Why Shaft Deflection Matters More Than Shaft Strength

Why this matters

Ask a shop why a shaft is the size it is and most will say strength. Almost no service shaft is sized by strength. It is sized by how much it is allowed to bend, because the bearings, seals and drive components hanging on it stop working at deflections hundreds of times smaller than anything that threatens the steel. Once you know which criterion governs, two very common decisions sort themselves out immediately: moving a sheave out an inch is a bigger change than it looks, and upgrading to a stronger steel to cure a flexing shaft does literally nothing.

Before you set an indicator

Isolate and lock out under 29 CFR 1910.147 before setting up any measurement, and confirm zero rotation rather than assuming the contactor drop stopped the wheel. Restrain a spring-loaded motor base before belts come off, since the tensioned base is stored energy under the same standard. Turn the shaft by hand from the coupling hub or the shaft itself with your hands out of the sheave nip, and without gloves, because a glove near anything that turns is an entanglement hazard rather than protection. Never hold a dial indicator against a running shaft: if you need a running reading, fit a non-contact displacement probe while the machine is locked out, then read it from outside a refitted guard, which is required by 29 CFR 1910.219 in general industry and 29 CFR 1926.300(b) on a construction site. If you pull a coupling to free a shaft, support anything the coupling was holding up before you separate it, and rig a heavy shaft rather than manhandling it.

The gate

Compute the deflection and the slope at the element that actually cares, compare each against that element's own tolerance, and check stress last.

Two cases below run through that single gate and come out opposite ways.

The geometry for both: a 1.500 in diameter shaft, bearing span 6.0 in, sheave overhung 2.0 in beyond the near bearing, belt pull 150 lbf.

The section property that resists bending is the second moment of area, I = pi d to the fourth over 64. For 1.500 in that is 0.2485 in to the fourth. The elastic modulus E for carbon and alloy steel is about 29 x 10 to the sixth psi. So 3EI is 21.6 x 10 to the sixth.

Two numbers come out of the geometry, and they scale differently, which is the part people get wrong.

Deflection at the overhung end is F a cubed over 3EI:

150 x 2.0 cubed / 21.6e6 = 0.000055 in

Slope at the near bearing. The overhung load reaches the bearing as a moment M = F x a, which is 300 lb-in, and the slope a simply supported span takes at the support where an end moment is applied is M L over 3EI:

300 x 6.0 / 21.6e6 = 0.0000833 radians = 0.29 arc-minutes

That model treats the supports as rigid and the shaft as uniform, so it ignores the bearing's own radial stiffness and any housing compliance. Both add to the real figure, so treat it as a lower bound rather than an answer, and use it for comparing before and after rather than for certifying an absolute.

Deflection scales with the cube of the overhang. Slope scales with the first power of it, because slope is driven by the moment and the moment is linear in a. Do not carry one exponent to the other quantity.

What each element will actually tolerate

Read the last row against the others. The steel's limit is four to five orders of magnitude away from the limits that everything attached to it lives by. That is the whole argument.

Outcome one: the sheave moved out one inch

A shop fits a larger sheave and needs a spacer to clear the guard, taking the overhang from 2.0 in to 3.0 in. Belt pull unchanged at 150 lbf.

Bearing slope. Moment becomes 150 x 3.0 = 450 lb-in.

450 x 6.0 / 21.6e6 = 0.000125 radians = 0.43 arc-minutes

Up by half, which is exactly the ratio of the overhangs. Against a deep-groove ball bearing's few-arc-minute allowance, this passes the gate. The bearing is not the finding here, and a tech who checked only the bearing would close the job.

Overhung deflection. 150 x 3.0 cubed / 21.6e6 = 0.00019 in, which is 3.375 times the original, or 1.5 cubed. The number itself is small. What matters is what it is: a deflection fixed in space, because belt pull points one way, with the shaft rotating inside it. Every revolution is a full reversal of bending stress at the shaft surface.

Bending stress at the near bearing. sigma = M c over I, with c = 0.75 in:

• At 2.0 in overhang: 300 x 0.75 / 0.2485 = 905 psi
• At 3.0 in overhang: 450 x 0.75 / 0.2485 = 1,358 psi

Both are trivially small against the steel, and both are fully reversed at shaft speed, which is where they matter. Fatigue at a stress raiser under reversed bending is the mechanism that actually breaks service shafts, and a sibling article works that case through.

Verdict on case one: passes the bearing check, and that is not the same as passing. The real consequences of the spacer are a 50 percent increase in reversed bending stress at every stress raiser on the shaft and a sheave face that now moves further under load. If there is a keyway or a shoulder near that bearing, the change landed on it.

Outcome two: the stronger steel

Same machine, different proposal. A shaft is flexing enough to be a problem, so the shop specifies a heat-treated alloy steel with roughly double the yield strength of the plain carbon original.

Run the gate. Deflection is F a cubed over 3EI. The change affects yield strength, which does not appear in the formula. E is essentially the same for every carbon and alloy steel, about 29 to 30 x 10 to the sixth psi, and heat treatment does not change it.

The deflection after the upgrade is identical to the deflection before it, to the digit. So is the slope at the bearing, the runout at the seal, and the contact pattern at any gear. Every number the gate cares about is unchanged.

Verdict on case two: fails the gate completely. The change bought only static strength, in a shaft running at a fraction of a percent of its shear capacity, on a machine whose problem was stiffness. Where the upgrade does earn something is fatigue strength, which does scale with tensile strength, but a stronger steel is also more notch sensitive, so at the sharp corner where the crack would actually start it collects less of that benefit than the strength ratio suggests.

One boundary worth stating so this does not become an over-general rule: the claim is that E does not vary meaningfully across steel grades, not that material never matters. Austenitic stainless runs a little lower than carbon steel, and aluminium is roughly a third of steel, so an aluminium shaft in the same geometry deflects about three times as much. Across the steels a service shop actually chooses between, the modulus is a constant.

The lever that does work

Geometry. Deflection depends on I, and I goes as the fourth power of diameter.

Go from 1.500 in to 1.625 in, an 8.3 percent increase in diameter. I goes from 0.2485 to 0.3423 in to the fourth, a factor of 1.377, so deflection falls to 1 divided by 1.377, which is 73 percent of what it was. An eighth of an inch on the diameter did what doubling the yield strength could not.

Overhang is the other lever, and it is usually cheaper. Deflection at the overhung end goes as a cubed, so pulling a sheave in from 3.0 in to 2.0 in is worth a factor of 3.375. That is why a bushing fitted with its flange the other way round, or a sheave pushed out to clear a guard, is a real reliability decision and not a fitting detail.

Span is the third. For a load between the bearings, deflection goes as span cubed, so a bearing moved out to make room is expensive in the same way.

The criterion that is not stress either: critical speed

A shaft with mass on it has a bending natural frequency. Run at a speed that coincides with it and deflection grows for the same force. Conventional practice keeps continuous operating speed well below the first critical, or well above it on a machine designed to pass through quickly; the separation margin is a design decision, so take it from the machine's own documentation rather than from a rule of thumb.

Critical speed goes as the square root of stiffness over mass, and stiffness goes as I, so the same fourth-power dependence appears here. The 8.3 percent diameter increase above raises stiffness by 37.7 percent and therefore the square root of that, 17 percent, in critical speed - but only where the vibrating mass is dominated by something attached to the shaft, like a fan wheel or a coupling. On a long shaft whose own mass dominates, the diameter increase raises mass as the square of diameter at the same time, and the two effects largely cancel. State which case you are in before you claim a gain.

Diagnosing a machine that is already resonating is a different job from sizing a shaft to avoid it, and the library has articles on reading a vibration signature and on separating a resonance from a genuine mechanical fault. This section is only the sizing criterion.

Where stress does govern

It would be a bad article that left you thinking stress never matters. It governs in specific, recognisable cases:

• Short, stubby shafts under high torque, where there is not enough length for deflection to accumulate. A stub shaft in a gearbox is a stress problem.
• Shock and reversing duty, where a peak torque several times the rated value arrives suddenly. Rated-duty arithmetic misses it entirely.
• Anywhere a stress raiser has been cut into the surface, because the local stress is the nominal figure times a concentration factor, and that is where the shaft actually fails.
• Thin-wall hollow shafts, where local buckling and wall stress become real limits that a solid-shaft calculation never sees.
• Corroded or fretted surfaces, where the endurance limit has been cut by the surface condition rather than by anything in the geometry.

The pattern in that list is that four of the five are fatigue cases at the surface, not static overload. Which is the same conclusion from the other direction: size for stiffness, then go and look at the surface.

References

• Trade-standard mechanical design references for beam deflection and slope relationships, second moment of area, and the elastic modulus of steels, which is not altered by heat treatment
• Bearing manufacturer catalogue for allowable misalignment by bearing type, which depends on internal clearance and load; seal manufacturer documentation for permissible dynamic runout against shaft speed
• 29 CFR 1910.147 (hazardous energy and stored energy in a tensioned drive); 29 CFR 1910.219 and 29 CFR 1926.300(b) (power-transmission guarding, general industry and construction)
• See related: How a Shaft Actually Carries Torque; What a Keyway Does and How It Fails; The Difference Between a Resonance and a Genuine Mechanical Fault