Why a Bolted Joint Loosens

Why this matters

"It vibrated loose" is the default explanation and it is right maybe half the time. A joint can lose every bit of its clamp with the nut sitting exactly where you left it, and a joint can spin a nut off in a machine with almost no measurable vibration. Those are two different failures. Threadlocker, lock washers and nylon-insert nuts act on one of them and do nothing at all for the other, which is why a shop can apply a real fix, correctly, to the wrong route and watch the same joint come back.

Before working on a mount, bracket or splice: lock and tag the drive so the machine cannot start, which is 29 CFR 1910.147 in general industry for mechanical and stored energy, or 29 CFR 1910.333(b)(2) where the isolation is electrical, since 1910.147 excludes exposure to electrical hazards from work on conductors and equipment in electric utilization installations at (a)(1)(ii)(C). Support the load independently before slackening anything and stay out from under it. Let a hot assembly come to ambient before you put a wrench on it: re-torquing hot metal is a burn, and on a bolt whose preload is temperature-dependent it is also a measurement of the wrong joint.

The call: the bracket that came back for two different reasons

An aluminum equipment bracket bolted to a steel frame with four 3/8-16 Grade 5 bolts, on a machine that heats to around 250 F in service and sits at shop ambient overnight. It arrived slack twice.

The first visit found the nuts turned back visibly. The tech tightened them and added nylon-insert lock nuts, which is a correct fix for a nut that is rotating. It held for a while.

The second visit found the joint slack again, and this time the marks put on at the first repair were still lined up. Nothing had rotated. The same symptom, a completely different failure, and the first repair had no bearing on it.

Route one: preload leaves without rotation

The fastener does not move. The stack under it gets shorter, so the stretch in the bolt is released, and clamp force goes with it. Four sources, which overlap in time rather than arriving in a fixed sequence:

• Embedment. No surface is perfectly flat at the microscopic level, and under clamp load the high points of the thread flanks and the bearing faces flatten out. This is over quickly, mostly in the first hours and the first load cycles. On clean machined steel faces it is a modest percentage of preload; on rough, soft, or painted faces it is a lot more, because there is more to crush.
• Coating and paint creep. Anything soft in the grip continues to squeeze out under sustained load. A thick zinc coating or a paint film left under a bearing face is a slow-motion embedment that never quite finishes.
• Gasket creep and relaxation. A compressible gasket keeps deforming under sustained load, and does it faster hot. This is why gasketed joints get a re-torque and dry metal joints often do not.
• Differential thermal movement. A grip made of two different materials changes length at two different rates. This one both gives and takes, and it is the one that got the bracket.

Route two: the fastener rotates

Here the nut actually turns, and the mechanism is not the one most people name. Axial vibration, shaking along the bolt axis, is a poor loosener on its own. What reliably rotates a fastener is transverse slip: the clamped members moving sideways relative to each other, even by thousandths of an inch. Once the faces slip, the thread helix turns that sliding into rotation, and the rotation is cumulative.

That is why the fix that works is usually not on the fastener. Slip happens when the sideways load exceeds the friction the clamp generates, so the sequence runs: low preload, then slip, then rotation, then lower preload, then more slip. It is self-feeding. The sibling article on preload and friction owns the arithmetic of the slip threshold.

Read the direction at both ends before you trust it. Raise clamp force enough that the faying surface never slips and the rotational mechanism has nothing to act on, so a well-preloaded joint stays put with no locking device at all. Let clamp force fall toward the slip point and rotation begins, and it accelerates as it goes. There is no middle state where a joint slips a little and holds.

Telling the two apart in the field

Marking fasteners at assembly is what makes the top two rows readable at all, and it is the cheapest evidence in this whole subject.

The thermal arithmetic that explained the second visit

Aluminum expands roughly twice as much as steel for the same temperature rise. Common mean coefficients over a moderate range near room temperature run about 13 microinches per inch per degree F for common aluminum alloys and about 6.5 for plain carbon steel. Those are mean values for that range and for those material families; the specific alloy and the specific temperature span change them, and a materials table or the manufacturer owns the number for the metal actually in the joint.

Grip: 2.0 inches of aluminum bracket. Temperature rise from 70 F to 250 F, so 180 F.

• Aluminum growth: 2.0 x 13e-6 x 180 = 0.00468 inch
• Steel bolt growth over the same 2.0 inch effective length: 2.0 x 6.5e-6 x 180 = 0.00234 inch
• The aluminum grows 0.00234 inch more than the bolt does, and since the bolt is holding it, that difference is imposed on the joint as additional stretch

The extra load is that differential times the combined stiffness of bolt and members. A 3/8-16 bolt with a tensile stress area near 0.0775 square inch over a 2.0 inch grip, at a steel modulus near 29,000,000 psi, has a stiffness around 1,124,000 lbf per inch. The aluminum members in that grip are stiffer than the bolt; take them at roughly five times bolt stiffness, which gives a combined figure near 936,000 lbf per inch. That is an estimate for this geometry, not a constant.

• Added load hot: 0.00234 x 936,000, about 2,190 lbf

Now compare it to what was installed. Grade 5 proof strength is 85,000 psi, so proof load on that area is about 6,590 lbf, and a preload target near 75 percent of proof is about 4,940 lbf.

• Preload hot: 4,940 + 2,190 = 7,130 lbf
• Proof load: 6,590 lbf

The hot peak is above proof by roughly 540 lbf, about 8 percent over. The bolt is yielding every time the machine comes up to temperature. Yielding is a permanent stretch, so on the way back down to ambient the bolt is longer than it was and the preload comes back lower. Even a modest permanent set of half a thousandth of an inch costs about 0.0005 x 936,000, near 470 lbf, which is roughly 9.5 percent of the installed preload per cycle until the joint shakes down at a much lower clamp.

That is a joint that loosens on a schedule, with nothing rotating, in a machine that a vibration meter would call quiet.

Matching the fix to the route

For rotation:

• Raise preload so the faying surface stops slipping. This is the primary fix and it addresses the cause rather than the consequence.
• Take the slip out mechanically: a dowel pin, a shear key, a fitted bolt in a reamed hole, or a serrated interface. A locking device holds the nut; a dowel stops the joint moving in the first place, which is upstream.
• Locking hardware, chosen for the duty and the temperature. It resists rotation and that is the whole of what it does. The sibling article on washers covers which devices actually resist rotation and which only look like they do.

For preload loss without rotation:

• Get the soft material out of the grip. Paint and thick coatings under bearing faces are removable causes.
• Lengthen the grip. A longer bolt through a spacer stores more stretch for the same preload, so a given amount of embedment or creep costs a smaller fraction of it. This is the single most effective structural fix for a joint that keeps relaxing.
• Re-torque after the settling cycle, cold, and mark it. Embedment is largely a one-time payment, so one properly timed re-check often ends it.
• For the thermal case, take the peak out: a lower preload target will not help because you still yield at the same absolute load, so either raise the fastener's grade to move proof load above the hot peak, or lengthen the grip so the same differential produces less added load, or break the material mismatch. Threadlocker does nothing here whatsoever.

Checking you got this right

• Before proposing any fix, say which route you are treating and name the evidence. If the only evidence is that the joint was loose, you have a symptom, not a diagnosis.
• Where you have added a locking device, confirm the marks show rotation. If they do not, you have treated a failure that was not happening.
• On a mixed-material grip, work out whether the hot peak lands above the fastener's proof load before touching the joint again. If it does, tightening it to the same value is scheduling the next call.
• Look for fretting between the faces on every repeat loosener. Fretting is proof of slip and slip is proof that clamp force is below what the joint needs.
• Time the re-check to the mechanism: after one full heat cycle for embedment and settling, and after several for a joint you suspect is yielding at temperature. One cycle cannot distinguish those two.

References

• 29 CFR 1910.147, control of hazardous energy, for isolating and restraining before a mount is slackened
• 29 CFR 1910.333(b)(2), for electrical isolation, which 1910.147 excludes at (a)(1)(ii)(C)
• Fastener and equipment manufacturer documentation for the specified preload, re-torque interval, and any locking device qualified for the temperature
• Published materials data for the thermal expansion coefficient of the specific alloys in the grip, over the temperature range in service
• See related: Why Preload and Not Friction Holds a Joint Together; What a Washer Is For and When It Is Not; How to Tighten a Bolted Joint and Know It Is Right