ModulesAlarm ManagementProcess Safety
Alarms worth the operator’s attention.One console’s log, read in minutes, kept to compare.
For control room supervisors and process safety engineers: drop the alarm and event journal you already export, see the load, the floods and the ten worst tags, and give each bad actor an owner. The file is read in your browser; only the summary is kept, and only if you save it.
630 alarms a day reached one operator, about four times the target. Ten tags made almost half of them.
| Measure | Value | Target | Action limit | Reads |
|---|---|---|---|---|
| Per operator position per day | 630 | ≤ 150 | > 300 | Beyond action limit |
| Per 10 minutes, average | 4.4 | ≤ 1 | > 2 | Beyond action limit |
| 10-minute periods in flood | 6.1 % | < 1 % | > 5 % | Beyond action limit |
| Peak in one 10-minute period | 84 | ≤ 10 | > 50 | Beyond action limit |
| Top-ten tags’ share | 47 % | ≤ 5 % | > 20 % | Beyond action limit |
| Stale (active > 24 h), worst day | 9 | < 5 | ≥ 5 | Beyond action limit |
| Priority split, low / medium / high | 41 / 33 / 26 % | ~80 / 15 / 5 % | < 50 / > 25 / > 15 % | Beyond action limit |
One console, one month, three views. An example check; the measures are the product’s, the values illustrative.
Where it connects
On the same loop: Barrier Management and 41 more modules in one workspace.
Method note
Fix the bad actors first, or rationalize first? Both schools are right.
The standards start from a philosophy and a full rationalization. Many sites start from the log, because a handful of tags make most of the noise.
Leandro · Founder, UProcess
One school writes the alarm philosophy and rationalizes every alarm before touching the console, so each change has a documented basis. The other starts with the worst ten tags, because removing chattering and standing alarms is quick and earns the operators’ trust for the long work.
So the check reads the log as it is, and the next check shows what changed.
- One checkA console’s export, read in your browser: the load, the floods, the ten worst tags.
- Monthly reviewThe saved check, with each bad actor given an action or a note, once a month.
- Before and afterTwo saved checks of the same console, compared tag by tag.
Where Alarm Management usually breaks
Four places alarm work goes wrong, and what the check does about each.
The flood arrives when it matters most.
In the upset that led to the 1994 Milford Haven refinery explosion, HSE found an excessive number of alarms in the emergency, which reduced how well operators could respond. A console tuned for normal days fails on the one day it is needed.
Source: HSE, Texaco Milford Haven case study
In UProcess The check counts annunciations per operator position per day, per hour and per 10 minutes, finds every flood period, and lists the worst five with start, end, peak and the first alarms of each.
Floods · Board 3 · 14 Sep 2026
Flood starts above 10 in 10 min
- Start
- 06:52
- End
- 08:31
- Peak
- 84 in 10 min · major
- First three
- PI-1123 · LAH-201 · FAL-310
The nuisance alarm hides the real one.
At a methyl chloride unit in 2010, an alarm with a history of false trips was dismissed as one more, and the release ran for five days. Alarms that chatter or stay in alarm teach people to stop looking.
Source: CCPS Process Safety Beacon, Oct 2017
In UProcess Chattering, fleeting and stale tags are flagged by name, and the top ten bad actors are listed with their count, their share of the load and the pattern behind them.
Top bad actors · Board 3 · Sep 2026
Chattering: 3 or more annunciations of one tag within 60 s.
The alarm you count on is not managed like a safeguard.
At Buncefield the level gauge stuck and the independent high-level switch did not operate, because its test lever had not been secured in the operating position. A protection counted on in the risk study was looked after like any other instrument.
Source: IChemE, Buncefield incident summary
In UProcess In a saved check, when a tag matches a barrier element at the same site, the check shows that element and its status from Barrier Management, next to the tag’s own figures.
LAH-201 · Tank T-201 high level
- In this check
- 1,212 annunciations · chattering
- Barrier element
- BE-0012 · safety alarm
- Status in Barrier Management
- Not assessed
- Site
- North site
In a saved check, matched by tag text at the same site. The check stores no link; Barrier Management keeps the element.
Nobody owns the fix.
After the same 2010 releases, the CSB found that corporate policy gave no clear guidance on dealing with nuisance alarms. A list of bad actors with no owner is the same list next month.
Source: U.S. Chemical Safety Board, recommendation status 2013-30
In UProcess Any bad actor in a saved check can become an action with an owner and a due date. The monthly review records each one as action added or no action, with a note.
Action Tracking · AT-0187
Widen the deadband on PI-1123 and add a 5 s on-delay
- Owner
- Priya Raman
- Due
- 15 Nov 2026
- From
- Alarm Management · Board 3 · Sep 2026
Next check of Board 3: PI-1123 below 100 a month · checker Tom Brennan
Console Board 3 · Tank farm and loadingLast review 8 Oct 2026Reviewer Tom Brennan
| Tag | Aug | Sep | Oct | Decision |
|---|---|---|---|---|
| PI-1123 | 3,120 | 2,904 | 211 | Action AT-0187 |
| LAH-201 | 1,301 | 1,212 | 1,180 | Action AT-0188 |
| FAL-310 | 988 | 1,060 | 1,002 | No action · instrument change planned |
| TAH-415 | 612 | 844 | 790 | No action · bearing replaced 2 Oct |
Counts from three saved checks of the same console. PI-1123: 2,904 → 211 after AT-0187.
The record an auditor asks for.
An alarm review is a habit, not a slide. The record shows that it happened, console by console.
- Each saved check: console, site, period, and the settings and column mapping it used.
- The measures, the floods and the top 50 tags, kept as a summary; the raw log is never stored.
- A monthly review per console: who, when, and each bad actor with an action or a note.
- The actions raised from bad actors, with owner, due date and the check that they worked.
What’s in a check
- Input
- An alarm and event journal export in CSV or Excel, ideally 30 days, mapped column by column. Company mapping templates are saved for the next run.
- Where it runs
- In your browser. The raw file never reaches UProcess; only the computed summary is stored, when you choose to save it.
- Measures
- Annunciations per operator position per day, per hour and per 10 minutes, time in flood, peak 10-minute count, top-ten share, chattering, fleeting and stale tags, and the priority split.
- Targets
- Each measure reads within target, above target or beyond action limit, in words and shape, never by colour alone.
- Output
- The result on screen, a one-page PDF, and, once saved, the summary to compare, review monthly and raise actions from.
Questions
Does my alarm log leave my computer?
No. The file is parsed in your browser. If you save the check, only the summary is stored: measures, floods and the top tags.
Which control systems does it read?
Any export you can map: one column for time, tag, event and priority. A library template is called supported only after a real file in that format has been read.
Does it connect to the DCS?
No. There is no live connection and nothing is written back. You export the journal; UProcess reads the copy.
Can I see a fix work?
Yes. Save a check before and one after, for the same console, and compare them tag by tag.