Alarms Aren't Controls: Interlock It When It Matters
An alarm is not a control. If a piece of equipment can eat itself in under a minute, no operator is going to save it with an acknowledge button. We designed it that way, then we blame ops for “not reacting.” That’s on us.
Example: a transfer pump kept losing bearings every few weeks. Suction strainer blinded, HMI threw low suction and high vibration. Night shift throttled a valve and hoped. Pump still cavitated to death. Well, there’s your problem. We added a DP across the strainer, blocked start if DP was high, and tripped the pump on a fast DP rise with a 5 s debounce and rate-of-change check. Also added a small bypass relief to avoid deadhead. Zero failures since, and the alarm list is quiet.
Not every trip belongs in logic. Pick interlocks by consequence to people/env, asset cost, and realistic operator response time. Then tune: hysteresis, latching with manual reset, and proof-test it.
Where do you draw the line between alarm-only and hard interlock? What logic blocks saved you the most grief?
One way to draw the line by method rather than feel: if the function is what stands between the initiating event and harm to people or the environment, and you take credit for it in a LOPA, it's a safety instrumented function under IEC 61511 / ISA 84. That means a SIL target, independent sensor and final element, a proof-test interval, and bypass management. If it only protects the asset, it can live in the BPCS, but IEC 61511 limits a BPCS protection layer to a risk reduction factor of no more than 10, and you generally can't credit a BPCS interlock and an alarm on the same BPCS as two separate layers.
For alarm-only, the test is operator response time against the process safety time. Your bearing case fails that outright. Commonly cited CCPS LOPA guidance only credits operator response to an alarm when there's roughly 10 minutes or more to detect, decide, and act, with a documented response and training. Anything faster belongs in logic.
ISA-18.2 rationalization is a good place to record that call for each alarm, so the alarm-versus-trip decision is written down with its reasoning and doesn't get relitigated every shift.
For alarm-only, the test is operator response time against the process safety time. Your bearing case fails that outright. Commonly cited CCPS LOPA guidance only credits operator response to an alarm when there's roughly 10 minutes or more to detect, decide, and act, with a documented response and training. Anything faster belongs in logic.
ISA-18.2 rationalization is a good place to record that call for each alarm, so the alarm-versus-trip decision is written down with its reasoning and doesn't get relitigated every shift.