Why a Bearing Fails in a Way That Names Its Own Cause

Why this matters

A failed bearing is a written record. Not of how bad the failure was, which is the part everyone looks at, but of where the load was and which ring it was moving with respect to. That information is not recoverable any other way, it is destroyed by the first pass of a shop rag, and it separates a bearing that did its job and reached the end of it from a bearing that never had a chance. A shop that reads it stops replacing the same part in the same worn housing every eighteen months.

A sibling article covers how bearings fail and the warning signs they give while running. This one is about the evidence in your hand after it is out.

Before you pull it

Isolate and lock out under 29 CFR 1910.147 and confirm zero rotation, because a fan wheel coasts long after the contactor drops. Restrain a spring-loaded motor base before the belts come off; the tensioned base is stored energy under the same standard. A bearing puller under load is also stored energy: never put your face or body in line with the puller screw, use a shield or a restraint, and relieve the load before releasing anything. If you heat a component to get it off, use a controlled heater within the manufacturer's stated temperature cap rather than an open torch, wear gloves rated for that contact temperature, and treat any plating or old paint on the part as an inhalation hazard, since heated cadmium plating on older hardware releases cadmium oxide fume that requires local exhaust and respiratory protection under a written program per 29 CFR 1910.134 alongside 29 CFR 1910.1027. Clean parts with ventilation and the SDS you are entitled to under 29 CFR 1910.1200, and never spin a bearing with compressed air, which is capped at 30 psi for cleaning with guarding and personal protective equipment under 29 CFR 1910.242(b) and can burst an unloaded bearing.

The gate

Is the damage stationary with respect to the ring it sits on, and does its position match where the load path says the load should be?

Everything below is that one question asked of both rings.

The physics behind it is simple. A load that is fixed in space, like belt pull or gravity, sweeps around a ring that rotates and stays put on a ring that does not. A load that rotates with the shaft, like unbalance, does the reverse. So the damage pattern on each ring tells you which kind of load was there, and that is a different fact from how big it was.

The four combinations

Two rules fall straight out of that table and they are worth stating because they explain a fitting practice most techs learn as an unmotivated rule:

The ring that sees a rotating load gets the interference fit. If it does not, it will creep, because the load zone is constantly moving around it and there is always a portion of the ring being lifted. The ring that sees a stationary load can take a looser fit, because the load holds it in one orientation.

On a normal machine that means an interference fit on the shaft and a closer-to-slip fit in the housing. Reverse the load conditions, as on a machine where the housing rotates and the shaft is fixed, and the fits reverse with them. The rule is not about shafts and housings. It is about which ring the load is moving with respect to.

Damage that names itself by spacing rather than position

Three patterns are identified by their geometry rather than by where they sit, and they are worth knowing because each has a cause that has nothing to do with load.

Fluting is a set of evenly spaced marks running across the raceway, at right angles to the rolling direction, present on both raceways and not confined to a load zone. It is electrical current passing through the bearing, discharging across the film. The spacing is set by an electrical and mechanical interaction, not by the ball spacing. The strong contextual clue is a variable frequency drive on the machine, because the switching produces common-mode voltage on the shaft. The fix is a current path that does not go through the bearing, which means a shaft grounding device, an insulated or ceramic-element bearing, or filtering at the drive. Fitting a new standard bearing changes nothing about the current.

Measuring shaft voltage to confirm it means working on a running, energized machine, which is electrical work under 29 CFR 1910.333(b)(2) with the arc-flash risk assessment and personal protective equipment that goes with it, using a purpose-built shaft-riding probe according to its own instructions, hands clear of the shaft. Note that the arc-rated gloves that protect you electrically are exactly what you must not have near a turning shaft, so this is a two-person or a do-not-do-it task in most service shops. The fluting pattern plus the presence of a drive is usually sufficient to act on.

True brinelling is a set of indentations at exactly the rolling element spacing, with sharp edges and displaced material at the rim. It is a mounting injury: force was driven through the rolling elements instead of through the ring being fitted. The fix is a mounting method, not a part.

False brinelling is also at rolling element spacing, but the marks are shallow, dull and often reddish-brown from fretting, with no displaced rim. It comes from small oscillating motion without rotation, so it is a transport, storage or standby-machine problem: a spare on a shelf next to a compressor, a fan idle in winter beside a running one. It is not a load fault and no amount of load-path analysis explains it.

Case one: the drive-end bearing

A belt-driven supply fan at 1200 rpm. Drive-end bearing, computed equivalent dynamic load 250 lbf, catalogue rating 6,000 lbf. It failed at about 30,000 running hours, five years in.

The evidence: outer race carries a clean matte band of about 130 degrees, centred on the direction the load path says the resultant points. Inner race is uniformly matte all the way round. Damage is subsurface fatigue spalling, a flake with a shell-shaped crater, sitting inside the band and nowhere else. Outer ring outside diameter clean and unmarked; housing bore in tolerance.

Run the gate. Stationary ring, localized band, position matches the computed resultant. Rotating ring, full circumference. Both normal. The bearing was loaded exactly the way the machine was designed to load it, and it failed by the mechanism a correctly loaded bearing eventually fails by.

Then check whether it lasted as long as that load says it should. Rating life goes as (C divided by P) cubed for a ball bearing, in millions of revolutions. Here C over P is 6,000 over 250, which is 24, and 24 cubed is 13,824 million revolutions. At 1200 rpm that is 13,824 million divided by 1,200, which is 11.52 million minutes, or 192,000 hours.

It failed at 30,000 hours, about 16 percent of that. L10 is a 90 percent reliability figure computed at the standard's reference conditions for lubrication and cleanliness, so one unit failing before it is not by itself a defect. A factor of six below is well outside the scatter you would shrug at.

The conclusion is that load is not the problem and there is no load-path fix here. The gap between 192,000 and 30,000 hours lives in the life-modification factor, which is where contamination and the viscosity ratio act, and they interact rather than adding. That is where the investigation goes next: seal condition, lubricant, and the operating temperature that sets the viscosity ratio.

Case two: the fan-end bearing on the same shaft

Same machine, other end. Same computed load class, no unusual duty.

The evidence: outer raceway is polished and loaded around the full circumference with no distinguishable band. Inner raceway is uniformly matte around its full circumference, with no localized zone anywhere on it. The outer ring outside diameter carries a bright polished ring and patches of reddish-brown powder. The housing bore measures above the top of the manufacturer's tolerance band for that size, and carries matching fretting.

Run the same gate and it resolves the other way. A full-circumference path on the stationary ring means the load is rotating with respect to it. There are only two ways that happens: the load itself rotates, or the ring does. The discriminator is the inner ring. Unbalance is a load fixed relative to the shaft, so it would leave a localized heavier zone on the inner raceway. There is none. The load is fixed in space, so the ring is what moved.

Verdict: the outer ring has been creeping in an oversize housing bore. The bearing did not fail from anything in its own duty. The housing did.

The part where being right is not enough

Case two is where good technicians lose the job. The diagnosis is correct, the part is genuinely destroyed, and a new bearing goes in and runs. Nine months later it is back.

Replacing a bearing that was doing its job correctly, in the bore that destroyed it, reaches the same end state as doing nothing, one step slower and with a part number on the invoice. Before the new bearing goes in, the housing bore is measured against the manufacturer's tolerance for this bearing size and this load condition, and if it is out, the durable repairs are boring and sleeving the housing, replacing the housing, or moving to a mounted unit with a replaceable insert. A retaining compound is a legitimate repair only within the gap range its own datasheet states, so read that number rather than treating the compound as a general-purpose fix for a worn bore.

The same discipline applies to a shaft seat that has fretted: the shaft is now undersize, the inner ring will creep on it in turn, and the fluting or the smearing you find next time will be a consequence of a repair, not of the machine.

How to preserve the evidence so the read is valid

The read above depends on knowing which ring was which and which way was up. Almost all of that is lost in the first minute of disassembly, so:

Mark before you move anything. A scribe or paint mark across the housing split line, the outer ring face, and the shaft, in the orientation as found. Once the bearing is on the bench there is no way to recover which direction the band was pointing, which is the entire content of the gate.

Photograph before you clean. Grease colour, fretting powder and discolouration are evidence, and solvent removes them. Photograph the housing bore and the shaft seat too, not just the bearing.

Keep the mating parts available. Case two is unreadable without the housing. If the bearing arrives at the bench alone, the best you can say is that the outer path went all the way round, which is two causes wide.

Split the failure from the consequence. A bearing that ran to destruction destroys its own evidence: everything is blue, everything is scored, and every ring is loaded all the way round because the geometry is gone. Where a machine is making noise, pulling it early gives you a readable bearing. Running it to seizure gives you a sample that names nothing, and that is the real cost of letting one go.

References

• ISO 15243 for the standard classification of rolling bearing failure modes and their appearances
• ISO 281 for the rating life relationship and the reference conditions the calculated life assumes
• Bearing manufacturer catalogue for shaft and housing fit tolerances by load condition and bearing size, which is where the creep judgement is made
• 29 CFR 1910.147 (hazardous energy, stored energy in a tensioned drive and a loaded puller), 29 CFR 1910.333(b)(2) (electrical work on an energized machine), 29 CFR 1910.242(b) (compressed air for cleaning), 29 CFR 1910.134 and 29 CFR 1910.1027 (respiratory protection program; cadmium)
• See related: How Bearings Fail and the Warning Signs; How to Tell Whether a Bearing Is Loaded the Way It Was Meant to Be; The Lubrication Regimes and Which One You Are In