Threaded inserts and compression limiters

A heat-set insert is a knurled brass sleeve melted into a hole in a printed part, so a machine screw has metal threads to bite. A compression limiter is a metal tube through a clamped plastic part, so the screw's clamp goes through metal and not through plastic that creeps. cadagent.mechanical.inserts designs the hole and the boss for an insert, works out where the screw's clamp goes, holds it against the insert's pull-out, torque-out and thread, the plastic under the head and the limiter, and says how much of the clamp the joint keeps. cad-agent inserts runs it on every joint in a file. The strengths are a shear model calibrated to published pull-out tests, so a result is a starting point to test, not a rating. The coupon plan at the end says how to measure it.

Check a joint

Declare each joint in a file and run the check. A joint with an insert is a screw into a heat-set insert. A joint with a thread and no insert is a screw into metal or a nut that clamps printed plastic.

# joints.yaml
defaults: {material: pla-bambu-basic}
joints:
  - {name: lid, insert: mcmaster-94459A140, boss_outside_mm: 9.5, head_mm: 7.0, clearance_mm: 3.4,
     clamped: {thickness_mm: 4.0}}
  - {name: bracket, thread: M4, thickness_mm: 6.0, head_mm: 7.0, clearance_mm: 4.5}

$ cad-agent inserts joints.yaml --output result.json
WARN  lid: M3 94459A140 in Bambu PLA Basic, through-insert; tighten to at most 0.20 N m (clamped on insert)
      warn: preload kept: the clamped plastic relaxes and takes the preload with it: ...
WARN  bracket: M4 clamping Bambu PLA Basic; tighten to at most 0.68 N m (head bearing)
      warn: preload kept: the clamped plastic relaxes and takes the preload with it: ...

The command exits 1 when a joint fails and 2 when the file is wrong. A warning does not fail it. The JSON holds every row with its value, limit, verdict and reason, the insert's capacities, the torque ceilings and the limits.

From Python:

from cadagent.mechanical import inserts

inserts.hole_design(inserts.insert("mcmaster-94459A140"))      # 4.2 mm in CAD (the maker's 4.0 + 0.2), 6.7 mm deep
r = inserts.check_joint("mcmaster-94459A140", "pla-bambu-basic", boss_outside_mm=9.5, head_mm=7.0,
                        clearance_mm=3.4, clamped={"thickness_mm": 4.0})
r["verdict"], r["torque"]["allowed_nm"], r["torque"]["governed_by"]   # 'warn', 0.20, 'clamped_on_insert'
r["capacity"]["pull_out"]                                         # low 811, mid 1152, high 1394 N
inserts.check_clamp("M10", "pa6-cf20-polymaker-wet", 40.0, head_mm=20.0, clearance_mm=12.0, torque_nm=20.0,
                    limiter={"outside_mm": 18.0, "bore_mm": 12.0}, limiter_length_mm=39.8)
inserts.coupon_plan()                                             # what to print and measure

Where the clamp goes

The insert's height against the joint face decides what carries the screw's clamp. This is the first thing the check works out, and it changes every limit after it.

• Flush, or up to 0.13 mm proud (PEM, SPIROL). The clamped part bears on the insert's end. The clamp goes through the brass, and the knurl carries only what pulls the parts apart. M3 in PLA Basic, clamping metal: tighten to at most 0.40 N·m, set by the insert's end face: the ring round the clamped part's hole takes nearly all the clamp, held to brass's 124 MPa. A printed clamped part sits on that same ring, so its plastic sets a lower limit (the lid above: 0.20 N·m). A washer or a limiter spreads it.
• Recessed, even by 0.05 mm. The screw pulls the insert towards the head until the gap closes. The knurl carries the whole preload, so the same joint 0.2 mm recessed may take only 0.13 N·m.
• A clamped part whose hole is as wide as the insert bears on the plastic round it. That jacks the insert out too, and the check says so.
• Pressed in from the far side, the insert is pulled into the boss in front of it, and that shoulder must not punch out.
• The screw's tip. Past the floor of a blind hole the screw bottoms and jacks the insert out. Past the insert's end it reaches the room left for the melt, which may have filled it. Give screw_depth_mm and the check reports both, and the thread length engaged.

Compression limiters

A limiter only helps if the head lands on it. If it is longer than the plastic, the part is loose round it. If it is shorter, the head must squeeze the plastic by the difference before it lands.

• Some printed plastics take less squeeze than SPIROL allows. SPIROL allows 3 to 5 % compression. A printed plastic reaches its compressive strength at 2.6 % (PETG Basic) to 3.6 % (PLA Basic), so the check warns past that.
• The seating force. SPIROL's formula 3 takes the plastic under the head as a column. A narrow ring of head on thick plastic spreads the load, and the far face sits on a broad face, so the check also takes Shigley's frustum through the whole thickness and uses the higher. An 18 mm steel sleeve 0.2 mm short of 40 ± 0.1 mm of wet PA6-CF, under a 20 mm washer, needs 550 N as a column and 5055 N as a cone. The ring then bears 85 MPa, so it warns.
• A limiter on a straight-knurl insert. The limiter must abut the insert. Its end then bears on the thin ring between its bore and the knurl, 10.00 mm² for M5. Brass's yield there caps the preload at 1240 N. Squeezing 6.1 mm of printed PETG by 4.9 % takes 3170 to 9146 N, so the limiter never seats and the check fails it. Make the limiter as long as the measured plastic, or use one with a proud knurl (Bulten), or a flanged insert.
• Limiters of your own. check_clamp takes a sleeve, a bushing or a spacer as limiter (its outside, bore, metal and, for the wall, its yield). Catalogue limiters are SPIROL's CL6000 (aluminium) and CL8000 (brass), M3 to M8, rated for 10.9 screws.

flowchart TD
    I["insert, hole, boss,<br/>material, axis"] --> P{"where does<br/>the clamp go?"}
    P -->|"flush or proud"| T["through the brass"]
    P -->|"recessed, wide hole<br/>or the tip bottoms"| J["jack-out: the knurl<br/>takes the preload"]
    P -->|"far side"| S["the shoulder<br/>takes it"]
    T --> C["torque ceilings: screw (75 % of proof),<br/>torque-out, insert thread, head,<br/>shoulder, limiter"]
    J --> C
    S --> C
    C --> L{"limiter?"}
    L -->|"no"| R["plastic in the clamp:<br/>it relaxes (warn)"]
    L -->|"yes"| K["length, seating force,<br/>wall, press fit, boss"]
    K --> V["pass, warn or fail<br/>with the most to tighten to"]
    R --> V
    classDef here fill:#fffdf5,stroke:#173d3c,color:#173d3c
    classDef out fill:#ffe1cc,stroke:#bd3a12,color:#173d3c
    class J out

What the numbers rest on

The pull-out force is the knurl's cylinder sheared, F = c τ π D L. τ is half the filament's tensile strength across the layers, from the maker's data sheet, and c is fitted to published tests.

• PLA: c from 0.62 to 1.06, from 13 results in four publications by three groups (László 2026, CNC Kitchen 2019 and 2026, SPIROL 2026). Medium confidence. SPIROL's two M6 results set the low end.
• PETG: c from 0.51 to 0.73, from three results: CNC Kitchen's M3 and SPIROL's two M6. Low confidence.
• SPIROL's M6 values are read off its bar charts (its 3D-printing paper, Figures 10 and 12), to about 20 N.
• Other families (ASA, nylon) have no printed test with a data sheet in the table. They take the lowest c of any family, and the check warns.
• Torque-out comes from pull-out through r = 2 T / (F D). The tests give r from 0.66 (SPIROL, printed glass-filled PA6) to 1.41 (PEM, M8 in ABS). CNC Kitchen's printed PETG gives 1.12. The low end is taken.
• Other sizes scale by the knurl's area. SPIROL's printed M6 results sit below the M3 ones, so area alone would overstate a larger insert. The low end is the M6's.

Worked example: a robot's inserts

Gyropup has 151 heat-set inserts and many clamped plastic joints, listed in its insert-joint audit (its PR #138, as merged). The checks ran on 41 of its 46 joint families and on 43 of the 47 printed clamps. Five joints fail:

• two screws reach 2.0 and 3.0 mm past the floor of a blind pilot, so they jack their inserts out;
• a 4 mm pilot is no deeper than its 5.7 mm insert;
• an M6 at 8 N·m puts 156 MPa on wet PA6-CF under its 10 mm head, against the 27 MPa allowed (114 MPa with an 11 mm washer), and an M4 at 2.3 N·m puts 160 MPa on PETG under a socket head (6.53 mm bearing face) if what it clamps is printed.

The audit found the same problems on its own. The printed clamps pass at the torque the check gives them (0.23 to 1.24 N·m) and warn only because their plastic relaxes.

Test coupons

The plan prints each insert upright and lying, pulls it out, turns it out, and clamps it with and without a limiter. The limiter stands proud of its plate, so that cell measures an all-metal clamp against the plastic one. python3 tools/insert_coupons.py --out exports/insert-coupons writes each coupon's STEP and STL, in its print orientation, and the plan as JSON.

Each cell is five specimens with the predicted value next to it, so a test can confirm or move the calibration. Record the filament, the printer settings, the hole as printed, the iron, the insert's height afterwards, the room and the failure mode. Pull at 2 mm/min with a reaction plate that clears the boss. Read a clamp's preload at 0, 1 h, 24 h and 168 h, then after 24 h at 50 °C.

Limits

• Not a rating. The calibration is M3 × 5.7 and M6 × 12.7 in PLA and PETG. Other sizes scale by the knurl's area, from the M6's lower fit, and other families take the lowest fit.
• Torque-out comes from pull-out through the tests' ratio. Push-out and an insert lying in the layer plane have no published number, so they are flagged, not modelled.
• Static, short-term, room temperature. A held load creeps. PLA inserts kept 26 to 55 % of their pull-out after a year outdoors (Materials 2025).
• Bosses as tested. SPIROL printed 5 wall loops at 55 % infill, László a 1.3 to 1.9 mm solid sleeve in 40 % infill. Fewer walls hold less: one loop at 15 % infill held about a quarter of eleven (SPIROL, Figure 8).
• Preload kept. Printed PLA was measured for an hour and printed joints for a week. The long-term figure is moulded polycarbonate's, and PLA, nearer its glass transition, will keep less.
• The seating force takes the plastic as an elastic column or cone, with no edge or free face nearer than the cone.
• CAD and closed forms do not prove a joint holds. Print the coupons and measure.

Sources and survey

Tests that keep it honest

tests/test_inserts_formulas.py checks each relation against a hand evaluation or a closed form it must reach. tests/test_mechanical_inserts.py reproduces the published tests from the calibration, checks the makers' hole and boss rules, each row's verdicts on built cases, the CLI's exit codes, and the coupons' CAD volumes. tests/test_insert_figures.py redraws every figure on this page from the module and fails if one has gone stale.