Alarm Management: From 2,000 Alarms a Day to a System Operators Trust

Alarm floods are not an operator problem; they are a configuration problem. Benchmarks from ISA-18.2 / IEC 62682 say a healthy system averages about one alarm per operator per ten minutes. Most SCADA systems we audit in Vietnam run ten times that, and operators acknowledge in bulk without reading — including, one day, the alarm that mattered.

Why systems drift into flood

Every integrator adds alarms "to be safe": each transmitter gets HI/HIHI/LO/LOLO by default, each device gets a communication-failure alarm, every pump gets a "not running" alarm even when it is stopped on purpose — and nobody ever removes anything. Ten years later the alarm list is a log, not an alarm system.

The rationalization workflow

1. Inventory. Export every configured alarm with its priority and its frequency over the last month. Most SCADA platforms can do this; if yours cannot, we extract it from the alarm database.
2. Kill the noise first. The ten most frequent alarms are typically more than half of all annunciations. Fix their deadbands, delays and chattering signals before touching anything else.
3. Ask the ISA question of every alarm: what action must the operator take? If there is no action, it is not an alarm — log it as an event instead.
4. Set priority by consequence and time to respond, not by instrument type. A high-high level in a chemical tank with a 2-minute response window is high priority; the same level in a buffer tank with hours of margin is low.
5. Document each alarm — cause, consequence, operator response — in a master alarm database. This is the document auditors and new operators will ask for.

Configuration mechanics that matter

• Deadband on analog alarms: 1–5 % of range, so a value sitting at the limit does not chatter.
• On-delay for transient-prone signals: 3–15 s, so a pressure spike during pump start does not alarm.
• State-based suppression: do not alarm "low flow" when the pump is intentionally stopped; do not alarm "communication lost" on a device that is switched off for maintenance.
• Shelving with automatic timeout instead of permanent inhibits that nobody remembers to remove.
• Alarm priority distribution target: roughly 80 % low, 15 % medium, 5 % high.

Measure it monthly

Track alarms per operator-hour, the top-ten offenders, standing alarms (active for more than 24 hours) and the priority distribution. Alarm health degrades silently as plants change; a one-page monthly report keeps it honest.

Where the hardware helps

Smart instruments with HART or IO-Link diagnostics let you distinguish "process out of range" from "sensor failing", so the two can be prioritized differently. Managed switches with SNMP make network alarms specific — which port, which link — instead of a blanket "communication lost". Both are cheap compared with the operator attention they save.

We deliver alarm rationalization as a service on Siemens WinCC, Rockwell FactoryTalk, Ignition and other platforms, including the master alarm database — see engineering services. Related: SCADA vs HMI.

Quick answers

How many alarms per operator is acceptable?

ISA-18.2 and EEMUA 191 benchmarks: about one alarm per 10 minutes per operator in normal operation (around 150 per day) is manageable; more than 10 per 10 minutes sustained is a flood in which important alarms are missed.