PETG bolts come loose. The part still looks fine.

The bracket was three printed pieces, Hatchbox PETG, held with three bolts in an engine bay. The first version was 20% infill and four walls. The bolts went slack in a few days. The reprint went to 60% infill, still four walls, and this time the nuts were nylocs with blue threadlocker. Over the next few weeks those bolts went slack too.

What makes the post useful is what did not happen. The washers had not buried themselves. The part was not bent or collapsed. Anyone glancing at it would have called the assembly fine and walked off. The only evidence was that a wrench could turn a fastener that had been tight.

That is an anecdote from one car, one filament brand, and one hot bay. It is not a measurement. It is the failure people recognise, and it rules out the explanation most of us reach for first: a crushed seat you can see with the naked eye.

In 2021 Catalin Gheorghe Amza and colleagues at Politehnica University of Bucharest published a creep test that is almost the same joint. They printed PLA and PETG coupons on an Ender 3. Each coupon had five 6 mm bores, and an M6 bolt and nut went through washers on both sides. The washers were 12 mm outside and 7.5 mm inside, so the load was already spread. Each bolt was tightened to 5.07 N·m. They left them for seven days, then measured the torque required to undo them.

The PLA coupons, Fillamentum Extrafill, came undone at 4.92 N·m on average, a 3.3% loss, give or take 0.04. The PETG coupons, PrimaSelect, came undone at 4.49 N·m, an 11.5% loss, give or take 0.07. They treated that untightening torque as the clamp that was left. Table 1 lists a 0.4 mm nozzle, 0.2 mm layers, two contours, and 100% infill. The discussion ties the bolt result to the 45°/−45° raster used to print the coupons. This was not a spongy 20% bracket. A solid coupon still gave the torque back.

A second set of coupons sat through 24 hours of UV-B first. The lamps ran in eight-hour dry stretches at 50°C, then four hours off, three cycles, which the authors treat as a rough stand-in for several months of outdoor ultraviolet. After that, the same week under the bolt: PLA barely moved from its unaged result. PETG came undone at 4.18 N·m on average, give or take 0.06, which the paper calls a further 7.5% below the unaged PETG. The discussion later says the sun-aged PETG loss reaches 21.3% against the original 5.07 N·m. That is a larger drop than the average of 4.18, so the number to repeat is the average, with the paper open beside it.

Four coupons carried the whole bolt test: one unaged and one sun-aged piece in each material, five bolts in each piece. The ± figures are the spread across those bolts. It is a small test. The authors already limit it to the filament, the machine, and the 45°/−45° raster they used, and they note that another brand can move the number. That caution is part of the result.

Creep, in the sense that matters at the bench, is slow plastic deformation under a load the part can already bear. The chains in the polymer slip. The coupon gets a little shorter between the washers. The bolt's grip is the stretch in that stack. Shorten the stack and the grip leaves, even though the nut has not turned a degree.

A nyloc nut and a drop of threadlocker are there to stop the nut turning. They did that job in the engine bay, and the bolts still went slack, which is what you would expect if the nut was never the part that moved. More infill makes a boss harder to crush outright. The Bucharest coupons were already solid, washers and all, and a week still took 11.5% of the PETG torque. Density was not the missing setting.

Most of the movement shows up early. In the same paper's tension test, loaded at about a quarter of ultimate strength, both PLA and PETG did the bulk of their elongating in the first hours, then crept more slowly out to the end of the week. A joint that passes a wiggle test at the end of the build can be the loose joint on Thursday. Nothing about the part announces the change.

A 2025 paper from the University of Latvia and Luleå, by Leons Stankevics and colleagues, pulled printed PETG coupons under a constant stress and watched them. This is a tensile test, not a washer crushing a boss, so it does not supply a torque. It supplies the clock, and the print direction.

Below half the coupon's strength, PETG printed with the filament running along the load picked up as much as an extra 1% strain in five hours. The crosswise coupons picked up 0.6%. At half strength, five hours of creep was about a twentieth of the elastic stretch, and that extra had doubled by twenty hours. Push the stress to 70% of strength and the along-filament coupons took on far more creep than the crosswise ones: a quarter of the elastic stretch in five hours, against 7% the other way.

The practical reading is narrow and worth keeping. A printed part under a steady load starts moving immediately, and turning the part on the bed changes how much. It does not say that any particular bracket will be loose by Friday. It says a manual that ignores print orientation is leaving out a variable the measurement can already see.

The clamp survives when the bolt tightens metal against metal. A sleeve, a spacer, or a shoulder inside the printed hole sets the length of the stack. The plastic can creep a little and the nut is still tight, because the distance that matters is now steel. The printed part is located. It is not the spring.

A heat-set insert is a metal thread. It is not a metal stack. The plastic under the screw head is still what the bolt is squeezing, and that plastic is the piece Amza's washers were bearing on. Size the hole from the tolerances note. Then decide, separately, whether this joint is allowed to relax.

If the design really does clamp plastic to plastic, write the maintenance where the builder will see it: check these screws after a day, and again after a week. That is the inspection the seven-day torque implies. Swapping the engine-bay bracket to PLA because PLA only lost 3.3% in a room-temperature week is a bad trade. The Fillamentum PLA in that paper is listed with a glass transition of 55°C. The PrimaSelect PETG is listed at 86°C. The bolt test itself was not run in an oven. An engine bay is hotter than the lab bench the 11.5% came from, and the sun-aged coupons are the warning that PETG gives up more clamp after heat and ultraviolet, not an invitation to print the same clip in a material that softens on a warm dashboard.