Predictive vs preventive maintenance is an asset-level choice. Preventive work runs on a fixed schedule; predictive work is triggered when condition monitoring shows a failure developing. DOE guidance puts preventive savings at 12% to 18% over reactive and predictive at a further 8% to 12%, but predictive pays only where an asset gives measurable warning before failing.
The evidence points one way. Buying sensors for the whole plant is the wrong question. The useful question is which maintenance strategy each asset has earned, and that depends on how the asset fails, what happens to the plant when it does, and how long a replacement part takes to arrive. This guide sets out the benchmarks, the reliability research behind them, and a ranking method a plant manager can apply to an asset list before the next capex cycle.
The benchmarks, and what they do not promise
The most widely cited public benchmark is the operations and maintenance best-practices guide that Pacific Northwest National Laboratory maintains for the Department of Energy's Federal Energy Management Program. It says preventive maintenance savings "can amount to as much as 12% to 18% on average" compared with reactive maintenance. It also says that "a properly functioning predictive maintenance program can provide a savings of 8% to 12% over a program that utilizes preventive maintenance alone." A facility that relies heavily on reactive maintenance, the guide adds, "could easily recognize savings opportunities exceeding 30% to 40%."
Read those numbers carefully. They are averages from past studies, not a guarantee for any given line. The predictive figure also carries a condition: the program has to be "properly functioning." A vibration sensor on a non-critical fan does not earn 8% to 12% of anything. The same PNNL page lists the risks for both approaches. For preventive maintenance these are catastrophic failures that are still likely, heavy labor, unneeded maintenance, and incidental damage caused during that unneeded work. For predictive maintenance they are higher investment in diagnostic equipment, higher investment in staff training, and "savings potential not readily visible to management."
Why calendar PMs often miss
Calendar-based preventive maintenance assumes that equipment wears out on a predictable schedule, so replacing or overhauling it at a fixed interval heads off failure. Reliability-centered maintenance (RCM) research tested that assumption directly.
The foundational study is the 1978 Reliability-Centered Maintenance report by F. Stanley Nowlan and Howard Heap. It was written at United Airlines under U.S. Department of Defense sponsorship (report AD-A066579), and RCM is now defined by SAE JA1011 (1998), with guidance in SAE JA1012 (2002). Nowlan and Heap sorted component failures into six conditional-probability patterns. Accendo Reliability's summary gives the shares as follows:
-
Pattern A, bathtub: about 4%
-
Pattern B, wear-out: about 2%
-
Pattern C, fatigue: about 5%
-
Pattern D, initial break-in: about 7%
-
Pattern E, random: about 14%
-
Pattern F, infant mortality: about 68%
Only patterns A, B and C show age-related behavior, and together they account for about 11% of failures. For everything else, a fixed-interval overhaul does little to stop the failure. Where infant mortality dominates, an overhaul can make things worse, because it resets the component to its riskiest period. That matches PNNL's warning about incidental damage during unneeded maintenance.
One caveat matters for an industrial audience. The Nowlan and Heap data came from airline components. A plant's mix of pumps, gearboxes, conveyors, presses and hydraulics will not match those percentages exactly. Treat them as a signal of direction: age-based replacement is the right tool for a minority of failure modes, not the default for every asset on the PM schedule.
The P-F interval test
If most failures are not age-related, the next question is whether they give a warning. Predictive maintenance depends on the P-F interval. ISA's InTech magazine (Kevin Clark, CMRP, 2019) defines it as "the time between when potential failure is detected in an asset and when it reaches the failed state."
The length of that interval depends on the detection technology. According to InTech, "With vibration, we can see indications of faults 12 to 18 months in advance." The article ranks oil analysis and ultrasound among the earliest indicators and thermography as a late one. It also makes a point that bears directly on budgets: "The more often assets are inspected and the more sensitive the method of inspection, the more time there will be between detection of potential failure and when failure actually takes place."
Condition monitoring earns its cost only when three things are true:
-
The failure mode produces a measurable precursor, such as vibration, wear debris in oil, ultrasonic emission or heat.
-
The warning arrives early enough to schedule labor and a production window.
-
The warning arrives early enough to get the replacement part. If a gearbox has a 20-week lead time and your monitoring method only flags trouble a few weeks ahead, the sensor tells you about a failure you still cannot prevent.
The third condition is the one most often left out of the business case, and it is where spare-parts strategy and maintenance strategy meet.
A three-question ranking for every asset
The framework below is ManufacturingMag's editorial synthesis of the RCM and P-F research above. It is not a published standard. Score each asset on the asset register on three questions.
1. How does it fail? Sort the dominant failure modes into three groups:
-
Age-related, meaning wear-out with a reasonably predictable life
-
Random, but with a detectable warning
-
Random, with no practical warning
2. What happens when it fails? Check whether it stops the constraint or bottleneck operation, whether there is redundancy, and whether the failure creates a safety hazard or a quality escape.
3. How long do parts take compared with the warning? Compare the lead time for critical spares, including repair or rebuild time, with the P-F interval your detection method can realistically deliver.
The answers point to one of four strategies: run-to-failure, calendar preventive maintenance, route-based manual condition monitoring, or continuous online monitoring. The table below shows how they map.
| Asset profile | Dominant failure behavior | Criticality and spares | Suggested strategy | | --- | --- | --- | --- | | Consumable and wear items with predictable life (belts, filters, wear liners) | Age-related wear-out | Varies; parts usually stocked | Calendar or usage-based preventive maintenance | | Critical rotating equipment on the constraint (main drives, large gearboxes, process pumps) | Random, with detectable precursors | High; long-lead or custom spares | Continuous online vibration and/or oil monitoring | | Moderately critical equipment with some redundancy or stocked spares | Random, with detectable precursors | Medium; spares available in reasonable time | Route-based handheld condition monitoring and inspection | | Cheap, non-critical, quickly replaced items (small motors, lighting, off-line auxiliaries) | Random, little or no warning | Low; off the shelf | Run-to-failure, with spares on hand | | Critical equipment that fails randomly with no practical warning | Random, no detectable precursor | High | Engineering fix: redundancy, redesign or strategic spares; neither PM nor PdM will prevent it |
The middle row is where plants tend to overspend. A Plant Engineering article by Bryan Christiansen, founder and CEO of the CMMS vendor Limble, argues that "no technique works for each asset class." Some equipment justifies IoT sensors, while for other equipment manual inspections plus basic sensors are enough. The author has a commercial interest in maintenance software, but the point is consistent with the P-F logic above: route-based checks with a handheld instrument can give adequate warning on an asset whose failure is survivable and whose parts are close at hand.
The last row is the uncomfortable one. Some critical failures give no usable warning. The answer there is an engineering or inventory decision, not a maintenance schedule.
What the survey data says about outcomes
The largest U.S. dataset on maintenance outcomes comes from NIST. In a 2021 paper in the International Journal of Prognostics and Health Management, Douglas Thomas and Brian Weiss estimated U.S. manufacturers' total annual maintenance costs and losses at an average of $222.0 billion. The paper builds on their earlier survey report, NIST AMS 100-34, Economics of manufacturing machinery maintenance (June 2020).
The comparisons are striking:
-
Respondents relying more on predictive and preventive maintenance than on reactive maintenance reported 52.7% less unplanned downtime and 78.5% fewer defects.
-
Respondents relying more on predictive than on preventive maintenance reported 18.5% less unplanned downtime and 87.3% fewer defects.
These are survey correlations, not results from a controlled trial. Plants that invest in predictive programs may also differ in management discipline, capital intensity or product mix. NIST is candid about the limits. Its June 2021 news release says maintenance "is estimated to be between 15% and 70% of the cost of goods sold." Manufacturers have reported "35% to 45% reductions in downtime and 65% to 95% reductions in defects" from advanced maintenance, and cited reductions in maintenance cost from predictive maintenance range from 15% to 98%. The release also says the effect at the national level "is not well documented." A range of 15% to 98% is not a forecast. It tells you results depend heavily on the asset and the execution.
The defect figures deserve attention from quality managers. Maintenance strategy affects scrap and escapes as well as uptime, so the business case for monitoring a spindle or a press may rest as much on quality as on availability.
Size the payback before buying sensors
Headline downtime figures are easy to find and hard to apply. AEMT's reporting of Siemens' True Cost of Downtime 2024 puts the loss from unplanned downtime at about $1.4 trillion a year across the world's 500 largest companies, equal to 11% of their revenues. It also says an idle line at a major automotive plant can cost up to $2.3 million per hour. The same report shows improvement: monthly downtime incidents fell from 42 in 2019 to 25, and average monthly unplanned downtime fell from 39 hours to 27.
Those numbers describe the Global 500. A mid-market machine shop or tier-2 supplier should not build a sensor business case on an automaker's hourly figure. The relevant number is your own, and it has two parts that are often lumped together:
-
Lost contribution: the margin on output you could not make and cannot recover. This applies mainly when the failed asset is the constraint and demand exceeds capacity.
-
Incremental recovery costs: overtime, expedited freight on parts, outside repair, scrap from restart, and premium freight to the customer.
If a non-constraint asset fails and the lost output can be made up on the next shift, lost contribution may be close to zero and only the recovery costs count. ManufacturingMag's free downtime cost calculator separates the two. Run it for each candidate asset before you get a monitoring quote. An asset whose realistic failure cost is small will not pay back continuous monitoring, however good the sensor is.
The costs that do not show up on the sensor quote
PNNL's risk list for predictive maintenance is short and accurate: diagnostic equipment, training, and savings "not readily visible to management." The first item is the hardware quote. The second and third are where programs stall.
-
Analyst time. Vibration spectra, oil reports and ultrasound readings need someone qualified to interpret them. Whether that is an internal analyst or a contracted service, it is a recurring cost.
-
Response workflow. An alert that does not become a planned work order, a parts requisition and a scheduled production window has no value. The CMMS and the planning process have to be ready before the alerts start.
-
Proving the savings. Avoided failures leave no trace in the P&L. Record each confirmed catch, the failure mode, the estimated avoided cost (using the same contribution and recovery split as above) and the part lead time. Without that record, the program is the first line cut in a downturn.
The same discipline applies to preventive maintenance, where the hidden costs run the other way: labor spent on tasks that prevent nothing, and the risk of damaging equipment while doing them.
A checklist for the next capex cycle
-
Start where three conditions overlap. Pick the few assets that are critical to the constraint or to safety and quality, have long spare-parts lead times, and fail in ways that give detectable warning. These are your candidates for online monitoring.
-
Keep calendar PMs where wear-out dominates. Belts, filters, lubrication and wear parts with predictable life still belong on a schedule.
-
Use route-based condition monitoring for the middle tier. Handheld instruments and trained inspectors cover moderately critical assets without continuous sensors.
-
Run cheap, non-critical items to failure on purpose. Stock the spares and document the decision so it is a strategy, not neglect.
-
Audit the PM list. For each task, name the failure mode it prevents. Drop or change tasks that have none, especially intrusive overhauls on components whose failures are not age-related.
-
Price each candidate asset first. Use the downtime cost calculator to separate lost contribution from recovery costs, then compare the result with the full program cost: hardware, analysis, training and planning time.
-
Match the detection method to the lead time. If the parts take longer than your monitoring method's warning, either choose an earlier-warning technology or stock the spare.
Assign each asset the strategy it earns, and the choice between predictive and preventive maintenance mostly takes care of itself.
Related reading
Sources
-
O&M Best Practice Issue Discussion: Maintenance Approaches, PNNL for DOE FEMP
-
Economics of manufacturing machinery maintenance (NIST AMS 100-34), NIST
-
Potential Cost Savings as US Manufacturers Spend Billions on Machinery Maintenance, NIST
-
How to extend the P-F interval for critical assets, Plant Engineering
-
The True Cost of Downtime 2024: A Comprehensive Analysis, AEMT