reviewer's first questions are always: where does this number come from, and does it apply here?
Sources, in order of preference for a specific device
- Device certificate and safety manual, with failure rates for the defined use and the conditions of validity.
- FMEDA report for the device, giving λSD, λSU, λDD, λDU (sometimes also "no effect" and annunciation failures) and the assumptions behind them.
- Your own field data: failures found in proof tests and in operation, divided by the cumulative device hours. Recorded properly, this is the most relevant data you can have.
- Generic industry databases, used under their licence terms. First Out AI does not include any licensed database; enter the value and cite the source.
Reading an FMEDA
- Check the unit: per hour, FIT (failures per 10⁹ hours) or per year. 100 FIT = 1 × 10⁻⁷ per hour.
- Check what is included: the transmitter electronics only, or the sensor too? The valve body only, or valve, actuator and solenoid?
- Check the assumed diagnostics: λDD is only detected if your logic solver acts on the diagnostic signal (out-of-range current, for instance).
- Check the environment and application: clean or dirty service, ambient temperature, process temperature.
Plausibility checks
- λDU of certified transmitters is typically in the 10⁻⁸ to 10⁻⁷ per hour range; valves with actuators are often higher, around 10⁻⁷ to a few 10⁻⁶. A value of 10⁻⁴ per hour means about one dangerous undetected failure per device per year: almost certainly a unit error.
- Field data that shows far more failures than the certificate says the certificate does not represent your service. Use the field data.
- Failure rates should come with a source per group. First Out AI lists every source in the report and the SRS.
Building your own data
Count dangerous failures found at proof tests per device type and service, and device-years in service. With n failures over T device-hours, the point estimate is n/T; with few failures use an upper confidence bound. Feed it back at each periodic review of the verification.
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.