When to Do Maintenance Work: The P-F Curve and the Cost of Work Balance
How to decide when maintenance work should happen: read the P-F curve, pick the right trigger for each failure mode, and balance reactive and proactive cost of work on your own assets.
The right time to do maintenance work is after a failure mode becomes detectable and before the asset loses its function. Reliability engineers call that window the P-F interval: the time between potential failure (P), the earliest point a problem can be detected, and functional failure (F), when the asset can no longer do its job [5]. Work done before P replaces parts that still had life in them. Work done after F is a breakdown repair. Good maintenance timing means detecting P early, confirming it, and finishing a planned repair before F.
That answer has three practical parts, and each one is a separate decision for a maintenance manager. The first is which trigger starts the work: a condition reading, run hours, a calendar date, or the failure itself. The second is how much proactive work is worth paying for on a given asset before the cost of inspections and replacements exceeds the cost of the failures they prevent. The third is an honest view of where your plant is today, because most plants still run a large share of their work reactively.
This guide covers those three decisions in depth. If you are still deciding whether to start a predictive program at all, our step-by-step roadmap from reactive to predictive maintenance covers the phases, data foundations and business case. This article assumes you want to get the timing of individual tasks right.
Where most plants really are on the maintenance mix
Most maintenance work in U.S. manufacturing is still triggered by failure. A 2020 NIST survey of manufacturers found that the average establishment's maintenance was 45.7% reactive, 31.8% preventive and 17.3% predictive [1]. Establishments in the top 25% for reliance on reactive maintenance were associated with 3.3 times more downtime and 16.0 times more defects [1]. The survey dates from 2020, so treat it as a baseline for comparison rather than a current census of your industry.
The federal O&M guidance describes what a mature mix looks like. The DOE Federal Energy Management Program guide, Release 3.0 from 2010 and still published by DOE, lists continually top-performing facilities at under 10% reactive, 25% to 35% preventive and 45% to 55% predictive [2]. PNNL's O&M best practice page lists the same ranges for leading programs, alongside broader targets of 20% to 30% reactive, 25% to 35% preventive and 30% to 40% predictive [3]. Neither source sets reactive work at zero. The DOE guide notes that maintenance on equipment that is inexpensive and unimportant to facility reliability may best be left reactive [2].
The three phases are easy to name. Reactive maintenance fixes equipment after it breaks. Planned maintenance performs scheduled activities on a time or run-hour basis. Predictive maintenance uses sensing data and analytics to decide when the equipment actually needs work [2]. Moving work from the first phase to the third is a timing problem as much as a technology problem, which is where the P-F curve comes in.
How the P-F curve tells you when to act
The P-F curve plots an asset's resistance to failure against time. Resistance starts high and declines as wear accumulates. Point P is where the decline first becomes detectable, and point F is where the asset stops meeting its required function [5]. Everything useful a maintenance team can do on a developing fault happens between those two points.
Where you put P depends on your detection method, and where you put F depends on the operating context. Fluke's worked example is a pump rated for 100 gallons per minute that might be treated as potentially failed at 95 and functionally failed at 90, with thresholds that vary by workload and environment, so two identical pumps in different parts of a plant can degrade differently [5]. Agreeing those thresholds with operations is part of setting up the task, not an afterthought.
Detection method decides how much warning you get. Non-intrusive methods that run while the equipment operates, such as vibration analysis or infrared imaging, tend to detect potential failures earlier and give a longer P-F interval than intrusive inspections that require shutdown or disassembly [5]. Near the end of the curve, detection comes from operator work requests and visual findings on rounds, which leaves little time to plan [5].
The four-step loop inside the P-F interval
Every planned repair on a developing fault follows the same sequence: detect the potential failure early, confirm it with a second measurement or an inspection, scope and plan the corrective action (parts, labor, permits, production window), then execute the repair. The interval has to be long enough for all four steps. If confirming and planning a bearing change on a conveyor head pulley takes longer than the warning your detection method gives, the work will still arrive as an emergency. Change the part at the right time: not too early, and never too late.
That loop is why the length of the P-F interval matters more than the elegance of the detection. A vibration route that catches outer race damage on a supply fan with enough lead time to order the bearing and book a weekend outage turns a failure into planned work. The same finding with too little lead time only tells you what broke. Predictive programs give lead time to purchase materials for repairs, which also reduces the parts inventory a plant has to hold [2]. For how that lead time connects to stocking decisions, see our guide on planning spare parts around condition signals and lead times.
The four triggers that start maintenance work
Once you know the failure mode and the warning it gives, you can choose what starts the work. There are four triggers, and a single asset usually needs more than one, because each of its failure modes behaves differently. The table maps each trigger to the situation it suits and the data it needs.
| Trigger | When to use it | Data you need | Example on a plant floor |
|---|---|---|---|
| Condition based (predictive with sensors or IIoT; prescriptive when analytics also recommend the action) | Critical equipment with random failure patterns and a detectable P-F interval | Vibration, temperature, oil, ultrasound or performance trends with an alarm limit and a confirmation step | Wireless vibration on a cooling tower gearbox raises a case when bearing tones trend up at steady load |
| Usage based (preventive by operating hours or production amount) | Wear that tracks run time, cycles or throughput rather than the calendar | Run-hour meters, cycle counts or production counts linked to the CMMS | Conveyor belt and scraper inspection every set number of operating hours, not every month |
| Time based (preventive at a set interval) | Consumables, known wear patterns and manufacturer recommendations | A calendar schedule in the CMMS and a task list | Quarterly filter changes on an air handler, annual inspection of a safety-related hoist |
| When it breaks (run to failure, found by inspection or process feedback) | Small, non-critical, redundant or inexpensive equipment where failure has little consequence | Operator work requests, inspection round findings, process alarms | A spare exhaust fan in a redundant pair is repaired when the operator reports it |
The DOE guide gives the same allocation in its RCM hierarchy. It assigns reactive tasks to small parts, non-critical equipment, equipment unlikely to fail and redundant systems, preventive tasks to equipment subject to wear, consumables, known failure patterns and manufacturer recommendations, and predictive tasks to critical equipment, equipment with random failure patterns and systems where incorrect preventive maintenance can induce failure [2]. That last category is worth noticing: for some equipment, the scheduled task itself is a cause of failure.
Usage-based triggers are often the cheapest improvement available. Many calendar PMs on conveyors, pumps and compressors exist because the CMMS had a calendar field and no meter reading. Linking run hours or production counts to the task moves the work closer to actual wear without adding a single sensor. It also exposes assets whose duty changed after the PM was written, such as a pump moved from standby to lead duty.
Three stages of predictive maintenance: detect, diagnose, predict
Condition-based work is not one capability. It matures in three stages, and each stage asks more of your data. The first stage is anomaly detection: spotting a trend that departs from normal behavior. The second is diagnostics: identifying the root cause of the anomaly, such as misalignment versus bearing wear versus looseness. The third is prediction: estimating remaining useful life, so the planner knows how long the P-F interval has left to run.
Deloitte's predictive maintenance position paper describes the data each stage depends on. Anomaly detection needs enough accumulated sensor data at a minimum measurement frequency to define the norm for a process, and an alert from it still needs a person to decide whether an actual failure has occurred [4]. Root cause analysis needs records of successful and unsuccessful repair approaches, and any predictive program needs process failures recorded with their exact date and time as its most basic input [4]. With enough process data, predictive maintenance can estimate the remaining runtime of assets with increasing accuracy [4].
The practical lesson is that each stage is earned by the one before it. A plant whose work orders say "pump repaired" with no failure code cannot train a diagnostic model, however good its sensors are. Confirmed findings closed against the right asset and component are what make the second and third stages possible. If your alerts already flow but technicians close them without findings, fix that first, as described in our guide to tuning condition alerts with technician feedback.
Finding the balance between reactive and proactive cost of work
Total cost of work is high at both extremes. With too much reactive work, the plant pays for breakdowns. With too much proactive work, it pays for inspections and replacements that the equipment did not need. Picture two curves on the same chart: reactive cost falls as you add proactive work, and proactive cost rises. The optimum for an asset is near the point where the two curves cross, and it is different for every asset and failure mode.
The reactive side has well-documented cost drivers. The DOE guide lists the cost of unplanned downtime, higher labor cost when overtime is needed, the cost of repair or replacement, possible secondary damage to equipment or the process, and inefficient use of staff [2]. Running equipment to failure also tends to require a larger inventory of repair parts [2]. The proactive side has its own: preventive programs are labor intensive, include unneeded maintenance, and carry the potential for incidental damage to components during that unneeded work [2].
To place a specific asset on the curve, price four inputs with your own records. None of them should come from an industry average.
| Cost input | What to measure | Where it usually lives | Pump, fan or conveyor example |
|---|---|---|---|
| Lost production | Hours of lost output per failure and the margin on that output | Downtime log, production reporting | A failed conveyor drive stops a packaging line until the gearbox is replaced |
| Labor | Straight time and overtime hours per repair, plus contractor callouts | CMMS work order actuals, payroll | An off-shift pump seal failure brings in two technicians on overtime |
| Replacement cost | Parts, expedited freight and secondary damage | CMMS parts usage, purchasing records | A seized fan bearing damages the shaft, so the repair needs a shaft as well as a bearing |
| Compliance risk | Safety, environmental or quality consequences of the failure | EHS incident records, quality holds, permit conditions | A failed scrubber circulation pump puts an emissions permit condition at risk |
Compare those four costs per failure with the cost of the proactive task that would prevent it, including the inspections that find nothing. A critical pump with a high lost-production cost justifies continuous condition monitoring. A redundant exhaust fan with low cost on every line belongs on the run-to-failure side of the curve, and spending more on it moves the plant away from the optimum, not toward it. Our risk-based framework for prioritizing maintenance backlogs uses the same consequence inputs to rank the work you already have.
Putting the timing decisions to work on your assets
Start with the assets where timing errors cost most, not with the whole register. A criticality ranking narrows the list: the P-F curve approach works best on critical assets with no backup, while lower-priority assets may suit simpler time-based methods or run-to-failure [5]. For a mid-sized plant, that might be the conveyor drives, process pumps and fans whose failure stops a line.
Then work through each critical asset failure mode by failure mode:
1. List the dominant failure modes from work order history and technician knowledge, such as bearing wear, seal leaks, belt misalignment and coupling wear. 2. Name the earliest detection method for each mode and estimate the warning it gives, using vendor guidance and your own failure records. 3. Check the warning against the four-step loop. If confirm, plan and execute take longer than the warning, choose an earlier detection method or hold the critical spare. 4. Assign the trigger from the table above, and replace calendar PMs with usage-based ones where wear follows run time. 5. Price the cost inputs and drop proactive tasks whose cost exceeds the failures they prevent. 6. Write the result into the CMMS as task triggers, alarm limits and job plans, so the decision survives staff turnover.
Then review the result against outcomes. Track how many failures on the selected assets were caught inside the P-F interval, how many arrived as breakdowns, and how many proactive tasks found nothing. A rising share of planned work with a falling share of no-finding inspections means the timing is converging on the balance point.
Where Monitory fits
Most of this method is a discipline your team can run with a spreadsheet and the CMMS. The hard parts are getting condition data early enough on critical rotating equipment and getting each confirmed finding into a planned work order before the P-F interval runs out. That is the work Monitory does: our condition monitoring services combine vibration, thermography, oil analysis, ultrasonic and other techniques so each failure mode is watched by the method that catches it earliest, with the alert routed into your work order process.
If you want to see what that looks like on real equipment, our field results from condition monitoring at operating sites show the alert, the finding and the repair for failures ranging from bearing failures to belt misalignment. If you want a starting point for your own plant, the free on-site predictive maintenance assessment builds a business case and a plan from your own maintenance and downtime records.
Frequently asked questions
What is the P-F interval in maintenance?
The P-F interval is the time between potential failure, the earliest point a developing fault can be detected, and functional failure, when the asset can no longer perform its intended function [5]. It is the window in which a maintenance team can confirm the fault, plan the repair and complete it before a breakdown.
How often should condition monitoring inspections run?
Often enough that a fault cannot start and reach functional failure between two inspections, with time left to confirm, plan and execute the repair. In practice, the inspection interval should be a fraction of the P-F interval for the failure mode you are watching. Failure modes with short warning periods need continuous monitoring rather than a monthly route.
Is preventive maintenance ever a waste?
It can be. Federal O&M guidance lists unneeded maintenance and incidental damage to components during that work as disadvantages of preventive programs [2]. Time-based tasks are still the right choice for consumables, known wear patterns and manufacturer requirements, so the fix is to move each task to the trigger that fits its failure mode, not to drop preventive work.
Should any equipment be run to failure?
Yes, when failure has little consequence. The DOE guide assigns reactive maintenance to small parts, non-critical equipment, equipment unlikely to fail and redundant systems [2]. Run to failure should be a written decision for those assets, with spares and response plans in place, not the default for everything.
What is the difference between predictive and prescriptive maintenance?
Predictive maintenance estimates when an asset will need work, using condition data and analytics. Prescriptive maintenance goes one step further and recommends the corrective action, such as which component to replace and when to schedule it. Both depend on clean failure history, because a recommendation is only as good as the confirmed findings it learned from.
Summary and next steps
Maintenance timing comes down to the P-F interval. Detect the potential failure as early as your methods allow, confirm it, plan the repair and finish it before functional failure. Choose a condition, usage, time or run-to-failure trigger for each failure mode, and set the amount of proactive work by pricing lost production, labor, replacement cost and compliance risk on each asset.
Three next steps for this month:
- Pick your five most critical assets and list their dominant failure modes with the earliest detection method for each.
- Find the calendar PMs on those assets that should be usage based, and link them to run hours or production counts.
- Price the four cost inputs for the last three failures on each asset, and compare them with the cost of the proactive tasks you run today.
If you want a second set of eyes on that list, our free assessment is built around exactly this exercise.
References
[1] NIST, "Research Suggests Significant Benefits to Investing in Advanced Machinery Maintenance", 2020. https://www.nist.gov/news-events/news/2020/06/research-suggests-significant-benefits-investing-advanced-machinery
[2] U.S. Department of Energy, Federal Energy Management Program, "Operations and Maintenance Best Practices Guide, Release 3.0", 2010. https://www.energy.gov/sites/prod/files/2020/04/f74/omguide_complete_w-eo-disclaimer.pdf
[3] Pacific Northwest National Laboratory, "O&M Best Practice Issue Discussion: Maintenance Approaches", PNNL. https://www.pnnl.gov/projects/om-best-practices/maintenance-approaches
[4] Deloitte Analytics Institute, "Predictive Maintenance: Taking pro-active measures based on advanced data analytics to predict and avoid machine failure", position paper. https://www.deloitte.com/content/dam/assets-zone2/de/de/docs/about/2024/Deloitte_Predictive-Maintenance_PositionPaper.pdf
[5] Fluke, "How To Use The P-F Curve To Stay Ahead of Equipment Failure", 2025. https://www.fluke.com/en-us/learn/blog/predictive-maintenance/p-f-curve-stay-ahead-of-equipment-failure
Related articles
Ready to put this into practice?
See how Monitory helps manufacturing teams implement these strategies.