Ask any reliability engineer what the P-F curve is and they'll give you a clean answer: the curve shows how an asset's condition degrades from the point of potential failure detection (P) down to functional failure (F), and the P-F interval is the window you have to intervene. Clean concept. Widely understood.
Ask those same engineers what P-F interval they've assigned to each asset in their fleet, and the answers get fuzzy fast. Most are using generic manufacturer estimates, or nothing formalized at all. The concept is understood theoretically but not applied operationally.
That gap — between knowing the curve and actually using it to set maintenance thresholds — is what this article is about.
The P-F Concept, Briefly
In RCM (Reliability-Centered Maintenance) theory, every asset follows a degradation path from new through functional failure. The P-F curve plots condition (or a condition indicator) against time. "P" is the point at which a potential failure becomes detectable by some monitoring technique. "F" is functional failure — the point at which the asset can no longer perform its required function.
The P-F interval matters because it defines the minimum detection frequency required for your monitoring program to be useful. If the P-F interval for a specific bearing fault mode is 4 weeks, and you're only doing monthly vibration rounds, you have almost no margin. A reading taken on day 1 of the month might catch it at P; a reading on day 29 might be looking at an asset that's already at F.
The standard guidance — found in John Moubray's RCM2 and subsequent literature — is that your inspection interval must be less than half the P-F interval if you want a reasonable probability of catching the failure in time to act. This is called the P-F interval halving rule, and most condition monitoring programs that underperform are violating it without realizing it.
P-F Intervals Vary Enormously by Fault Mode
One of the most practically important things about P-F intervals is that they're not a property of an asset — they're a property of a specific fault mode on a specific asset in specific operating conditions. This distinction matters enormously when you're trying to set detection frequencies.
Consider a horizontal centrifugal pump:
- Outer race bearing defect — P-F interval typically 4–8 weeks at normal load. The defect frequency appears in the spectrum weeks before the bearing reaches the noise-generating stage. High-confidence detectable with continuous vibration monitoring.
- Impeller imbalance — P-F interval can be much shorter, days to weeks depending on severity and operating speed. Detectable via 1X vibration amplitude increase in the radial plane. Can accelerate rapidly if the imbalance is from material buildup or erosion rather than assembly.
- Cavitation — P-F interval is essentially zero in practical terms. When a pump is cavitating badly, it's already causing impeller erosion. Detectable via broadband high-frequency noise (sub-harmonic and random bursts), but the degradation is concurrent with the detectable signal. Early detection here means catching mild cavitation before it becomes severe — a different kind of intervention than bearing replacement.
- Seal face degradation — P-F interval is highly variable. Detectable via increased process fluid leakage rate (flow/pressure monitoring), temperature at the seal gland, or in some configurations vibration signature changes. Could be weeks to months depending on process conditions.
If you're managing monitoring frequency at the asset level ("check this pump monthly") without thinking about which fault modes you're targeting, you'll catch some fault modes comfortably within their P-F interval and miss others entirely.
How P-F Intervals Inform Monitoring Frequency — And Thresholds
The practical translation of P-F interval into a monitoring program has two components: detection frequency and alert threshold calibration.
Detection frequency: Per the halving rule, if your bearing outer race P-F interval is 4 weeks minimum, your effective detection frequency should be 2 weeks maximum to maintain reasonable probability of catching it. For continuous monitoring systems (data collected every few hours), this is trivially satisfied. For route-based programs, this means fortnightly rounds at a minimum for this asset class — which most monthly PM programs don't do.
Threshold calibration: The P-F interval also tells you where to set your alert threshold. If you want to catch a bearing fault at P (earliest possible detection) with 4 weeks of P-F interval remaining, your alert threshold needs to fire at Stage 2 bearing degradation, not Stage 3. Setting an alert at "overall vibration exceeds 0.3 in/s" might be appropriate for a Stage 3 detection program with 2-week intervention turnaround. But if you want a 4-week window, you need to detect Stage 2, which means watching specific bearing defect frequencies at amplitudes as low as 0.05-0.1 in/s in the relevant bands.
Most generic "ISO 10816 zone D" thresholds are functional failure indicators, not potential failure indicators. They're catching assets at Stage 3 or 4. Using them as your primary alert is better than nothing, but you're consistently working with compressed P-F windows.
The Curve Shape Matters as Much as the Interval Length
P-F curves are typically drawn as smooth, gradual degradation lines. In practice, many fault modes have nonlinear degradation curves — they degrade slowly for a long time, then rapidly near failure. This "hockey stick" shape has a direct implication: if your detection system catches the fault in the slow-degradation phase, you have a long time to act. If you miss it until the rapid phase, the apparent P-F interval looks short even though the underlying interval was long.
Bearing defect frequency growth fits this pattern well. BPFO amplitude growth from Stage 2 to Stage 3 can take 3-5 weeks. Growth from Stage 3 to Stage 4 often takes less than a week. A maintenance team that only catches Stage 3 defects will consistently perceive short P-F intervals and conclude that condition monitoring barely gives them enough time to act. The actual problem is that they're detecting too late on the curve.
We're not saying that late detection always means inadequate monitoring setup — sometimes the fault mode genuinely has a short P-F interval, or the asset operates in conditions that accelerate degradation. But if you're finding that your condition monitoring program routinely gives you only days to act, the first question to ask is whether your detection threshold is set at the right point on the curve.
Assigning P-F Intervals: Practical Approaches
There are three practical approaches to assigning P-F intervals for a rotating equipment fleet, each with different data requirements.
Manufacturer data + engineering judgment: Many bearing manufacturers publish L10 life calculations and provide guidance on detectable degradation stages. Equipment OEMs sometimes include P-F interval estimates in their maintenance documentation. These are generic estimates, not asset-specific, but they're a reasonable starting point for new installations without historical failure data.
Historical failure data analysis: If your CMMS has records of failure modes and detection dates, you can reconstruct actual P-F intervals from past failures. The data is rarely clean, but even rough estimates from 5-10 failure events for a common asset class give you better calibration than generic manufacturer data. The challenge is that most CMMS failure records capture "work order closed" dates, not condition indicator trend data from the weeks prior. You need some condition history to calculate the interval.
Continuous monitoring trend analysis: With continuous vibration monitoring and a health score trend, you can observe the actual degradation curve as it happens for each failure event. After 2-3 confirmed bearing replacements with full trend data, you have site-specific P-F interval estimates for that asset class under your specific operating conditions. This is the highest-quality input because it accounts for local load cycles, process conditions, and maintenance practices.
At Fleetpio, we track the health score trend from first significant anomaly detection to the point of confirmed maintenance intervention, and accumulate that data to refine per-asset-class P-F interval estimates over time. After a fleet has been monitored through 6-10 maintenance events, the threshold calibration becomes much more precise than any manufacturer generic estimate.
The Practical Takeaway for Setting Your Program
If you're building or refining a condition monitoring program and want to use P-F intervals operationally, here's a starting framework:
First, enumerate the fault modes you're targeting for each asset class — not just "bearing failure" but outer race, inner race, rolling element, cage, plus whatever other fault modes are relevant (imbalance, cavitation, misalignment, looseness). Each fault mode gets its own P-F interval estimate.
Second, for each fault mode, identify the earliest-detectable indicator — which parameter, which frequency band, which analysis technique gives you detection earliest on the curve. This determines what your monitoring system needs to measure and at what granularity.
Third, set your detection frequency to satisfy the halving rule for the shortest P-F interval in your priority fault mode list. If you have one fault mode with a 2-week P-F interval among assets that otherwise have 6-week intervals, that short one sets your minimum detection frequency.
Fourth, set your alert thresholds to fire at the earliest practical detection point — Stage 2 bearing defects, mild cavitation signatures — not at the functional failure boundary. You want to catch assets at P, not at F minus two days.
The P-F curve doesn't automatically improve your maintenance program. Understanding it clearly enough to translate it into specific detection frequencies and alert thresholds is where the operational value actually lives.