Run-to-failure maintenance is a deliberate decision to let a specific failure mode happen without trying to prevent it. You make that decision before the failure, because preventing it would cost more than allowing it.
SAE JA1011, the standard that sets out what counts as reliability-centred maintenance, defines it at clause 3.29 as "a failure management policy that permits a specific failure mode to occur without any attempt to anticipate or prevent it."
It's a legitimate strategy. Used well, it cuts PM work that prevents nothing.
The trouble is that 3 different terms get used for it, and most of what you'll read lumps them together.
Below I separate those 3 terms, show where run-to-failure comes from, and give you the 5 questions that decide whether a failure mode qualifies. Then the examples, including a flange in hydrocarbon service.
Run-to-failure, corrective and reactive are 3 different things
Run-to-failure is a decision. Corrective maintenance is the work. Reactive maintenance is the condition your plant ends up in when nobody made the decision.
| What it is | When it happens | Who decides | |
|---|---|---|---|
| Run-to-failure | A failure management policy, 1 of 6 in SAE JA1011 | Before the failure | You do, during analysis |
| Corrective maintenance | The work you do to restore the item | After the failure | It follows from the failure |
| Reactive maintenance | A plant dominated by failures | Continuously, until you change it | Nobody decides. It builds up by default |
You'll often see run-to-failure listed as "also known as" breakdown maintenance or reactive maintenance. I don't agree.
I also see reactive and proactive as descriptions of your maintenance culture. Are you chasing failures, or are you continuously improving?
When you opt for run-to-failure, you're opting for deferred corrective maintenance. Deferred, because the work can wait. It goes through planning and scheduling like any other job.
A run-to-failure decision should never lead to emergency maintenance. Emergency work is new work, not in the CMMS before the week started, so urgent it breaks into the frozen weekly schedule.
If the failure would force that kind of repair, it was the wrong call.
Where does run-to-failure come from?
Nowlan and Heap's 1978 report for United Airlines gave reliability-centred maintenance its name and its first full write-up. Search the full report for "run to failure" and you won't find it.
Their term is "no scheduled maintenance". On their decision diagram, Exhibit 4.4, it's a box you arrive at by answering a sequence of questions. And it sits on 2 branches only, both of them economic. You can't reach it on the safety branch or the hidden-function branch.
Page 80 puts the economics plainly: "It is less expensive to leave them in service until they fail than it would be to prevent the failures."
SAE JA1011, published in 1999, uses the name run-to-failure. It lists run-to-failure as 1 of 6 failure management policies, alongside on-condition tasks, scheduled restoration, scheduled discard, failure-finding and one-time changes.
Naval aviation calls it "No PM". The US Navy's RCM process makes that option the baseline every proposed task has to beat.
Apply that test to your own PM schedule and see how many tasks survive it.
| Term | Where it comes from | What it means |
|---|---|---|
| Run-to-failure | SAE JA1011 (1999) | A policy that lets a failure mode happen with no attempt to anticipate or prevent it |
| No scheduled maintenance | Nowlan & Heap (1978) | The same decision, 21 years earlier, on the 2 economic branches only |
| Corrective maintenance | ISO 14224 | Work after a fault to restore the item. It says nothing about whether the failure was chosen |
| Deferred corrective maintenance | EN 13306 | Corrective work delayed in line with your maintenance rules. What a good run-to-failure decision produces |
| Reactive maintenance | US DOE O&M Best Practices Guide | "Basically the 'run it till it breaks' maintenance mode." A catch-all for a plant in firefighting mode |
Why do so many failure modes qualify?
More failure modes qualify than you might expect. Nowlan and Heap's failure pattern chart shows why.
Exhibit 2.13 in Nowlan and Heap splits the United Airlines data into 6 failure patterns. Only 11% of items showed a pattern where an age limit might help. The other 89% didn't.
In their words: "Some 89 percent of the items analyzed had no wearout zone; therefore their performance could not be improved by the imposition of an age limit."
Only a handful of items followed the bathtub curve: "Just 4 percent of the items fell into this pattern."
Page 48 goes further: "in many cases scheduled overhaul actually increases the overall failure rate by introducing a high infant-mortality rate in an otherwise stable system."
You can see the effect on page 5. Under the old hard-time logic, the DC-8 program put 339 items on scheduled overhaul. The DC-10 program, built on the new MSG-2 logic, put 7. Spare engine inventory fell by more than 50%.
The 70% to 90% band I teach for failures that aren't age related comes from 4 studies, 2 in airlines and 2 in the US Navy. The Navy data shows more wear-out, mostly because ships work in a far more corrosive environment.
Where the likelihood of failure is constant over time, a time-based task can't help you. For an evident failure mode with no wear-out, that leaves you condition-based maintenance, run-to-failure, or a change to the design.
The 5 questions that decide it
Ask these in order, and assess the consequences as if nothing is currently being done to prevent or detect the failure. "We've always run it to failure and it's been fine" doesn't count as evidence.
1. Is the failure hidden or evident?
If the failure is hidden, you need a failure-finding task, or a very good reason why not.
2. What consequences does the failure mode have?
In most cases, if a failure mode has safety, environmental or reputational consequences, you won't accept run-to-failure. The flange later in this article is the exception, and I'll show you why.
Realistically, you'd only consider run-to-failure where a failure mode has financial consequences only. By financial I mean the repair cost plus the impact on production.
SAE JA1011 sets the boundary in mandatory language at clause 5.8.2. Where the failure is evident, the failure mode "shall not have safety or environmental consequences".
2 things to be clear about here. Reputational consequences are my addition, and the standard doesn't list them. And where JA1011 splits the economic side into operational consequences (it hits production) and non-operational ones (a repair bill and nothing more), I fold both into financial.
3. Would letting it fail cause consequential damage?
Would the failure of this component take other components with it, or damage the production line?
Consequential damage is easy to overlook. Miss it and a cheap bearing failure can take a gearbox with it.
4. Would the restore break into your frozen weekly schedule?
If yes, what you've chosen is reactive maintenance with a nicer name.
Emergency maintenance is expensive and inefficient, and it's often less safe. The operational split from question 2 matters here. A failure that only costs you a repair bill can usually wait a week. A failure that stops production usually can't.
5. Is the job plan written?
Write it now, before the failure. It needs the part number and bill of materials, minimum and maximum stock levels, craft hours, tools and lifting arrangements, isolation and permit requirements, and a target duration.
That plan is what turns the failure into deferred corrective work.
A hidden failure, a safety or environmental consequence, consequential damage, or a restore that breaks the frozen week: any one of those ends the conversation. Then sanity-check the likelihood. A failure mode with small consequences that fails every few weeks can still cost you more than the PM would.
When everything passes, the standard is clear. Where no cost-effective proactive task can be found, SAE JA1012 says run-to-failure "should be selected as the appropriate failure management policy."
It also warns against "default logic", where decisions get made automatically because good data isn't available.
Hidden failures need a failure-finding task first
Treat every hidden failure as a failure-finding candidate first. You'd only leave one undetected if the multiple failure it sets up has no safety or environmental consequence. In my experience that's rare.
A hidden failure mode is one you can't see in normal operation. It only becomes evident when something else fails as well. Most hidden failures sit in protective functions and standby equipment.
Take a duty pump and a standby pump. If the standby has failed, you don't know about it. You find out when the duty pump stops and the standby doesn't start.
Nowlan and Heap put the rule this way on page 96: "we can allow such items to fail, but we cannot afford the possible consequences of undetected failures."
Note the word undetected. Their arithmetic on pages 28 and 29 shows why it matters.
Take 2 items, each with a 99% chance of surviving a flight. The chance both fail on the same flight is 0.0001, or once in 10,000 flights. Now let the first one sit there failed and undetected. The probability becomes 0.01, or once in 100 flights.
The risk goes up 100 times, just because the first failure went undetected.
The answer is failure-finding maintenance: a periodic functional test, ideally end to end, that checks whether the hidden function still works. It finds a failure that has already happened, and you still do the corrective work.
If you reconfigure 2 identical pumps from both running to duty and standby, you've just created a hidden failure mode on the standby, and it needs a failure-finding task.
The same fan, 3 different answers
The numbers below illustrate the method. They aren't site data.
The asset is an auxiliary air-cooled heat exchanger fan, 22 kW, 1 of 6 on a common bank. The failure mode is motor bearing seizes.
A replacement motor costs US$3,000 delivered, and the swap takes 2 people 4 hours, so call the restore US$4,000. Site history gives about 1 failure per fan every 8 years, or US$500 a year. The PM on the table is a quarterly vibration route plus annual greasing, at US$250 a year.
Case A. The bank has 20% design margin at summer ambient.
The failure is evident. The consequence is financial only. The restore fits into the next scheduled week, and the motor is a stock item.
A vibration route would give you warning. With a motor on the shelf and 20% margin on the bank, that warning buys you nothing you need.
Nowlan and Heap's rule on page 96 settles it. If the 2 costs are of the same order of magnitude, the answer is no scheduled maintenance, because any task may disturb a stable mechanism. Run it to failure.
Case B. Same fan and failure mode, but no spare in stores and a 14-week lead time on the motor.
This time the consequence is bigger. Running the bank 1 fan down through a summer is an operating restriction.
Buy the spare. US$3,000 of stock removes 14 weeks of exposure and takes you back to case A. In JA1011 terms that's a one-time change, another policy on the same list as run-to-failure.
Case C. Same fan and failure mode, but this bank cools a lube-oil circuit protected by a high oil temperature trip.
Go back to question 1. The fan failure is still evident, but the trip is a hidden function. If the trip has also failed when the fans go, you get a lube-oil temperature excursion on a machine that won't shut itself down.
The fan stays on run-to-failure. The trip's multiple failure has safety consequences, so it gets a failure-finding task, tested end to end.
Examples beyond the light bulb
You don't go around your house replacing light bulbs before they fail, do you? They have a constant probability of failure during their useful life, so replacing them early wastes time and money.
The same common sense doesn't always make it into industrial plants. As a rule of thumb, run general area lighting to failure.
There are 2 exceptions. Emergency lighting may need failure-finding maintenance, because those lights aren't normally on and a failure would be hidden.
Lights on exhaust stacks and flares are there for aviation, and in most countries they're a legal requirement. Where you can only change them at shutdown, go for time-based replacement on a conservative OEM service life. You're deliberately over-maintaining those lights and accepting the cost, because the business can't accept the consequence of failure.
The bigger savings are usually in process instrumentation.
I know a lot of plants that religiously calibrate all their process instrumentation every 1, 2 or 4 years. My advice would be don't.
Modern smart transmitters perform well, and most of their failure modes show a constant probability of failure across their service life. A failure will often cause a process upset that a control room operator can correct, and a separate trip instrument usually backs that up. So the real consequence is low.
That advice has 2 limits. It applies to instruments without a trip or alarm function, because protective instruments need failure-finding. And some instruments are critical, or do have age-related failure modes.
Even so, many plants can cut their instrumentation maintenance substantially by dropping most routine calibrations. If you're worried about instrument reliability, use the diagnostics your smart transmitters already give you. That beats a traditional calibration round.
Gas turbines look like the obvious no. They're critical, complex and expensive, and any stoppage hurts. But break the turbine down into its components and failure modes, and you'll almost certainly find some where run-to-failure is right.
Make the run-to-failure decision one failure mode at a time, whatever the asset costs.
Why you can run a flange to failure
I've spent most of my working life in plants handling hydrocarbons. Oil, gas, sometimes hydrogen sulphide, which is highly toxic. People's lives depend on flange integrity.
So a flange leak would have unacceptable consequences, and you'd never accept run-to-failure. Right?
Well, wrong. But only if one condition is met.
With flanges, the failure pattern is one of infant mortality. Once a flange has been made up well and to specification, the likelihood of a leak is very small and pretty constant. Quality management during make-up gets you through that phase, plus leak testing before start-up where the service calls for it.
This is a likelihood call. Once a flange is made up right and leak tested, no scheduled task will lower the chance of a leak. There's no on-condition task that tells you a compliant joint is about to leak, and no meaningful age limit either. A periodic break-and-remake would bring back the very infant mortality you just paid to remove.
The other option is a line on a PM schedule that somebody signs off without doing anything useful. Paper is cheap, and paper is patient. That line can sit in your PM schedule for 20 years at 100% compliance and prevent nothing.
The failure modes that do matter are managed outside the PM schedule.
| Failure mode | How it's managed | Where it's managed |
|---|---|---|
| Infant mortality, poor make-up | Make-up sheets, joint integrity procedures, torque control, competency | Work execution and quality control |
| Infant mortality, leak on start-up | Leak testing before the joint goes into service, where the service calls for it | Commissioning and return to service |
| Dissimilar-metal corrosion | Designed out, or the metals physically separated | Design and construction |
| External corrosion | A risk-based coating program, with regular inspections setting its scope and timing | Coating and inspection program |
| What's left | A very low, constant likelihood of a leak, with no task that lowers it | Run-to-failure |
So you accept run-to-failure on the flange itself, on one condition. Your flange management practices have to be in place first.
What does run-to-failure cost?
You'll see it claimed everywhere that reactive work costs several times more than planned work.
I use 3 to 5 times as a rule of thumb, and I still believe the direction. But I can't point you to a study behind that number, and I haven't seen anyone else point to one.
The US DOE guide most often cited for it claims far less: preventive maintenance saving 12% to 18% over reactive.
Work out the cost at your own site. An emergency job has no kitting and no staging. There's no scheduled window, so you pay for premium freight and overtime. A crew stands around waiting for parts. You get secondary damage, and you carry the production loss.
Then check your emergency work. The best plants keep emergency work under 2%, measured in hours.
What's yours running at? A good run-to-failure program won't push that number up.
Where run-to-failure sits
Reliability-centred maintenance is where the decision logic comes from. Run-to-failure is the RCM default action once no proactive task is worth doing.
Reactive maintenance is what this decision turns into when it goes wrong. If your choices are generating emergency work, go back to the analysis.
Risk-based maintenance sets the amount of maintenance by the risk of each failure mode. I've used risk-based inspection to cut inspection workloads by 30% to 40%.
Don't put run-to-failure on a P-F curve. The P-F interval belongs to condition-based maintenance only.
Failure is an option
We've all been conditioned to avoid failure. But in maintenance, failure is an option, and any good preventive maintenance program has some of it built in on purpose.
Go through the failure modes you've assessed as low risk. Be objective, and mark the negligible ones as negligible. That list is where your run-to-failure candidates come from.
Run-to-failure is sometimes the best choice, and occasionally the only sensible one. Make the decision in order, and write the job plan before the failure.



