New AI Trick Makes 3D-Printed Metal Parts Stronger and More Reliable

Metal 3D printing works a bit like a very precise welding robot. A laser melts metal powder, layer by layer, until a solid part appears. But how fast each layer cools changes the metal's internal structure — the same way a blacksmith plunges hot steel into water to harden it. Cool it wrong and the part can crack, warp, or fail when it matters.

A second laser can fix this by reheating the metal just behind the first one, slowing the cooling down. The problem: figuring out the right laser power, speed, and spacing is brutally hard, because the physics inside that tiny molten puddle is chaotic. Engineers usually tune it by trial and error, burning through time and material.

The new approach, from researchers Lauren Bogo, Adam Clare, and Dominic Liao-McPherson, uses a controller that learns from its own mistakes. A computer model suggests where to search; real test results then correct the model's errors. In simulation — including deliberately sloppy models, where absorption, heat storage, and powder conductivity were all wrong on purpose — the controller drove error down by more than a factor of ten. A model-only approach stalled at much worse results, even though it looked like it had succeeded.

The catch: this was tested in computer simulation, not on an actual factory machine. Real printers have dust, vibration, and quirks that simulations miss. Still, the idea is promising, because it could squeeze the guesswork out of making metal parts for planes, cars, and medical implants. That means less waste, lower costs, and parts you can trust.