Understanding Expected Operating Ranges

Why this lesson matters

Structured readings tell you what the value is. An expected operating range tells you whether that value is acceptable. This is the point where monitoring stops being record keeping and starts being condition assessment, and it is also the point where most sites accidentally destroy the credibility of their own alerts.

What a range is, and what it is not

An expected operating range is an engineering statement: under defined conditions, this asset should read between a low bound and a high bound, and a value outside those bounds warrants attention.

It is not:

• a tolerance band drawn around a historical average
• the same as a safety trip or relief setting
• a fixed number that applies in all plant states
• a substitute for the manufacturer's design data

Get this wrong and you build an alert system that cries wolf. Once a range produces routine false exceptions, people stop reading the alerts, and from that point the monitoring has negative value because it creates a false sense of coverage.

The four inputs

1. Design and manufacturer data. The commissioning record, the nameplate, the O and M manual, the pump curve, the burner setup sheet. This is the primary source and it is the one most often not consulted.

2. Statutory and safety settings. Relief valve set pressure, high limit thermostat, trip points, regulatory limits. These are boundaries you must sit inside, not targets. Your expected range should be comfortably inside them, so that an exception gives you time to act before a trip.

3. Instrument uncertainty. Your range cannot be tighter than the instrument can resolve. For a pressure gauge to EN 837-1:

permitted error = accuracy class % x full scale span

A class 1.6 gauge on a 0 to 16 bar span: 1.6% x 16 = +/- 0.256 bar. Add reading uncertainty from parallax and interpolation, typically half a graduation. Your practical range must be wider than the sum of those.

This also explains why a gauge operating at 10 percent of its full scale is a poor measurement: the permitted error is a percentage of span, not of reading, so at 1.6 bar on a 16 bar gauge the error is 16 percent of the value you care about. Size instruments so the normal operating point sits in the middle third of the scale.

4. Observed behaviour. Once you have a few weeks of structured readings you can characterise how the asset actually behaves, separated by plant condition. Observed behaviour refines a range derived from the first three inputs; it should not be the only input.

Ranges are conditional

A single pair of numbers rarely fits a real asset. The value depends on state:

• Plant condition: running, idle, cold start, shutdown, standby.
• Load: a compressed air header pressure at full production and at night are different populations.
• Ambient conditions: chilled water and condenser circuits shift seasonally; so does gas consumption in a heating application.
• Time of day and occupancy in buildings.
• Position in a maintenance cycle: a filter differential pressure legitimately rises across its service interval. The useful limit is not a static number but a rate of rise and a replacement threshold.

The practical approach is to define a small number of named conditions per asset with their own ranges, and to record the plant condition with the reading. If the capture process cannot record condition, keep one wide range rather than a narrow one that is only valid in one state.

Static limits versus rate of change

Some of the most valuable signals are not limit breaches at all.

Rate of change catches drift long before an absolute limit is reached. A pump discharge pressure falling 0.05 bar per week is a strong impeller wear or recirculation signal, and it may be a year away from breaching a low limit.

Baseline shift is the clearest consumption signal available. Overnight gas or water use rising from a stable floor indicates a continuous load or a leak, even while the total remains within any plausible range.

Variance change matters for instantaneous values. A pressure that used to be steady and now oscillates between the same bounds is a changed condition even though every individual reading is in range.

For cumulative meters, absolute limits on the register index are meaningless. Set limits on the derived rate instead: consumption per day, normalised to elapsed time, and optionally normalised further to production units, degree days or occupancy.

Setting a range: a worked sequence

For a boiler house gas meter, read weekly:

1. Establish the derived quantity: daily consumption in m3 per day, from the register delta and elapsed time.
2. Gather design input: burner rated input and expected duty hours give a theoretical maximum.
3. Collect twelve weeks of structured readings, tagged by season and production level.
4. Compute the distribution of daily rate, separated by heating and non heating periods.
5. Set the high bound above the observed heating season maximum plus a margin, because exceeding it indicates a genuine anomaly rather than a cold week.
6. Set the low bound above zero but below the summer minimum, so that a stuck register or a missed meter exchange is caught.
7. Set a separate overnight baseline limit if you can obtain a non production reading. This is usually the most diagnostic limit on the asset.
8. Record the justification for every number against the asset, with the date and the person who set it.

That last step is the one that gets skipped, and it is why nobody six months later is willing to change a limit: they do not know why it was set.

Reviewing ranges

A range is a live engineering parameter, not a configuration item set once.

• Review after any plant modification, control change, meter exchange or instrument recalibration.
• Review seasonally where the asset is weather dependent.
• Review any limit that generated more than a handful of exceptions in a period, because either the plant has changed or the limit is wrong, and both require action.
• Review any limit that has never generated an exception in a year, because it may be set so wide it can never trigger.

Keep a record of range changes with their justification. Auditors and incoming staff both need it.

Distinguishing exception types

Not every out of range value means the same thing, and treating them identically is what makes alert fatigue inevitable. Separate:

• Operational exception - the asset is genuinely outside its expected range. Something is wrong with the plant.
• Data quality exception - the reading is missing, late, unreadable, estimated, or implausible given the previous value. Something is wrong with the data.
• Instrument exception - the reading is impossible for the instrument, such as a value above full scale or a negative cumulative delta. Something is wrong with the measurement or the entry.

These go to different people and drive different actions. A maintenance supervisor needs the first. A monitoring or reporting owner needs the second and third.

Common mistakes at this stage

• Deriving limits only from historical average and standard deviation, with no reference to design data. This encodes whatever the plant was doing at the time, including existing faults, as normal.
• Setting limits tighter than instrument uncertainty. Guaranteed false exceptions.
• Using a single range for all plant conditions.
• Applying absolute limits to cumulative register values instead of to the derived rate.
• Setting the expected range equal to the safety trip setting, which removes all warning time.
• Not recording the justification, which freezes the limits permanently because nobody dares change them.
• Leaving noisy alerts running. An ignored alert is worse than no alert, because it produces the appearance of monitoring without the function.