Why are there so many inspections with on-condition maintenance?

The complaint, and why it's incomplete

A common criticism of on-condition maintenance is that it generates a huge number of inspections relative to the failures it actually catches — particularly after an RCM project has just set the periodicities. It's a fair observation on the surface. But it comes from treating on-condition maintenance as a single technique, when in practice it splits into two quite different models that most textbooks lump together.

Two models, not one

The first model, long-term deterioration, covers components whose condition starts degrading from manufacture or from the moment they enter service — a rubber hose curing, a hull exposed to seawater. The second, incipient-failure detection, covers components that deteriorate rapidly only after a distinct initiating event, which can happen at any point in their service life — a bearing overloaded by a sudden shock, for example. The two differ mainly in how long the degradation runs and when it starts, and that difference is what drives how many inspections each one needs.

What has to be true for on-condition to work at all

Before comparing the two models, it's worth being precise about on-condition maintenance's prerequisites, since gaps here are often the real source of "wasted" inspections:

• There must be a genuine time lapse between the start of degradation (or its initiating event) and functional failure.
• The deterioration must be detectable early enough to avoid the failure, let recovery be planned with resources in place, and ideally leave room to time the shutdown to minimise operational disruption.
• Detection must be reliable — low on false positives and false negatives — and must isolate both the failure mode and the specific component needing corrective work.
• Time-to-failure for a given failure mode on a given component must be reasonably consistent across instances.
• Remaining useful life (RUL) must actually be predictable once degradation is detected.

It's worth flagging a semantic point too: on-condition maintenance sits in the preventative maintenance family, but it doesn't prevent failure. It only lets you avoid the worst consequences of a failure you can now see coming — corrective action at the later stages of deterioration is still required.

Long-term deterioration and the P-F interval

Some conditions, like fatigue cracking in a vehicle chassis or aircraft component, can only be reliably measured once the crack has grown large enough for non-destructive techniques — ultrasonics, eddy-current, dye-penetrant — to pick it up. That gap between "we can now detect it" and "it fails" is the P-F interval. It's a fixed window, unlike RUL, which counts down after diagnosis.

RCM sets inspection frequency at a fraction of the P-F interval, which is exactly why early inspections in the cycle often find nothing — the crack simply isn't there yet to detect. That's not waste in itself; it's the cost of guaranteeing the crack is caught before it reaches critical size. Where the criticism does have teeth is when an inspection regime ignores the second prerequisite above and keeps sampling components that can't actually be usefully diagnosed at that stage.

Cutting inspections without losing the protection

Two approaches bring the inspection count down responsibly. The first, damage tolerance — adopted in aerospace after the 1977 Dan-Air cargo aircraft crash — sets inspection frequency so that even the fastest credible crack growth can't reach a critical length between checks, rather than defaulting to a fraction of the P-F interval. It requires solid data on crack growth rates under real operating stresses, manufacturing and environmental defect rates, minimum detectable defect size, and maximum safe defect size — all of which need validating in any safety-regulated setting.

The second applies where condition can be measured at any point in the deterioration lifecycle, but full inspection is impractical — a gas turbine hot-end inspection that requires shutdown and cooling, or a pipeline network too extensive to monitor continuously (pigging scrapers and drones are the practical answer there). In these cases, engineering knowledge can target inspections at the sections most likely to fail first — the most heavily stressed points in a chassis, welded heat-affected zones, pipeline bends and low points — rather than treating the whole asset uniformly. Where a Weibull analysis of time-to-failure is available, a half-life scheduling approach — inspecting at roughly half the remaining deterioration life, then tightening as a credible shock event becomes more likely — further minimises the number of inspections while keeping the same protection. This isn't a scheduling method RCM explicitly describes, but it addresses the "too many inspections" criticism directly. You can read more about how our Reliability module automates Weibull analysis for exactly this kind of scheduling decision.

Incipient failure detection: the real cost-benefit paradox

Bearings illustrate the second model well. A properly specified, well-manufactured, correctly lubricated bearing operated within its limits rarely fails — but a lapse anywhere in that chain can trigger an initiating event. Detection methods (vibration, oil-debris sampling, temperature) each carry a different effective P-F interval, and fixed sensors with automated monitoring can eliminate periodic sampling altogether by watching continuously.

Here's where the paradox is sharpest: if a fleet of fifty machines with two bearings each experiences only one or two bearing failures a year, but the P-F interval is two months, RCM's own rule (sample at less than half the P-F interval) requires roughly twelve hundred samples a year to reliably catch those two or three failures. That ratio is precisely what looks wasteful at face value — but stretching the sampling interval without reference to the actual P-F interval means missing a real fraction of failures, or catching them too late to act on. Done properly, the "waste" is the price of catching nearly every failure early enough to matter.

It's also worth not underrating the oldest on-condition sensor available: an experienced maintainer's own senses. A trained person doing daily walk-downs can catch failures — by touch, sound, or smell — that never get logged in a CMMS, but that doesn't make them any less a legitimate and valuable part of the maintenance regime.

Making the cost case explicit

Because there's no equivalent of damage-tolerance scheduling for incipient failure detection, the honest answer to "is this inspection regime worth it" is a cost-benefit analysis, ideally built into the RCM process itself. That means capturing the P-F interval and its variation, the failure rate for the specific failure mode, the cost of inspection and analysis, the cost of an unplanned functional failure, and the cost of planned corrective action taken early. Where Weibull data and component ages exist, a simple discrete-event Monte Carlo simulation can project failure rates forward; where they don't, experienced maintainer and operator knowledge (or manufacturer and industry benchmarks) is a reasonable starting point that gets refined as real data accumulates.

It's also worth accounting for diagnostic quality itself — missed failures on one side, false positives triggering nugatory corrective work on the other. If the cost case still holds once both are priced in, then a high ratio of inspections to detected failures isn't a flaw in the regime; it's a predictable, acceptable consequence of doing on-condition maintenance properly.

The bottom line

Splitting on-condition maintenance into these two models explains why inspection counts look disproportionate to failures found, and shows there are legitimate, principled ways to bring that count down — damage tolerance and half-life scheduling for long-term deterioration, tighter cost justification for incipient failure detection — without weakening the protection the regime exists to provide. Done right, RCM's own logic justifies the ratio rather than exposing it as waste.