What Misalignment Does to a Bearing in Mechanical Terms
Why this matters
Everyone knows misalignment kills bearings. Almost nobody can say what it actually does inside one, which is why the same shop replaces the same bearing three times and never revisits the alignment. There are two separate mechanisms at work, they leave different marks on the raceway, and only one of them is fixed by fitting a self-aligning bearing. Knowing which one you have turns a failed bearing from a part you throw away into the piece of evidence that closes the job.
Before the bearing comes out
Open and lock the motor disconnect and prove the terminals dead with a live-dead-live check on a known live source (29 CFR 1910.333(b)(2); NFPA 70E-2021, 120.5). Work the mechanical side under 29 CFR 1910.147: confirm zero rotation, block the rotor, and isolate, lock, relieve and drain any pressurized or hot line into the machine before you open a bearing housing, because a housing on a hot pumped line will discharge scalding fluid when the cover comes off. Let hot surfaces cool or read them with a non-contact infrared thermometer; above about 140 F they burn on contact.
Bearing removal has hazards the entry lockout does not touch. A puller on an interference-fit inner ring stores a large amount of energy and the ring departs as a projectile when it releases, so use a rated puller with a shroud or a heavy blanket over it and keep the line of the screw clear. Induction heaters or a controlled hot-oil bath are the right way to expand a ring; an open flame on a greased bearing produces smoke you do not want to breathe, so if heat is used, work with local exhaust and select respiratory protection under a 29 CFR 1910.134 program rather than reaching for a nuisance dust mask. If you cut a seized outer ring out with an abrasive wheel, that is abrasive wheel machinery work under 29 CFR 1910.215: guard in place, face shield over safety glasses, and metal grinding dust controlled at the source with local exhaust and respiratory protection selected for a metal fume and particulate route, because a face shield does nothing about what you inhale.
What a healthy bearing is doing
A radially loaded rolling-element bearing does not carry the load on one ball. The load is shared across the rolling elements inside a load zone, typically somewhat less than half the circumference, with the element at the center of the zone carrying the most. Each element rolls into the zone, takes its share, and rolls out. Every pass puts one cycle of subsurface shear stress into a small volume of raceway just below the contact.
That is the fatigue clock. Rolling contact fatigue is a subsurface phenomenon: the peak shear stress sits below the surface, a crack initiates there, works its way up, and lifts a flake of raceway. That flake is a spall, and the noise and vibration everyone recognises come after it, not before.
The first mechanism: more load, and the exponent is brutal
Misalignment makes the coupling deflect, the deflected coupling produces a reaction force, and that force adds to the bearing's load. The reason a modest force increase matters so much is the exponent in the rating life relationship.
Bearing rating life scales with the load ratio raised to a power: cubed for ball bearings, and to the ten-thirds power for roller bearings, under the ISO 281 basic rating life model. That model is derived for a bearing running under a steady load, with an adequate lubricant film, correctly mounted, clean, and operating within its permissible internal misalignment. Every one of those is a condition, and two of them are exactly what a misaligned set violates.
Inside those conditions, the arithmetic is unforgiving:
- A 25 percent load increase gives a life factor of 1 divided by 1.25 cubed, which is 1 divided by 1.953, or about 0.51. A quarter more load halves a ball bearing's life.
- A 60 percent load increase gives 1 divided by 1.6 cubed, which is 1 divided by 4.096, or about 0.24. Life drops to under a quarter.
- The same 25 percent on a roller bearing, at the ten-thirds power, gives 1 divided by 2.10, or about 0.48.
Nothing here requires a dramatic misalignment. A reaction force that would not register as a problem in a static load calculation is a life problem because life is not linear in load.
The second mechanism: tilt, and where the model stops applying
The reaction force is only half of it. Misalignment also tilts the inner ring relative to the outer ring, and a bearing's tolerance for that tilt is small. Manufacturers publish a permissible misalignment for each bearing type, and for a deep-groove ball bearing it is a matter of minutes of arc rather than degrees, so get the figure from the bearing catalog for the specific type and series rather than assuming.
Past that limit, the contact patch between the rolling element and the raceway stops sitting in the middle of the raceway and runs toward the shoulder. When it reaches the edge, the contact ellipse is truncated and the stress at the truncation spikes far above the nominal contact stress the load calculation assumed. This is edge loading, and here is the part that matters: once you are edge loading, the rating life model no longer applies at all, because the condition it was derived under - the contact fully supported within the raceway - is violated. Any life multiplier you compute from the nominal load understates the damage, and it understates it by an amount nobody can put a number on from the outside.
That is why a bearing on a badly misaligned set can fail far earlier than even the cubed relationship predicts, and why the answer to "the calculation says it should have lasted three years" is sometimes that the calculation was never valid.
Read the ball path, because it separates the two mechanisms
Cut the failed bearing open, with the abrasive wheel guarded and a face shield over safety glasses and the grinding particulate captured at the source rather than left in your breathing zone, and the raceways carry a record of how the load was applied. Bag the failed bearing rather than binning it; the marks are the only physical evidence you get.
| What the raceway shows | What was happening |
|---|---|
| Even band, centered, uniform width, on the rotating ring's full circumference | Rotating radial load, no tilt; the bearing was loaded as designed |
| Band centered but wider and darker than expected | More radial load than design, no tilt; look at the reaction force, first mechanism |
| Band that wanders diagonally across the raceway, wider on one side of the ring and shifted toward the opposite shoulder on the other | Tilt between the rings; second mechanism, and edge loading if the band touches a shoulder |
| Band riding hard against one shoulder all the way around | Steady axial thrust or an axial preload, not misalignment |
| Two separate bands on the non-rotating ring, one on each side | Load zone position varying, commonly from a housing fit or a cyclic force reversing |
The diagonal wandering band is the misalignment signature and it is unmistakable once you have seen one. It tells you the tilt was real, which tells you the alignment work is not optional and that a self-aligning substitution would address something.
Worked example: back-calculating the load from a repeat failure
A 3,570 rpm pump set. The pump inboard bearing, a ball bearing, has been replaced three times: at 14 months, 16 months and 13 months of service. The bearing manufacturer's rating for this bearing at the design load and this speed works out to a basic rating life of about 5 years, which is 60 months.
Take the median observed life, 14 months, against 60 months: the life ratio is 0.233. Inverting, the load ratio cubed is 1 divided by 0.233, which is 4.29, so the load ratio is the cube root of 4.29, about 1.62. The bearing has been running at roughly 1.6 times the load the rating assumed.
Three conditions govern whether that 1.6 means anything, and they all have to be stated before anyone acts on it.
First, the observed median is not L10. Rating life is the life 90 percent of a population reaches, so the L10 of a population whose median is 14 months is shorter than 14 months. Using the median as if it were L10 therefore understates the load multiplier, which makes 1.6 a floor rather than an estimate.
Second, the model assumes an adequate lubricant film and a clean bearing. If these bearings were also running a marginal film or ingesting contamination, part of the shortfall belongs to those and the true load multiplier is smaller than 1.6. Check the grease condition and the seal before you take the number to a customer.
Third, and this is the one that settles it: if the teardown shows the diagonal band running to a shoulder, edge loading was present and the whole calculation is void, because it was derived for a contact fully inside the raceway. In that case the honest statement is not "1.6 times the load" but "the bearing was being loaded in a way the rating does not describe."
So the sequence is: compute the multiplier to know whether you are looking at a small problem or a large one, then cut the bearing open to find out which mechanism produced it, then measure the alignment. A shop that stops after the multiplier orders a heavier-duty bearing and buys itself a longer interval on the same defect.
The failure mode of getting this wrong is subtle and common: upgrading the bearing. A heavier series bearing has a higher dynamic capacity, so at the same 1.6 times load its computed life goes up and the shop believes the problem is solved. But a heavier series usually has a lower permissible tilt, not a higher one, so if the actual mechanism was the second one, the upgrade made the bearing less tolerant of the thing that was killing it. Interval improves for a while and then gets worse.
What genuinely changes the answer
Fitting a self-aligning bearing - a self-aligning ball or a spherical roller, whose outer raceway is spherical so the inner assembly can swivel - removes the tilt mechanism and leaves the load mechanism completely intact. The rings stop fighting each other; the reaction force from the coupling is still there in full, still cubed. That capability is specified for static misalignment between the rings from mounting and shaft deflection, not as a licence to run a coupling out of tolerance, because the coupling reaction does not care what kind of bearing is downstream of it.
Speed changes it too, in the direction people underestimate: rating life is in revolutions, so the same life in revolutions is half the calendar time at twice the speed. The same absolute misalignment on a 3,570 rpm machine and a 1,780 rpm machine costs about twice the calendar life on the faster one before any tolerance argument is made.
How to verify you have found the real cause
Do not close the job on the calculation. Confirm three things line up: the reverse-calculated load multiplier says the load was high, the raceway band says tilt or no tilt, and the measured alignment at the coupling agrees with what the raceway showed. When all three point the same way, you have a cause. When the raceway band is clean and centered and the alignment measures inside tolerance, the load is coming from somewhere else - belt tension, pipe strain, a hydraulic radial thrust on the impeller - and the alignment was never the answer.
References
- ISO 281, rolling bearing dynamic load ratings and rating life, for the basic rating life relationship and its load exponents; bearing manufacturer catalogs for permissible misalignment by type and series, and for the modified rating life factors covering lubrication and contamination.
- 29 CFR 1910.147 for mechanical isolation, stored energy and rotor blocking; 29 CFR 1910.333(b)(2) with NFPA 70E-2021, 120.5, for the motor circuit; 29 CFR 1910.215 for abrasive wheel work; 29 CFR 1910.134 for respiratory protection program requirements.
- See related: "How Bearings Fail and the Warning Signs"; "Why a Flexible Coupling Does Not Fix Misalignment"; "The Vibration Signature Reference".