These pages explain the metal laser powder bed fusion (LPBF) failures that trace back to the shape of the part, in the order they usually get expensive. Each one gives the mechanism, the specific measurement that predicts it, where the threshold came from, and — the part most guides leave out — what that measurement cannot tell you.
They exist because the industry’s answer to this question is a PDF of design rules, and a design-rules PDF cannot look at your part. The rules themselves are also shakier than they read: the 45° overhang rule and the h/t 8–12 slenderness rule are both contradicted by published builds, by margins of 2× and 4–8× respectively.
| Failure | Reads like | Predictable from geometry? |
|---|---|---|
| Recoater strike | “It failed halfway through” / “it pulled off the supports” | Yes — overhang, slenderness, section jump and support burden are all shape properties |
| Overhang angle & edge curl | “The downskin is rough” / “the edges lifted” | Yes for super-elevation; the staircase roughness term is exact, the powder-adhesion term is not |
| Layer separation | “It split along a layer” / balling, bead-up | Partly — the cross-section jump is computed; thermal history dominates the rest |
| Trapped powder | “It came out heavy” / “the channel is blocked” | Yes — enclosed cavities and flow continuity are pure topology |
Of the 18 LPBF failure modes in this tool’s published taxonomy, 4 are fully geometry-predictable and 12 are informed by at least one of its 20 gates. The remaining modes — keyhole porosity, element evaporation, phase fractions, spatter from neighbouring parts, GD&T violations, optics damage — depend on energy density, powder, plate layout, heat treatment or a drawing spec, and no amount of mesh analysis reaches them. Every report lists them as out of scope by name rather than omitting them.
This matters for what a screen is for. It is a necessary-condition checker: it can tell you a part is at high risk, and it cannot tell you a build will succeed. The failure that defines the boundary is NASA’s combustion chamber that ran nine seconds — it passed inspection and died from a mid-build process excursion. Design-stage risk and process execution are different jobs.
The tool these pages document is free to use, requires no sign-up, and never modifies your geometry. That last one is deliberate: the most-used tool in this category shut down after users discovered its repair step was silently changing part dimensions. A referee that does not touch the mesh is a design decision, not an omission.
What it wants in return is disagreement. If you have printed a part and measured what happened to it, and the screen’s reading disagrees with your measurement, that case is worth more than a hundred passes. There is no lab behind this yet; the calibration has to come from people who have one.
Screen a part →