Alarms people act on rather than silence
An alarm system that cries wolf is worse than none, and the failure is almost always in the design rather than the operator.
Short answer
An alarm should require an operator action, and anything that does not should be an event or a log entry instead. The standard failure is alarm flood: hundreds of alarms from one root cause, which trains operators to acknowledge everything without reading. Deadbands, delays and suppression of consequential alarms fix most of it.
The test for an alarm is whether an operator can and should do something about it. Almost every bad alarm system fails that test in the same way.
The three faults
- Chatter. A value hovering at a threshold with no deadband annunciates every scan. Two percent of hysteresis fixes it.
- Transients. A pressure dip during a valve change is not a fault. An on-delay of a few seconds removes most of these.
- Flood. One root cause produces fifty alarms. The operator acknowledges all fifty without reading any, which is the behaviour the design taught them.
Suppression is not hiding
If a pump has tripped, its low-flow alarm is a consequence, not information. Suppressing consequential alarms while their cause is active is the single highest-value change in most systems.
The measure that matters is alarms per operator per hour during an upset. If it exceeds what one person can read, the system is not communicating, it is broadcasting.
Priority means something or nothing
Three levels, used honestly. If everything is high priority then nothing is, and operators learn to treat the colour as decoration.
Common questions
- What makes a good alarm?
- It requires an operator response, it arrives in time for that response to matter, and it identifies a root cause rather than a consequence. If an operator can do nothing about it, or if it always arrives with fifty others, it is not an alarm; it is noise wearing an alarm's colours.
- What is alarm flood and how do you prevent it?
- Alarm flood is many alarms arriving from one root cause, faster than an operator can read. Prevent it by suppressing consequential alarms when their cause is already alarming, adding on-delays so transients do not annunciate, and using deadbands so values hovering at a threshold do not chatter.
Keep reading
- Safety
SIL or PL: which one does your machine need?
Two standards, two scales, and one machine. Which one applies, how they map to each other, and why the answer is usually ISO 13849.
- Safety
Categories B, 1, 2, 3 and 4, in plain terms
Five architectures, what a single fault does to each, and the practical wiring that goes with them.
- Safety
Safety relay or safety PLC: how to decide
One is a wiring decision, the other is a programming one. The count of safety functions, not the size of the machine, is what settles it.