The P-F interval is the time between the point at which a developing failure first becomes detectable (P) and the point at which it becomes a functional failure (F). That window is all the warning you get. It's the time you have to detect the problem, plan the work, get the parts and act before the equipment stops doing its job.
Get it wrong and you'll pay for a monitoring program that finds failures you can't do anything about.
Below I cover where the P-F interval came from, what point P means, why there's no single P-F interval for a piece of equipment, and how to set an inspection interval you can act on. I'll also cover who named the curve, and where the half rule comes from.
What do P and F stand for?
P is potential failure. F is functional failure.
Nowlan and Heap defined the first one on page 19 of their 1978 report, and the definition has held up for nearly 50 years:
"A potential failure is an identifiable physical condition which indicates a functional failure is imminent."
A functional failure is when your equipment can no longer do what you need it to do, to the standard you need. That standard is your operating requirement, which is often well below the design spec. A pump that still runs but can't deliver the flow your process needs has functionally failed.
We cover that distinction properly in failures vs functional failures vs failure modes.
The interval between P and F is usually expressed in days, weeks or months. It can also be expressed in cycles, starts and stops, or units produced. Use whatever your degradation tracks.
Where did the P-F curve come from?
Nowlan and Heap introduced the concept in their 1978 report on reliability-centred maintenance. They never called it the P-F curve.
In the original document the diagram is Exhibit 3-1. The vertical axis is resistance to failure, and the horizontal axis is operating age. The points are labelled A, B, C and D:
- A is the new item
- B is where the crack first becomes visible
- C is the potential failure
- D is the functional failure
Between C and D the report marks inspection intervals, ΔT, and that's the basis of the on-condition task: inspect often enough to catch the potential failure before the functional failure arrives.
The labels P and F appear nowhere on the 1978 curve, and a full-text search finds no "P-F" anywhere in the report. It's public, filed as AD-A066579, so you can check for yourself.
The name came later, with John Moubray, who developed the modern version of the curve in RCM II and called it the P-F curve.
I prefer the version Moubray uses in RCM II, with condition on the vertical axis, because condition is what you're looking at in the field. Where I part ways with him is point F. Moubray stops his curve there. I carry mine on, because a functional failure often leaves equipment running in a degraded state.
Is point P where the failure starts?
No. Point P is where the onset of the functional failure can first be detected by the technique you're using.
The failure mechanism is already running when you get to P, often for a long time. What changes at P is your ability to see it.
That sounds academic until you change technique. Take a single failure mode on a single bearing. A vibration analyst with envelope detection picks it up at one point. A fitter walking past with a torch picks it up much later.
Point P depends on your detection method as much as on your equipment.
You'll often see point P defined as the point where degradation begins.
If you believe P is the birth of the failure, you'll read the curve as a picture of how long the failure takes to develop. It only shows how much of that development your technique can see.
Treat P as the birth of the failure and you'll under-invest in detection, then get caught out when the warning turns out shorter than the diagram suggested.
There's no P-F interval for "a pump"
Ask what the P-F interval is for a centrifugal pump and somebody will give you a number. It won't mean anything, because the question is wrong.
A P-F interval belongs to a failure mode. And "pump fails" isn't a failure mode.
I teach people to avoid generic verbs like fails, breaks or malfunctions, because they give you nothing to work with. Write "coupling bolts come loose", not "coupling fails". Only at that level of detail can you ask what the warning period looks like.
The interval also differs between failure modes on the same technique. As a rough illustration, thermography might pick up a bearing problem a few months out, and an electrical fault much earlier.
| The P-F interval depends on | Example |
|---|---|
| The failure mode | Inner-race fatigue behaves nothing like lubrication starvation |
| The detection technique | Vibration, ultrasound, oil analysis and a visual check each find it at a different point |
| The operating context | Duty cycle, load, ambient temperature, contamination |
| The individual occurrence | The same mode on the same machine varies from one event to the next |
The shape of the curve varies between failure modes, and it varies for the same failure mode.
If the curve is steeper next time, you can get a functional failure between 2 condition checks. If it's flatter, you're inspecting more often than you need.
The RCM guide, SAE JA1012, says the same thing in standards language. P-F intervals aren't always constant, so set the task interval below the shortest likely interval. Where the variability is extreme, drop the task altogether.
2 public bearing run-to-failure datasets, PRONOSTIA and the IMS set, show nominally identical bearings under identical load and speed failing at very different times.
What about the published tables of warning times?
You've seen them: detection techniques down one side and months of warning across the other, with "hot to touch" at the bottom and oil analysis or ultrasound at the top.
Most of those tables cite nothing, and they contradict each other. One puts vibration at about 1 month of warning on rolling-element bearings, another at 12 to 18 months. These numbers were never measured on a plant.
I've used one of these charts in my own training, as a concept and nothing more. That's all they're good for.
ISO 17359 and ISO 13379 don't publish technique-specific durations either. They describe process and methodology. For a properly sourced starting point on choosing parameters by machine type, use our tables derived from ISO 17359.
There's no published study of P-F interval distributions in an operating process plant, and I don't expect one. Collecting that data would mean letting failures run to F on purpose. Get your real numbers from your own failure history.
How often should you inspect?
There's one hard requirement and one rule of thumb, and they get mixed up a lot.
The hard requirement: your inspection interval must be shorter than the P-F interval. SAE JA1011 clause 5.7.2 puts it as less than the shortest likely P-F interval.
The rule of thumb is that the inspection interval should be about half the P-F interval, and sometimes a third. Why half? To answer that you need the net P-F interval.
The net P-F interval
The net P-F interval is the P-F interval minus your inspection interval. It's the minimum warning you're guaranteed, as opposed to the maximum the curve shows.
Say you have a P-F interval of 8 months and you inspect every 3 months. Your net P-F interval is 5 months. Those 5 months are the minimum time between finding the potential failure and the functional failure, so they're what you have to act in.
Moubray worked the same calculation with different numbers:
| P-F interval | Inspection interval | Net P-F interval | Source |
|---|---|---|---|
| 8 months | 3 months | 5 months | Our own worked example |
| 9 months | Monthly | 8 months | RCM II (Moubray), repeated in SAE JA1012 |
| 9 months | 6-monthly | 3 months | RCM II (Moubray), repeated in SAE JA1012 |
Inspect less often than the P-F interval and the net interval goes negative. You then find some failures at F and some well past it.
Set the inspection interval at half the P-F interval and the net P-F interval also lands at half.
Once you see the half rule as arithmetic, you can choose a different fraction on purpose.
Who said "half"?
- Nowlan and Heap didn't say it. No half or fraction rule attaches to on-condition inspection intervals in the 1978 report. They set ΔT from the measured rate of decline for the specific failure mode.
- SAE JA1011 doesn't say it either. It requires a task interval shorter than the shortest likely P-F interval. The word half doesn't appear.
- John Moubray put it in RCM II: half the P-F interval is usually enough. SAE JA1012 later wrote the same guidance into the standard, with an escape clause for other fractions.
Half is a good start. Sometimes a third is the better call, and that takes judgement.
The half rule assumes you find it the first time you look
Every version of the half rule assumes that once your equipment passes point P, the very next inspection detects it. That's a probability of detection of 1.
Inspection doesn't work that way in the field.
Gary West, writing on Aladon's site, sets out how the US Navy's RCM process (NAVAIR 00-25-403) handles this. It uses the probability of detecting the potential failure on any one inspection, and an acceptable probability of the failure getting through.
In his worked example, an acceptable failure probability of 0.001 with a detection probability of 0.90 calls for 3 inspections inside a single P-F interval. That's 1 inspection every third of the P-F interval, with a technique you'd already call reliable.
Probability of detection can be measured. MIL-HDBK-1823A sets a standard way to measure how reliably non-destructive testing finds a flaw of a given size.
Operator skill widens the spread. The UK Health and Safety Executive ran a program on manual ultrasonic inspection and found that results varied widely between the best and the weakest operators, even with better procedures.
That's why effective condition-based maintenance needs much more than technology. You need the right people and practices, so the measurement taken in the field is accurate and repeatable.
So when you see "a half or a third", treat the third as the answer to imperfect detection.
Does your net P-F interval survive your planning process?
This is the test that decides whether condition monitoring is worth doing on a given failure mode:
Net P-F interval ≥ the time to plan the work, get the parts and do it.
Fail that test and you detect failures you can't act on. You get the pain of a breakdown plus the cost of the monitoring program that watched it coming.
I see this a lot. Organisations run condition monitoring programs and then don't intervene in time, so they still end up with failures they saw developing. Too often those failures are well past point F before anyone puts a spanner on them.
Usually the site is too reactive to act on what the monitoring finds. Your weekly schedule is already full of emergency work, so an alert with 5 months of warning can sit in the backlog for 6.
So my advice is simple. Don't start a condition monitoring program until you've got effective planning and scheduling in place.
We've written that argument up in full in why CBM fails without planning and scheduling.
A worked example
These numbers illustrate the method. They aren't measured data.
Asset: process fan, 315 kW, single-stage centrifugal, spared on a shared header.
Failure mode: drive-end rolling-element bearing spalled, from inner-race fatigue.
Technique: monthly route-based vibration with envelope bearing bands.
Across the last 4 occurrences on this fan class, the time from the first confirmed bearing-band alarm to the point where the fan could no longer hold duct pressure was 11, 14, 8 and 26 weeks.
Design to the shortest, 8 weeks. The average, about 15 weeks, would leave you exposed.
| Inspection interval | Net P-F interval | What it gives you |
|---|---|---|
| Monthly | 4 weeks | Half the P-F interval |
| 6-weekly | 2 weeks | Very little. Miss one route and you have nothing |
| Quarterly | Negative | You find some failures at F and some past it |
Now apply the planning test. Does a 4-week net interval get the job done? The repair is only 10 hours on the tools, so everything depends on the parts.
- Bearing set on the rack: 1 week to plan it into the schedule, and the next window is in 2 weeks. That's 3 weeks against a 4-week net interval. It works, with a week to spare, and only if its priority earns it a place in next week's load sequence.
- Bearing set not in stores: 12 working days of procurement, plus a week of planning, plus the window. That's 5 to 6 weeks against a 4-week net interval. It fails, and your monthly vibration route on this fan is giving you warning you can't use.
There are 3 ways to close a gap like that, in the order I'd try them:
- Stock the bearing set. It's the cheapest fix. It cuts the time to act from 5 to 6 weeks to 3, so the route you already pay for gives you warning you can use.
- Go fortnightly. The net interval rises to 6 weeks. It also doubles the route effort on this asset and does nothing about procurement, so save it for cases where stocking isn't possible.
- Allow for imperfect detection. With monthly checks you get 2 looks at that bearing inside the 8-week P-F interval. The NAVAIR arithmetic wants 3. That pushes you toward fortnightly checks, or a permanently mounted sensor on that one bearing.
If the shortest observed interval had been 3 weeks, no inspection frequency would save you. That's a design conversation.
What if there's no usable P-F interval?
Sometimes there isn't one, and forcing a condition-based task into that gap is how PM programs fill up with work that manages nothing.
SAE JA1011 lists 5 criteria for technical feasibility. Moubray put them as 4 conditions, and you need all 4:
- A clear, identifiable potential failure condition exists.
- The P-F interval is reasonably consistent.
- It's practical to monitor at intervals shorter than the P-F interval.
- The net P-F interval is long enough to be of some use.
And you need people who can read the results, and planning and scheduling that acts on them in time. Without those, the other 4 won't save you.
Fail any of them and condition monitoring is the wrong policy for that failure mode. Your options are scheduled restoration, scheduled discard, redesign or run-to-failure, depending on the consequence of the failure.
There's one more option the standards don't emphasise. Where the P-F interval is very short or wildly inconsistent and the consequence is significant, consider an automated trip. Instrument the machine to shut itself down safely, so you're not relying on a person on a route to catch it inside a 3-day window.
2 other cases land here too. Hidden functions rarely have a useful on-condition point P, which is why they get failure-finding tasks. Shutdown valves with smart positioners are an exception: partial-stroke data can show you how the valve is trending. And infant mortality is best managed by quality control and testing before equipment goes into service.
Don't buy a more sensitive sensor and hope. It may move P earlier on some failure modes, but your supply chain and your schedule still have to act inside the warning.
Where the P-F curve applies, and where it doesn't
The P-F curve is a condition-based maintenance concept only. Don't use it to represent time-based preventive maintenance or run-to-failure.
The other strategies fit like this:
- Condition-based maintenance is the parent concept. The P-F interval decides whether a CBM task is worth setting up at all.
- Time-based maintenance is the right answer where a failure mode has a clear wear-out age. For those modes it's often the first choice.
- Predictive maintenance is a later, model-based extension of CBM. Get the basics working first.
- Corrective maintenance and run-to-failure remain legitimate answers where the consequences are purely economic and no proactive task pays.
The on-condition task comes out of reliability-centred maintenance, and the P-F interval is what makes that task work or fail.
Do the arithmetic before you buy the monitoring
The P-F interval is the warning period your chosen technique gives you for one specific failure mode.
Subtract your inspection interval to get the warning you're guaranteed. Then compare that with how long your organisation takes to plan, procure and do the work. If the numbers don't work, change the policy.



