A 3D printed part that broke is not a dead object. It is a witness statement. Somewhere in that part is a fracture surface, an orientation, a colour, a location, and a time-to-failure, and those five things together will tell you more about what went wrong than any amount of staring at your slicer settings.
The problem is that most of us were taught to read metals, and plastics do not break like metals. Steel gives warning: it yields, bends, whitens, and then tears. A printed part usually does none of that. It behaves perfectly for four months and then splits down a layer line on a Sunday afternoon, two centimetres from where you expected it to fail if you had thought about it at all. That gap, between where it broke and where you assumed it was strong, is the actual subject of this article.
This is a field guide, not a textbook. The goal is that the next time something snaps in your hands, you spend twenty minutes learning why instead of five seconds feeling annoyed and reprinting the same geometry that is going to snap again.
Why printed parts fail in ways that look wrong
Three structural facts about FDM parts make their failures look counterintuitive to anyone who is used to isotropic materials.
First, the part is not one material, it is two. Extruded roads bond to each other laterally along a layer, and layers bond to each other vertically. Those two interfaces have different strengths, often by a factor of three or more. When you decide that a part is "PETG" you have decided almost nothing about how it will fail, because the road-to-road and layer-to-layer bonds are not PETG at all. They are the welded remains of two cooling polymer surfaces.
Second, the geometry you designed is not the geometry that cracked. Corners are not sharp, holes are not round, and thin walls are thicker at the bottom than at the top because of how the first layer squashes. Stress concentrates at exactly the features a slicer rounds off for you, so the fracture path often follows a curve you never drew.
Third, time is a variable. Plastics creep, relax, absorb water, and fatigue. A part that survives a hundred aggressive load cycles can die quietly after ten thousand gentle ones. This is why "it held fine for months" is not evidence of adequate design; it is evidence that a slow mechanism had not finished yet. If you want the longer version of that argument, it is covered in creep, fatigue and preload.
The evidence you already have
Before you touch a tool, write down five things. Not because the list is magic, but because the act of writing them down stops you from skipping to a conclusion.
| Evidence | What it rules in | What it rules out |
|---|---|---|
| Where it broke | Design cause: a concentration, a thin section, an unsupported span | Material lot problems, which fail everywhere at once |
| How long it lasted | Seconds = load; days = geometry; months = creep or chemistry | A single mechanism, if the number is inconsistent |
| The face (texture, colour, gloss) | Whether the crack followed layers, roads, or crossed both | Assembly damage, if the face is fresh and unoxidised |
| Orientation at failure | Whether the load crossed the layer planes | Load direction guesses, if you can read the part |
| Environment | Heat, water, solvent, UV, vibration exposure | Pure mechanical explanations |
Two of these are usually available for free. The location is in your hand. The time is in your head or in a message you sent someone complaining about it. The other three require you to look properly, which brings us to the part most people skip.
Reading the break face
A fracture surface is a map. It has a starting point, a direction, and a texture that records the speed at which it moved. Learn four textures and you can classify most printed failures in a couple of minutes.
Five failure modes and what they leave behind
| Mode | What the face looks like | Typical time to failure | The honest fix |
|---|---|---|---|
| Layer separation | Flat, matte, dusty; individual layer edges clearly visible like a stack of sheets that came apart | Seconds to days under load | Change orientation, raise hotend temperature, reduce part cooling, or accept the limit and add material across the joint |
| Road separation / voiding | Long grooves running along the direction of travel; a comb-like appearance with gaps between lines | Hours to weeks | Increase flow slightly, check extrusion consistency, slow down, verify the nozzle is not partially clogged |
| Brittle snap | Glossy, smooth, sometimes with a radiating fan from the origin; little or no whitening | Instant, at peak load | Wrong material for the temperature or the impact; a tougher grade or a warmer environment |
| Ductile tear | Stretched, whitened, drawn out; edges pulled thin; visible necking | Instant, after visible deformation | Usually this is a good failure; the part was overloaded but not misdesigned. Increase section or reduce load |
| Progressive crack | Two zones: a smooth slow-growth region near the origin, then a rougher fast-fracture region | Weeks to months | Find the origin and remove the concentration; the part was carrying a crack, not a load |
The single most useful distinction in that table is between the first two rows and the last three. Layer and road separation are process failures: the design was probably fine and the printer under-delivered. Brittle snap, ductile tear and progressive crack are design failures: the printer did what it was told and the part still was not strong enough. Getting this classification right saves you from the two classic wastes of time, namely re-slicing a part that was simply too thin, and thickening a part whose layers never bonded in the first place.
The four-question triage
Once you have the face classified, run four questions in order. Each one, if answered yes, sends you down a different path. Most people run them in the wrong order and end up tuning the slicer to fix a design problem, which is roughly as effective as adjusting your car mirrors to fix a flat tyre.
- Did it break along a layer plane? If the fracture surface is flat and follows a horizontal line in the print orientation, the interlayer bond was the limiting factor. Nothing you do to the XY geometry will help until that is addressed.
- Did it break at a feature? A hole, a fillet, a boss, a sharp internal corner, the end of a rib. Concentrations multiply nominal stress by two to four times, and printed parts have concentrations you did not model.
- Did it break in the middle of a uniform section? No feature, no layer plane, just a clean break across a flat wall. That is a genuine section-modulus problem: the part needed more material or a different shape.
- Did it break where two parts met? Screw, insert, adhesive joint, press fit. Then the joint, not the body, is the design. There is more on that in threads and heat-set inserts.
If none of the four answers are satisfying, the part probably failed by a mechanism that is not visible at the fracture at all: creep, chemical attack, or fatigue. Those need a different kind of investigation, and they are the ones that make people say "it broke for no reason".
Design causes versus process causes
The table below is the one I would tape to a workshop wall. It splits the usual suspects by which side of the printer they live on, because the fix depends entirely on that split.
| Symptom | Design side | Process side |
|---|---|---|
| Crack starts at a hole | Hole too close to an edge; no boss; load path runs through the hole | Hole printed undersized, creating a press-fit stress that was never designed |
| Crack follows a corner | Sharp internal corner, no fillet, or a fillet too small to matter | Corner was rounded by the slicer but the wall thinned at the same point |
| Part splits in the middle of a wall | Section modulus too low for the span | Severe under-extrusion or a partial clog on that region |
| Failure at the base | Bending moment peaks at the root; root not reinforced | Elephant foot creating a stress riser at the first layer |
| Failure after weeks at low load | Creep: sustained stress above roughly a third of yield | Moisture pickup softening the material over time |
| Failure when warm | Service temperature within 20 °C of the material's glass transition | Chamber too hot, part cooling insufficient, part annealed unevenly |
| Cracks radiating from a screw | Boss wall too thin; insert too close to an edge | Oversized pilot hole splitting the boss on installation |
Notice how many rows have a plausible explanation on both sides. That is exactly why guessing produces so much reprinting. The only way to separate them is to change one thing and test, which is the next section.
Isolating the cause with tests that settle it
The temptation when a part fails is to fix everything at once: warmer, slower, more walls, different material. That approach has a fifty percent chance of working and a zero percent chance of teaching you anything. If it works you will not know which change did it, and if it fails you will have no idea what to try next.
Instead, print a small set of specimens that isolate the variable. You do not need a tensile machine. You need four coupons and a way to break them the same way twice.
| Test | Specimen | What it isolates | Pass criterion |
|---|---|---|---|
| Layer bond | Bar 10 × 4 × 60 mm, printed standing on end so every layer is a potential break | Interlayer strength at your current temperature and cooling | Breaks with visible whitening, not a clean flat split |
| Road bond | Flat plate 60 × 60 × 3 mm, printed flat | Extrusion consistency and lateral bonding | No visible gaps between adjacent lines on the top surface |
| Feature sensitivity | Same bar with a 3 mm hole at mid-span, and an identical bar without | How much the feature costs you | If the holed bar fails at less than half the load of the plain bar, your hole treatment needs work |
| Orientation | Identical bar printed flat, on edge, and standing | Which axis your part is actually weak in | The strongest orientation tells you how to re-orient the real part |
Break them by hand, or with a lever, or by clamping one end in a vice and hanging a known weight off the other. The exact force does not matter much. What matters is that you break four of them the same way and compare the fracture faces, because the comparison is the experiment. A coupon whose face looks completely different from the failed part has just told you that the failed part did not die of that mechanism.
One caution: print all the coupons in the same session, on the same spool, in the same orientation as the real part where relevant. A coupon printed after a nozzle change and an eight-hour drying cycle is not a control, it is a different experiment.
The hidden variables nobody writes down
There is a category of cause that only shows up in the field and never on the bench, and it is responsible for a large share of "inexplicable" failures.
Moisture, absorbed quietly
Nylon is the famous case, but PETG and even PLA will pick up water from the air over weeks of sitting on a spool holder. Wet filament does not fail dramatically; it foams slightly, loses a fraction of its strength, and produces parts that look fine and are ten to twenty percent weaker than the same part from a dry spool. Because it is a gradual change, it never triggers an investigation. If your failure sits in the "progressive crack" row and your spool has been open since spring, dry it and reprint before you redesign anything. A filament drying box or a dryer is cheaper than a redesign.
Heat, and the glass transition you ignored
A part that sits in a car, near a window, on top of a printer enclosure, or inside a device that generates its own heat is not at room temperature. PLA softens noticeably around 55–60 °C. A black PLA bracket on a sunny dashboard can reach that. If the fracture face is glossy and smeared rather than clean, suspect that the part was warm before it was loaded, and the load finished a job that heat started. Heat management is its own discipline.
Creep the design never accounted for
A bolt tightened to hold a part in place is a load that never goes away. Printed plastic under a sustained load will slowly yield, and the preload you set on day one will be gone by month three. The part then rattles, the joint loosens, and something else breaks as a consequence. This is not a material defect; it is a design assumption that was true for metal and false for plastic.
Chemistry with a long fuse
Solvents, oils, cleaning agents, threadlocker, and even some adhesives will attack printed parts slowly. The failure often happens months later, far from the chemical, with no obvious connection. If a part failed inside a machine or a workshop with solvents around, read chemical resistance and stress cracking before assuming mechanical overload.
When it is cheaper to redesign than to reprint
There is a decision point that most people reach but few acknowledge: sometimes the correct response to a failure is not a better print of the same geometry. The signals that you are at that point are fairly consistent.
- The same part has failed in the same place three times, with different materials or settings.
- The failure face says design (brittle, feature-driven, progressive), not process.
- Adding walls or infill has not changed the failure location.
- The part needs to be strong in an axis the printer cannot give you.
- The service load includes heat, solvent, or a sustained preload the material cannot hold.
When those are true, the productive move is to change the load path rather than the material. That usually means adding a rib, moving a hole, splitting the part so the joint sits at a low-moment location, or changing the orientation so the main stress runs along the layers instead of across them. The thinking behind that is in load paths and ribs and gussets.
| What you are tempted to do | Why it often fails | What to do instead |
|---|---|---|
| Switch to a "stronger" material | The weak point was the layer bond, which is material-independent to first order | Re-orient, or add material across the joint |
| Add infill density | Infill rarely carries the peak stress; shells do | Add walls, or change the section |
| Slow the print down | Helps bonding slightly, but does not fix a concentration | Fillet the corner or move the hole |
| Print hotter | Improves bond up to a limit, then causes stringing and sag | Verify with the layer-bond coupon, then stop |
| Make it thicker everywhere | Adds mass and print time without addressing where the stress is | Thicken only the region inside the stress path |
Building a failure record that pays for itself
The value of diagnosis compounds only if you write it down. A single line per failure, kept in one place, turns into a pattern in about three months, and the pattern is what tells you that your brackets always fail at the root, or that your nylon parts always fail in August.
| Column | Why it belongs | Example |
|---|---|---|
| Date and part name | Lets you find the failure again when the pattern appears | 2026-08-14 / bracket-v3 |
| Time in service | Separates instantaneous from time-dependent mechanisms | 11 weeks |
| Where it broke | The single strongest design signal | At the lower screw boss |
| Face classification | Design versus process, immediately | Progressive crack |
| Material and spool age | Catches the moisture and lot variables | PETG-CF, opened 5 months ago |
| Environment | Catches heat, solvent, UV and vibration | Inside enclosure, 42 °C |
| What you changed | Makes the next failure informative rather than repetitive | Moved boss inboard 6 mm, added fillet 2 mm |
Seven columns, one row per failure, and within a season you know more about your own parts than any generic guide can tell you. The record is also the only honest defence against the most common form of self-deception in this hobby, which is remembering the failures that confirm your theory and forgetting the ones that do not.
Five checks before you call a part fixed
- Reprint and test the same way it failed. If it failed under a sustained load, hold it under a sustained load. A part that survives a hand squeeze has proved nothing.
- Look at the new part under the same conditions for the same duration. A progressive-crack failure needs weeks of observation, not a visual inspection.
- Verify that your change did not move the failure elsewhere. Stiffening one region commonly transfers stress to the next weakest point, which is progress but not success.
- Check the dimensions after your change. A rib, a fillet or a moved boss can shift a mating face; a stronger part that does not fit is not a fix. The measurement routine is covered in inspecting printed parts.
- Write the row. No row, no learning. This is the whole system.
Material notes that change the reading of a break
| Material | How it typically breaks | What that means for diagnosis |
|---|---|---|
| PETG | Ductile, whitens, stretches before giving way | If a PETG part snapped cleanly, suspect that it was cold, old, or wet |
| PETG-CF | Stiffer and more brittle than plain PETG; snaps with less warning | Clean snaps are expected; look for feature origins rather than blaming the material |
| PETG Matte | Similar to PETG but slightly lower elongation; matte surfaces hide whitening | Check the face texture with a light at a low angle, not straight on |
| PA6-CF20 | Tough, but moisture-sensitive; wet parts fail progressively | Always confirm the spool was dried before you accept a fatigue conclusion |
| PA12-CF10 | More dimensionally stable, less water pickup than PA6 | Failures here are more likely genuinely mechanical |
| PA612-CF15 | Middle ground on moisture; good toughness retention | Useful as a control when you cannot tell if moisture is the variable |
| PA6-GF25 | Stiffer, more brittle than the CF grades | Expect less warning before fracture; concentrate on feature design |
| ePA-CF | Consistent but still hydroscopic | Same moisture rule applies; dry, then judge |
| PETG-ESD | Conductive filler can act as a crack initiation site | Look for fractures starting at filler clusters or print seams |
| PA612-ESD | Same as above, with better moisture behaviour | Separate electrical requirements from mechanical ones before redesigning |
| TPU 95A high speed and eSUN TPU 95A | Rarely snaps; tears, delaminates, or takes a permanent set | A "break" in TPU is usually a bond failure or a fatigue tear, not a strength limit |
| PLA Matte | Brittle, snaps with almost no warning, creeps under sustained load | Clean brittle snaps are normal; check the temperature before concluding overload |
| Impact resin and ONYX Impact Plus | No layer direction; fractures are smoother and more isotropic | Do not apply FDM layer logic; suspect geometry, cure, or post-cure instead |
Two practical notes on reading those rows. First, printed parts often contain more than one material region: a PETG body with a TPU gasket, or a nylon part with a brass insert. In that case the weakest interface, not the weakest material, is usually the culprit. Second, abrasive filaments such as the carbon and glass filled grades are harder on nozzles, and a worn nozzle changes extrusion consistency without any visible warning. If your failures cluster after several spools of filled material, check the nozzle before you check the design; nozzle material selection covers what wears and how fast.
Quick reference: symptom to likely cause
| What you see | First thing to check | Second thing |
|---|---|---|
| Clean split along a layer, no whitening | Print temperature and part cooling | Orientation: the load is crossing the layers |
| Comb-like grooves along the extrusion direction | Partial clog or under-extrusion | Flow calibration and filament diameter consistency |
| Cracks radiating from a hole | Edge distance and boss design | Press-fit stress from an undersized hole |
| Break at the base of a cantilever | Root fillet size | Bending moment at the root, not nominal stress |
| Failure after months at light load | Sustained preload and creep | Moisture uptake |
| Soft, smeared fracture face | Service temperature versus glass transition | Enclosure heat or nearby electronics |
| Rattling, then a break somewhere else | Fastener preload lost to creep | Vibration loosening the joint |
| Cracks at the top of a tall wall | Layer adhesion at the coolest part of the print | Chamber draught or fan hitting the top layers |
The habit that makes all of this worth doing
Diagnosis is not a skill you use once per disaster. It is a habit that turns every failure into a small amount of information instead of a small amount of money. The parts you print next year will be better because of the ones that broke this year, but only if you looked at them.
The minimum viable version is twenty minutes: classify the face, run the four questions, write one row. That is enough to stop repeating the same failure. The full version adds coupons, a drying check, and a dimension check on the reprint, and it is what separates a workshop that keeps having the same problem from one that keeps finding new ones.
And if you only remember one sentence from this article, make it this one: the fracture face tells you whether to change your slicer or change your geometry, and those two answers are almost never the same. Getting that wrong is why so many people spend a weekend tuning settings to fix a part that needed a fillet.