Before the route changes
Start with how much warning the failure gives.
A bearing on a critical process fan fails three weeks after a monthly route reading that looked normal, a week before the next one was due. At the route review, the first proposal is to move every machine on the route from monthly to fortnightly readings.
That more than doubles the work on the route, from 12 rounds a year to 26, and can still miss a failure that gives less than two weeks of warning. NASA's reliability-centred maintenance guide calls fortnightly and monthly monitoring the most common practice in industry, and gives two reasons for it: a lack of information and poor monitoring techniques.
An interval can only be judged against a particular failure mode. Before the route changes, the team needs three numbers for that failure: how many weeks of warning it gives, how many weeks the plant needs to act, and how many readings it wants in between.
A developing failure passes two points. P, the potential failure, is the earliest point at which your method can detect the fault. F, the functional failure, is the point at which the machine can no longer do its job. The time between them is your warning, usually called the P-F interval. ISO 17359, the ISO guideline for condition-monitoring programmes, uses a closely related term, lead time to failure, and says the monitoring interval depends mainly on the type of fault and how fast it progresses.
Warning varies widely. SKF's bearing failure guide notes that a bearing can go from first damage to unserviceable in a few seconds at high speed, or over months in a large, slow machine.
Your own records are usually the best guide. For each past failure, find how long the first useful sign was visible before the machine had to stop. Treat the answer as an estimate, and plan around the shortest warning you have seen, not the average. Where a failure has safety or environmental consequences, the US Navy's reliability-centred maintenance manual, NAVAIR 00-25-403, specifically asks for the shortest P-F interval in the range. If you have no history, use a cautious assumption and write down that it is one.
Before you pick an interval
A measurement that sees the fault sooner gives you more warning.
A developing fault shows up in different measurements at different stages. In SKF's damage sequence for rolling bearings, high-frequency envelope measurements detect the damage first, then standard vibration monitoring, then what a person can hear and feel. Temperature comes late: NASA's guide treats high temperature together with high vibration as a sign that failure is imminent.
So the method and the interval are one decision: more warning can allow a longer interval. If a route takes only an overall velocity reading at a bearing that matters, adding an envelope reading at the same point may buy more time than visiting twice as often. Background noise in the signal matters too. SKF notes that the pre-warning time can vary with it, so a fan bearing beside a vibrating screen or a mill gearbox may warn later than the same bearing on a quiet machine.
Oil analysis follows the same logic, with one addition: laboratory turnaround comes out of the warning time. If the report takes two weeks to come back, two weeks of warning have gone before anyone reads it. For larger, critical or high-cost machines, NASA's guide describes routine oil sampling as typically quarterly or every six months once a machine has a year or more of history, and more often when a problem is emerging.
Between the first sign and the stop
Only the warning left after the time to act is useful.
Warning helps only if there is time to act on it. On a kiln line or in a steel mill, a critical fan cannot always stop the day a fault is confirmed. Someone has to agree the stop with production, confirm the spare, arrange the crane and issue the permits.
NAVAIR 00-25-403 turns this into a formula for equipment that cannot be stopped as soon as the warning appears and may keep running, without restriction or further checks, until the repair. Take the warning time, subtract the time needed to act, and divide what remains by the number of readings you want inside it. The result is the planned interval for a healthy machine.
Three readings is a sensible minimum: enough to see a change start and then confirm it. NASA's guide asks for at least three before failure is expected. The popular shortcut of measuring at half the P-F interval is practitioner convention, not a rule in these sources, and it assumes each inspection almost always catches the fault.
The arithmetic works in both directions. A slow wear mode with a long, stable history, such as a lightly loaded gearbox whose trends have been flat for years, may need fewer visits than the calendar gives it. NASA's guide says calendar intervals set for lack of information often lead to excessive monitoring, and that in those cases they can be increased significantly without harming reliability. The route hours saved can go to the failure modes with short warning times.
Illustrative field method
Does a monthly route give the ID fan enough warning?
This worked example uses a different fan, one with failure records to learn from. In a fictional cement plant, the kiln's induced-draught (ID) fan is on a monthly vibration route, with an acceleration envelope reading at a marked point on the fan-side bearing. The question is whether that interval gives the plant a fair chance to plan a bearing change before the bearing fails.
Records for two earlier outer-race failures on similar fans show the first envelope sign 12 and 10 weeks before the bearing had to come out. The plan uses the shorter figure.
Agreeing a kiln stop and having the spare bearing, a crane and permits in place usually takes about three weeks. That leaves a useful window of seven weeks.
About every 2.3 weeks: seven useful weeks divided by three readings. A fortnightly check fits inside that and is easy to schedule.
In the worst case, the fault becomes detectable just after a routine reading. The next readings come at weeks 4 and 8. Week 8 is a week after the stop had to be booked, so the decision would rest on the week 4 result alone.
Worst case for the fictional fan: the fault becomes detectable (P) just after a routine reading, and the bearing must come out at F, ten weeks later. The monthly route gets one reading in the useful window; a fortnightly check gets three. Illustrative arithmetic, not measured data.
When the trend moves
After an alert, measure more often and book the stop.
The planned interval assumes a healthy machine. Suppose the fan's fortnightly reading in week 4 crosses its alert level. Before changing anything, check that the reading is comparable with the last one: same marked point, same mounting, similar fan speed and load. The five questions in our earlier field note cover this check.
If the rise is real, measure more often. NASA's guide recommends cutting the interval to between one-third and one-quarter of the previous one once a reading passes its alert level, and shortening it further while the level keeps rising. For this fan, a quarter to a third of its 14-day interval is 3.5 to 4.7 days: in practice, a reading twice a week.
Watch the slope as well as the level: NASA notes that resistance to failure often falls sharply near the end of life. If readings stabilise, the interval can be relaxed again.
Review the interval after a change in load, speed or duty as well, for example when the kiln feed rate or the fan's damper setting changes. ISO 17359 notes that operating conditions such as duty cycles also influence the monitoring interval, and that a change in a reading may come from a change in duty rather than a fault.
Beyond the route
When no route is fast enough, change the method.
Sometimes the warning time minus the time to act is shorter than any route you could realistically run. Then a route is the wrong tool. You need an earlier indicator, a permanently installed sensor or a protection system.
ISO 10816-7, the vibration standard for rotodynamic pumps for industrial applications, says periodic measurements alone will not detect suddenly occurring defects, and recommends continuous on-line monitoring where a pump is relevant to safety. The clause is written for pumps, but the reasoning is worth applying to other machines. ISO 13373-1 lists what should decide between permanent and portable monitoring, including criticality, the cost of downtime, how fast the failure mode progresses, access and safety.
Check what "continuous" means before relying on it. ISO 13373-1 notes that scanning systems read their measuring points in turn, which leaves gaps at each point. Ask how often your point is actually updated.
Use on your route
Six entries make an interval you can test.
- Failure modeName the component and the mechanism, for example outer-race damage at the fan-side bearing.
- Method and pointRecord the measurement that detects it earliest and the marked point where it is taken.
- Warning timeWrite the estimate and where it came from: failure history, manufacturer data or a cautious assumption. Use the shortest figure you have seen.
- Time to actInclude planning, spares, permits, access, laboratory turnaround and the production window, not only the repair itself.
- Readings wantedThree is a sound minimum for seeing a trend start and then confirming it.
- Interval and review triggerState the interval and what will make you revisit it: an alert, a failure, a false alarm or a change in duty.
After every surprise failure or false alarm, compare the warning you actually had with the warning you assumed. Reliability-centred maintenance calls this age exploration: you adjust intervals as you learn how fast the machine really deteriorates. After a year, the route list should show which intervals have been tested against a real alert or failure and which are still estimates.
Technical sources
Primary references behind the interval method.
These references inform the method in this field note. Standards links point to public catalogue pages. They do not imply ConditionCore certification or conformity with a particular standard.
- NASA RCM Guide (2008)Reliability-Centered Maintenance Guide for Facilities and Collateral Equipment: monitoring practice, three readings before expected failure, response to alerts, oil sampling periodicity and resistance to failure near the end of life.
- NAVAIR 00-25-403 (2016)The Naval Aviation Reliability-Centered Maintenance Process, a US naval aviation management manual approved for public release. No official public link was found. Interval from the P-F interval, time to correct and number of inspections; shortest P-F interval for safety or environmental consequences.
- ISO 17359:2018General guidelines for condition-monitoring programmes: lead time to failure, the monitoring interval and operating conditions.
- ISO 13373-1:2002Factors for choosing permanent or portable vibration monitoring, and gaps in scanning systems.
- ISO 10816-7:2009Periodic and continuous vibration monitoring of rotodynamic pumps for industrial applications.
- SKF bearing failure guide (2025)Bearing damage and failure analysis: time from first damage to failure, the order in which methods detect bearing damage, and the effect of background noise.
- Aladon (2020)On-condition task interval determination: practitioner discussion of the half-interval convention and its assumptions.
Back at the route review
Change an interval only when the warning time supports it.
The review does not need to move the whole route to fortnightly. Only the failure modes whose warning, minus the time to act, leaves too few readings need a shorter interval. Some points on the same route may need fewer visits.
The fan that failed needs a different answer. On the measurement its route used, it gave less than three weeks of warning. If that plant also needs about three weeks to arrange a stop, no route interval is short enough. Add an earlier measurement, such as an envelope reading at the bearing, or move the bearing to an installed sensor.
A useful first step fits into one planning meeting. Pick the five machines on your route whose failure would stop the kiln, mill or caster. For each one, fill in the six entries for its most likely failure mode, then mark which intervals the numbers support and which they do not.
When a reading on a shortened interval crosses its alert level, take it into an integrated machine review before the work order changes. For a second view of your intervals, bring the route list, your last few failures and the readings taken before each one to ConditionCore.
Review a machine with ConditionCore