How to Tighten a Bolted Joint and Know It Is Right

Why this matters

Most shops can tighten a joint. Far fewer can say afterwards, with evidence, that it is right. The wrench reading is not that evidence, because it is a snapshot of resistance at the moment of assembly and it says nothing about what the joint did in the hours that followed. So this procedure is built the other way round: every step ends in something you can observe and write down, and the joint is proved by what those observations say across one settling cycle. If a step produces no observable, it is not finished.

Safety first, and it applies at every step below, not just at the start. A support, hanger or bracket is holding weight, so rig or block an independent support under the load before you slacken the first fastener, and stand clear of the piece rather than beside it. Where the joint clamps a pressurised or spring-loaded part, isolate and relieve to zero and confirm it at the joint, which is the stored-energy duty at 29 CFR 1910.147 in general industry; an electrical termination isolates under 29 CFR 1910.333(b)(2) instead, because 1910.147 carves out electric utilization installations at (a)(1)(ii)(C). Working off a ladder or a platform brings fall protection in at 29 CFR 1910.28 in general industry and 29 CFR 1926.501 on a construction site, and a field-service shop can fall under either depending on the job.

Step 1: Record what you found before a wrench touches it

Look at the fasteners and the faces and write down their state: thread condition, corrosion, whether the fasteners have been used before, whether the bearing faces carry paint or leftover compound, and whether the mating faces sit flat.

The observable: a written condition line. "Six 1/2-13 Grade 5 through-bolts, previously used, light surface corrosion on two, paint under all six heads."

Skipping this loses you the ability to interpret anything later. A joint that fails after a careful assembly is a puzzle only if nobody wrote down what went in.

Step 2: Fix the value and its lubrication state, from a source

Get the torque value from the equipment manufacturer, the fastener manufacturer, the engineered drawing, or the shop's own card for that size and grade, and note which of those you used. Published values state whether they are for dry or lubricated threads. Match your assembly to the state the value assumes.

The observable: the value, its source, and the lubrication state, written down together.

This is where the joint is most often lost before it starts. A dry value applied to lubricated threads over-tightens the fastener substantially, and the sibling article on torque and friction has the arithmetic. Where the value genuinely varies by application, the manufacturer or the drawing owns the answer and the shop card does not.

Step 3: Clean the interfaces and prove they are clean

Take paint, plating debris, burrs and old compound off the bearing faces under the head and nut and off the mating faces. Run a thumb over each one.

The observable: what came off, and that every bearing face is bare metal and flat to the touch.

Hazard in this step, and it is an inhalation route rather than a contact one. If you are scraping, wire-brushing or grinding coatings, the dust goes in the air. Coatings on older equipment can contain lead, controlled in general industry under 29 CFR 1910.1025, and abrading a substrate that contains crystalline silica falls under 29 CFR 1910.1053. Use wet methods or local exhaust and a respirator selected and fit-tested under a written program meeting 29 CFR 1910.134, not a nuisance dust mask. If thermal insulation has to come off a pipe to reach the joint and the system is old enough that the material is unknown, treat it as asbestos-containing until sampled and follow 29 CFR 1910.1001 in general industry or 29 CFR 1926.1101 on a construction site rather than pulling it dry by hand.

Step 4: Run every fastener home by hand

Every bolt goes in by fingers until it seats or reaches the nut's bearing face. No wrench.

The observable: each fastener turned freely by hand the whole way, or a named one that did not.

This is the cheapest test in the procedure and it catches the two faults that make everything downstream meaningless. A fastener that binds part way in is cross-threaded, damaged, or the wrong pitch, and a wrench will hide that under a rising reading. A fastener that runs free and then stops early has bottomed in its hole and will never clamp anything no matter what the wrench says.

Step 5: Snug pass, then measure the gap

Bring every fastener to light contact in a crossing pattern, then measure the gap at the joint at four points spread around it with a feeler gauge.

The observable: four gap readings.

A joint that will not close evenly at snug will not close evenly at full value. If the readings spread widely, stop and find out why: a warped face, a missing shim, a burr, a pipe or duct pulling the assembly out of line. Pulling misalignment out with bolt tension buries a permanent load in the joint and the fasteners pay for it.

Step 6: Rising passes in a crossing pattern

Work up in passes at roughly 30 percent, then 60 percent, then 100 percent of the value, crossing the pattern each time, then one final pass around the pattern at full value until nothing turns.

The observable: the pattern used, the pass percentages, and whether anything still moved on the final pass.

The sibling article on tightening sequence owns why the pattern matters. What matters for the record is the last pass: if fasteners are still turning on a full-value lap after three passes, the joint is still compressing and it needs another lap, not a note saying it is done.

Step 7: Mark every fastener

Draw a single line with a paint pen or marker across the nut, the washer and the member, so all three line up.

The observable: the mark itself, and a photo of it.

This is the step that makes the whole procedure verifiable, and it is the one that gets skipped. A torque wrench cannot tell you later whether a fastener rotated, because what it reads on a stationary fastener is breakaway resistance and that has no dependable relation to remaining clamp. A mark answers the question directly and costs seconds.

Step 8: Settle it, then read the marks

Put the joint through one full service cycle, meaning a full heat cycle where the joint runs hot, then come back and read the marks.

The observable: the rotation at each mark, in degrees, and the ambient state you read it in.

Read it cold. Do not re-torque a hot line that is still pressurised: the burn and the release are the same motion, and the thermal growth means you are reading a different joint. Shut down, isolate, relieve, let it come to ambient, then read.

The record, filled in: a bolted splice in a hot-pipe trapeze

A six-bolt splice in the steel trapeze carrying a heating water main, 1/2-13 Grade 5 through-bolts, dry, grip length about 3 inches through the steel stack.

• Step 1: six bolts previously used, two with light thread corrosion, paint under all six heads. All six replaced from one box.
• Step 2: value taken from the shop card's dry figure for 1/2-13 Grade 5, held at one value across all six. Dry state, no compound.
• Step 3: paint removed from six bearing faces, wet-scraped with the run isolated and cool.
• Step 4: all six ran home by hand.
• Step 5: gaps at four points read 0.012, 0.010, 0.011 and 0.010 inch at snug, a spread of 0.002 inch. Acceptable, joint pulling flat.
• Step 6: 30, 60, 100 percent in a crossing pattern; two fasteners still moved slightly on the first full lap, so a second full lap was run and nothing moved.
• Step 7: all six marked and photographed.
• Step 8, after one heat cycle: four marks unmoved, two showed about 5 degrees of backoff.

Five degrees looks like nothing. Put a number on it. A 1/2-13 thread advances 1/13 inch, about 0.0769 inch, per full turn, so 5 degrees is 5/360 of that, about 0.0011 inch of lost stretch. Over a 3 inch grip, a 1/2-13 bolt with a tensile stress area of about 0.1419 square inch and a steel modulus near 29,000,000 psi has a bolt stiffness near 1,372,000 lbf per inch. Treating the bolt as the only elastic member, which is an upper bound because the clamped members are three to five times stiffer and share the strain, the lost preload is at most 0.0011 x 1,372,000, about 1,470 lbf, and realistically nearer 1,200. The preload target is 75 percent of proof load: Grade 5 and its 85,000 psi proof strength are SAE J429 values that reach you through the drawing or the purchase specification and are verified by the head marking rather than by the grade being written on a shop card, the metric equivalents live in ISO 898-1 and are not interchangeable, and 75 percent of proof is a common shop practice for a reusable elastic joint rather than a universal target, so where the equipment manufacturer states one, that governs. At 0.1419 square inch and 85,000 psi, proof load is 12,060 lbf and 75 percent of it is around 9,000. Those two bolts lost somewhere between 13 and 16 percent of their clamp, and the upper bound is the one to act on.

That number belongs to that geometry and moves with it. Halve the grip to 1.5 inches and the bolt is twice as stiff, so the same 5 degrees costs roughly twice as much, around a third of the preload. A short-grip joint is far less forgiving of small rotation than a long-grip one, which is the practical reason a longer bolt through a spacer is a real fix on a joint that keeps relaxing.

The close-out: the two bolts were re-torqued to full value cold, re-marked, and read again after a second heat cycle. Both read zero rotation. Four unmoved, two moved and were corrected, and the second cycle confirmed the correction. That is the sentence that goes on the ticket, and it is an observation, not a claim.

What the failure mode would have looked like without step 7. Same joint, no marks. The two relaxed bolts sit at roughly 84 percent of clamp through a winter of thermal cycling. Each cycle the splice slips a little at the faying surface, the faces fret, grip length erodes, preload drops further, and the callback arrives in the spring as a sagging run with two elongated holes. Nothing in the file says anything except "torqued to spec," which was true on the day.

What changes this procedure

• A gasketed joint adds a creep term that steel-to-steel does not have, so it needs its re-check timed to the gasket manufacturer's guidance rather than to one heat cycle, and it may need more than one.
• A joint you cannot revisit cannot use step 8 at all, and the settling has to be bought before close-up. The sibling HowTo on unre-checkable joints owns that case.
• A joint with a specified angle or stretch method replaces steps 6 and 8 entirely with the manufacturer's snug value and angle. Do not blend the two methods.

References

• 29 CFR 1910.147, control of hazardous energy, for stored energy in a loaded or pressurised joint
• 29 CFR 1910.333(b)(2), for isolating an electrical termination, which 1910.147 excludes at (a)(1)(ii)(C)
• 29 CFR 1910.28 general industry and 29 CFR 1926.501 construction, for fall protection where the joint is worked at height
• 29 CFR 1910.1025 lead, 29 CFR 1910.1053 respirable crystalline silica, 29 CFR 1910.1001 asbestos general industry and 29 CFR 1926.1101 construction, for the airborne hazards created by cleaning faces and disturbing insulation; respirator program under 29 CFR 1910.134
• See related: The Torque Sequence That Matters; How to Torque a Joint You Cannot Re-Check Later; What Torque Actually Measures and What It Does Not