Bolt preload

Why does a bolt stretched hard on assembly outlive a gently tightened one carrying exactly the same working load?

Two identical flanges, two identical bolts, the same fluctuating load pulling them apart a few times a second. One bolt was run up snug and nipped; the other was stretched on assembly until it carried a large tension before the machine was ever switched on. The gently tightened one breaks in weeks. The hard-stretched one runs for years.

The instinct is that this must be backwards: the hard-tightened bolt already carries a heavy load before the working load is added on top, so it should be closer to breaking, not further from it. What is wrong with that sum?

The sum assumes the working load simply adds to whatever the bolt already had. Ask whether that is true.

Picture the assembly before any external load. The bolt has been stretched, so it pulls the flanges together; the flanges, squeezed, push back with exactly the same force. Two elastic members, equal and opposite, in a closed loop. Now pull the flanges apart. The bolt gets slightly longer — but the flanges, released by that same small amount, spring back and their squeeze falls. The two are one movement.

So how does the external load divide? By stiffness. The extra bolt tension is the external load times the bolt's stiffness divided by the sum of bolt and member stiffnesses; the rest shows up as lost clamp. And a short fat column of steel flange is much stiffer than a long thin bolt. The fraction landing on the bolt — the joint stiffness factor — is commonly around 0.2 for a metal-to-metal joint, rising towards 1 as a soft gasket goes in. That figure is geometry-specific and illustrative, not a constant to look up.

Take a 10 kN cyclic load on such a joint. About 2 kN reaches the bolt; about 8 kN is subtracted from the clamp. The bolt sees a fifth of the swing it otherwise would, and it is the swing, not the mean, that fatigue answers to. Fatigue life climbs far faster than in proportion as alternating stress falls, so a fivefold cut in range can be the difference between a finite life and none ever being reached.

Now find the threshold, because this arrangement has a cliff rather than a slope. Clamp force falls as the external load rises; where does it reach zero? If the preload is 30 kN and 80 per cent of each newton of external load eats into the clamp, separation arrives at 30 divided by 0.8, which is 37.5 kN. Below that the bolt is protected. Above it the flanges part, the whole external load transfers to the bolt, and the stress range jumps fivefold in an instant.

That is the answer to the puzzle. Preload does not meaningfully add to the bolt's peak stress; it decides whether the bolt is inside the sheltered regime or outside it. The gently tightened bolt separates on every cycle and does the entire job alone.

A second, separate payoff in one sentence: self-loosening in vibration is driven overwhelmingly by tiny transverse slips at the mating faces, so clamp force sufficient to prevent slip prevents loosening — which is why the reliable answer to a bolt that keeps backing off is more preload, not a locking compound.

The load-bearing claim is that an external load on a preloaded joint is shared between raising bolt tension and relieving clamp force in proportion to the two stiffnesses, so while the joint stays closed the bolt's stress range is a small fraction of the applied range. It is directly falsifiable: bond a strain gauge to the bolt shank, tighten to a measured preload, and apply a slowly rising external load while recording strain. The account predicts a shallow line — bolt tension rising at roughly a fifth of the applied rate — then a sharp knee at the calculated separation load, beyond which it rises at the full rate. The refuting observation would be a bolt whose tension tracked the applied load one-for-one from the start, which is what the naive addition predicts, or a knee that did not move when the preload changed.

One misconception to correct. Preload is not "extra load the bolt must survive"; it is stored elastic energy that makes the joint behave as one body. It does raise the bolt's mean stress, a real cost that consumes static margin against proof strength. The conventional target of roughly 75 per cent of proof load, which I am recalling rather than deriving, is a compromise between that ceiling and the separation floor — a shop convention embedded in standards and torque tables, not an optimum anybody computes per joint.

That points at what actually binds. Not the physics of load sharing, which is uncontroversial, but knowing the preload achieved. Tightening is almost always done by torque, and most of the applied torque goes on friction under the head and in the threads; only a small part becomes tension. Friction varies with lubrication, plating, finish and reuse, so a torque-controlled bolt scatters widely around its target. The accepted failure follows: the design targets a preload comfortably above separation and below yield and simply tolerates the scatter — accepting that some bolts are far more highly stressed than intended, in exchange for a method a fitter can execute. Where that margin will not stretch, the answer is to measure the tension itself, by bolt elongation, turn-of-nut past snug, or hydraulic tensioning.

• Why a soft gasket raises the share of external load reaching the bolt, and how that changes the design.
• How turn-of-nut and bolt-elongation methods sidestep friction scatter, and why torque persists anyway.

Every term the collection defines is gathered in the glossary.