Simulate before printing: what an LPBF process simulation really predicts

An LPBF part will warp. This isn't a defect; it's the inevitable consequence of melting metal layer by layer and allowing it to cool unevenly. The useful question isn't whether it will warp, but how much and in which direction—and that can be calculated before using up a build plate.

What a simulation actually does

Simulating the complete physics—the molten bath, vaporization, gas flow—is a research problem and takes days for a single cubic centimeter. What's used in production is something else entirely: a mechanical model that applies an equivalent deformation layer by layer, calibrated with tests, and calculates how the entire part responds.

This achieves three things:

Predicts	With what reliability
Final deformation of the part	Good, if the model is calibrated for that machine, that alloy, and that layer thickness
Where the stresses are concentrated	Relatively good: it says where, not always how much
Whether the part will hit the squeegee or detach from the platform	Very good, and it's what prevents costly failure

And it doesn't predict: surface finish, porosity, microstructure, or final mechanical properties. Anyone who claims otherwise is exaggerating.

The real value: compensate, don't avoid

Here's what almost no one explains well. Simulation isn't so much about preventing deformation as it is about compensating for it: if you know that a face is going to curve 0.4 mm upwards, you export the geometry pre-deformed 0.4 mm downwards, and during manufacturing it will come out where it's supposed to be.

This makes simulation a precision tool, not a preventative one. And that's what allows it to compensate for large, flat, or thin parts, where subsequent machining cannot correct what has shifted.

When it pays off, and when it doesn't

It pays off when	Why
The part is expensive or time-consuming to manufacture	A 40-hour build that fails costs much more than the simulation
It's tall, flat, or thin-walled	The three geometries that move the most
There are few opportunities to test	Titanium, cobalt-chromium, single-unit runs
It will be repeated many times	The cost of simulation is spread
Dimensions are lost if it deforms	And there's no excess material to recover them by machining

It's not worth it when	Why
The part is small and compact	It moves little and testing is cheaper than simulating
It's the first of a family that you'll iterate on anyway	You'll learn more from the first real part
The model isn't calibrated for your combination	An uncalibrated simulation gives a figure with two decimal places and no truth behind it

That last row is the industry's trap: simulation is only as good as its calibration, and calibrating it requires manufacturing and measuring standard specimens on your machine, with your powder and your parameters. Without that, it provides guidance; it doesn't guarantee anything.

What you can always do and it only takes minutes

Before undertaking a full simulation, there are two checks that prevent most errors and require no effort:

See the layer-by-layer build in the slicer. The layers with enormous sections, the islands that appear unconnected to what's below, and the unsupported overhangs are all easily visible. This is where almost everything originates.

Look at where it rests and how it's held. Most platform lift-offs and squeegee collisions can be predicted by calmly looking at the supports for ten minutes.

In SamyStudio, you can view the build layer by layer before launching, along with an estimated time, and correcting incorrect orientation is still free. It's located in the slicer.

How to calibrate, in short

If you're going to rely on real simulation, this is the way forward and there are no shortcuts:

Fabricate a known standard geometry—the classic example is a bridge with several arms cut from the deck, and the height of their rise is measured.
Measure the actual deformation.
Adjust the model parameter until the prediction matches.
Repeat for each alloy and each layer thickness, because the results are not inherited.
Recheck the model whenever something in the process changes.

It's work, and it's exactly the same work required to qualify a part. Anyone who has already done the work for one has half the work done for the other.

In summary

Simulation isn't essential to get started, but it's very useful for avoiding repetition. Its best use is to pre-deform the geometry to ensure it works correctly the first time, and its biggest risk is relying on a model that no one has calibrated for your specific combination of machine, powder, and parameters.

And a consequence that links to the rest: a model can only be calibrated if you know and can adjust its parameters. With a machine of closed parameter sets, simulation is an estimate about a black box. It's in what it means to fully develop an LPBF technology.