What does the bathtub curve tell you about maintenance?

It tells you which strategy fits each phase. Infant mortality calls for burn-in, commissioning tests, and install quality control. The random useful-life phase calls for condition monitoring, because you cannot schedule random failures away. Wear-out calls for time- or usage-based replacement set just before the failure rate climbs. Applying a fixed replacement schedule during the random phase wastes effort and can even raise failures by resetting items to infant mortality.

Why do machines fail right after maintenance?

Because intrusive maintenance resets equipment to the start of the infant-mortality curve. Rebuild errors, wrong parts, disturbed connections, and contamination introduced during the work create a temporary spike in failure rate. If your failures cluster in the days after PMs, it is a signal about maintenance quality and intrusive-PM risk, investigate the work rather than shortening the interval.

Do all machines follow the bathtub curve?

No. Nowlan and Heap's 1978 study of aircraft components found six failure patterns, and only about 11% showed an age-related wear-out zone where scheduled replacement helps; roughly 89% failed randomly with no wear-out age, and about 68% followed an infant-mortality-then-constant pattern. Complex assets tend to fail randomly; simple single-failure-mode items are more likely to show real wear-out. Your own failure history decides which is which.

What is the difference between infant mortality and wear-out failures?

Infant mortality failures happen early and their rate falls over time, they come from defects, design errors, bad installation, and maintenance mistakes, not age. Wear-out failures happen late and their rate rises over time, they come from age-related mechanisms like fatigue, erosion, and corrosion. The two look opposite on the curve: one weeds out weak units, the other reflects an aging population reaching the end of its life.

How does the bathtub curve relate to the Weibull distribution?

The three phases map to the Weibull shape parameter beta. Beta less than 1 gives a decreasing failure rate (infant mortality), beta equal to 1 gives a constant rate (random failures, the exponential case), and beta greater than 1 gives an increasing rate (wear-out). Reliability engineers fit Weibull distributions to failure data to identify which phase a component is in and whether a wear-out age exists.

Should I replace parts on a fixed schedule?

Only for failure modes that have a real wear-out age, where the item genuinely becomes more likely to fail as it ages. For random failure modes (the majority in complex equipment), fixed-interval replacement does not lower the failure rate and can introduce infant-mortality failures. Use the asset's failure history to tell wear-out modes from random ones, and reserve scheduled replacement for the former while monitoring the latter.

The Bathtub Curve in Reliability Engineering

Q: What are the three phases of the bathtub curve?

Infant mortality, useful life, and wear-out. In infant mortality the failure rate is high but falling, driven by defects, bad installs, and maintenance errors. In useful life the rate is low and roughly constant, and failures are essentially random. In wear-out the rate rises as age-related mechanisms, fatigue, erosion, corrosion, take over. Only the wear-out phase has an age at which failure becomes more likely.

Infant mortality, useful life, and wear-out, what each phase of the failure-rate curve means, and why most equipment doesn't actually follow it.

The bathtub curve is a reliability-engineering model of how an asset's failure rate changes over its lifetime. It has three phases: infant mortality (a high but falling failure rate from defects and bad installs), useful life (a low, roughly constant rate of random failures), and wear-out (a rising rate as components age). Plotted together, they trace the cross-section of a bathtub.

It is one of the most-drawn diagrams in maintenance, and one of the most misapplied. Used well, it explains why brand-new machines fail, why a constant failure rate is normal, and when scheduled replacement finally makes sense. Used badly, it convinces plants to overhaul equipment on a calendar that has nothing to do with how that equipment actually fails. This guide walks the three phases, what each means for your PM schedule and the crucial finding that most modern equipment does not follow the bathtub at all.

What are the three phases of the bathtub curve?

Each phase has a different dominant failure cause, which is why each calls for a different response.

The three phases, each mapped to the maintenance strategy that fits it. Time-based replacement only makes sense in the wear-out region on the right, and that region does not exist for most equipment.

Infant mortality (decreasing failure rate). New or freshly repaired equipment fails at an elevated but falling rate. The causes are not age, they are manufacturing defects, design errors, incorrect installation, and maintenance-induced problems. The population weeds out its weak units early, and the survivors settle into a lower rate. The maintenance lesson: failures right after commissioning or a PM are usually a quality problem, not a wear problem.
Useful life (constant failure rate). The long, flat middle. Failures still happen, but at a low and roughly constant rate, and they are essentially random a surge here, a stray contaminant there, an operating overload. Because the rate is constant, an item is no more likely to fail tomorrow than it was last month. This is the phase where MTBF is most meaningful and where scheduled replacement does the least good.
Wear-out (increasing failure rate). Eventually, age-related mechanisms, fatigue, erosion, corrosion, insulation breakdown, take over and the failure rate climbs. This is the only phase where an item genuinely becomes more likely to fail as it gets older, and therefore the only phase where time-based replacement is well matched to the physics.

Why does the shape matter for maintenance?

Because the curve tells you when time-based maintenance helps and when it is wasted, or harmful. In the constant-failure-rate middle, replacing a component on a fixed schedule does not lower the failure rate, since failures are random rather than age-driven. Worse, every intrusive PM drops the item back to the start of the infant-mortality curve, temporarily raising its failure rate. That is the mechanism behind a familiar plant experience: things break right after you work on them.

The curve maps cleanly onto the reliability strategies on the equipment reliability ladder:

Infant mortality → quality and burn-in. Commissioning tests, incoming inspection, installation standards, and disciplined rebuild quality catch weak units before they reach production. This is why correct installation is a reliability lever, not just a maintenance task.
Useful life → condition monitoring. When failures are random, you cannot schedule them away, you can only watch for the ones that give warning. Vibration, temperature, and oil analysis under condition-based maintenance and predictive maintenance catch degradation in time to act.
Wear-out → scheduled overhaul or replacement. Once an item is provably in its wear-out zone, a time- or usage-based replacement interval set just before the failure-rate climb is exactly right.

Does every asset follow the bathtub curve?

No, and this is the finding that reshaped modern maintenance. When United Airlines analyzed the actual failure patterns of aircraft components, published by Nowlan and Heap in 1978, they found six distinct patterns, and the classic bathtub described only a small minority. About 89% of the items showed no wear-out zone at all their failures were random, while only about 11% had an age-related pattern where a scheduled replacement age would help.

The six patterns behind the average. Only patterns A, B, and C have a wear-out age where scheduled replacement helps, and together they are 11% of items. The other 89% fail randomly, which is why condition monitoring beats the calendar for most equipment.

The caveat: that study was aircraft components, and the exact percentages should not be copied onto a food plant or a machine shop. But the qualitative lesson holds broadly, complex, multi-component assets tend to fail randomly, and blanket time-based overhauls of such equipment often add cost and infant-mortality risk without cutting failures. Simple single-failure-mode items (a specific bearing, a belt, a filter) are far more likely to show a real wear-out age. The art is knowing which of your assets is which, and your own failure history is the only reliable guide.

How is the bathtub curve different from the P-F curve?

These two curves get confused constantly, and keeping them straight is worth the effort because they answer different questions. The bathtub curve describes a population over its whole life: what fraction of many identical items are failing at each age. It is a planning tool, it tells you whether a scheduled replacement age makes sense for a class of parts.

The P-F curve by contrast, describes a single failure as it develops: the interval from the point a defect first becomes detectable (P, potential failure) to the point of functional failure (F). It is a detection tool, it tells you how much warning a given condition-monitoring technique buys you before a specific item quits. Vibration analysis reading bearing defect frequencies catches a fault early on the P-F curve; temperature catches it late.

The connection is that the P-F curve is what makes condition-based maintenance viable precisely in the random, useful-life stretch of the bathtub where scheduled replacement fails. When you cannot predict which item will fail or when from age alone, you monitor each one for the onset of its own P-F interval instead. One curve tells you the calendar is useless here; the other tells you what to do about it.

How do you use the bathtub curve in a real PM program?

Classify the failure mode, not the machine. Ask which pattern a specific failure mode follows, using that asset's history. A pump might have one wear-out mode (impeller erosion) and three random ones (seal, bearing contamination, electrical). Each mode gets its own strategy.
Match the strategy to the phase. Wear-out modes get usage-based replacement just before the rate climbs; random modes get condition monitoring; infant-mortality risk gets tighter commissioning and rebuild quality control.
Stop scheduling intrusive PMs against random failures. If a failure mode is random, a fixed-interval teardown will not prevent it and may cause new infant-mortality failures. Replace the calendar task with an inspection or a condition trigger.
Watch for post-maintenance failure clusters. Failures in the days after a PM are the infant-mortality wall, a signal about rebuild quality and intrusive-PM risk, worth a root cause analysis not a reason to shorten the interval.
Let the data move the intervals. Track failures per asset and per mode over time. Rising failure rate with age confirms a wear-out interval; a flat rate says stop replacing on a schedule and monitor instead. Feed it all into your MTBF trend and maintenance KPIs.

What the research and numbers say

Statistically, the three phases correspond to the Weibull shape parameter β: β < 1 is a decreasing failure rate (infant mortality), β = 1 is a constant rate (random, the exponential distribution), and β > 1 is an increasing rate (wear-out). The bathtub is these three regimes stitched together (overview: bathtub curve).
The landmark finding that roughly 89% of failures are random and only ~11% are age-related with about 68% following an infant-mortality-then-constant pattern comes from Nowlan and Heap's 1978 reliability-centered-maintenance study of aircraft components, the foundation of modern RCM. The economics of shifting failures from unplanned to planned are documented by the U.S. Department of Energy and PNNL: predictive programs save an estimated 8–12% over preventive-only maintenance (PNNL, O&M Best Practices: Maintenance Approaches).

Using the curve well depends on knowing how each asset actually fails, and that requires failure history that is complete and honest. Harmony pulls machine signals, downtime reasons, and maintenance records into one operational data layer, so wear-out modes separate from random ones in the data instead of hiding in a paper log, and it can flag the pattern and draft the work order for a human to approve. It layers onto the CMMS and machines you already run, no rip-and-replace; see how it works or the CLS case study. For the strategy this feeds, see equipment reliability predictive maintenance and how the phases connect to availability versus reliability.