3D printing promises to change how we build aerospace engines and gas turbines. But there is a catch: the high-performance superalloys required for these extreme environments are prone to tearing themselves apart during the printing process.
In International Journal of Extreme Manufacturing , Prof. Jianxin Xie's team at the University of Science and Technology Beijing developed a machine learning framework to perfect the laser settings, reducing internal crack density by 99% and boosting the metal's high-temperature strength well beyond that of traditional cast components.
The trial-and-error trap
Laser powder bed fusion works by using lasers to fuse fine metal powder layer by layer. It is a massive advantage for printing geometrically complex parts, but when manufacturers attempt to print crack-susceptible superalloys like CM247LC, the intense localized heating and rapid cooling cycles often ruin the build. Severe internal cracking and interlayer delamination routinely lead to scrapped parts and substantial economic losses.
To fix these defects, engineers typically rely on a trial-and-error approach, manually adjusting the laser power, scan speed, and powder thickness. This method generates a massive parameter space that is practically impossible to explore exhaustively, creating a costly and inefficient bottleneck for scaling 3D-printed superalloys onto the factory floor.
A data-driven rescue
Seeking an alternative to the guesswork, Prof. Xie's team developed a data-driven strategy to navigate the complex physics of the printing process.
They started with a small set of 30 physical experiments to map the baseline behavior of the metal. To avoid the expense of thousands of physical tests, they used a statistical algorithm called a Gaussian mixture model to simulate 100 virtual data points, enriching their dataset. Finally, they applied Bayesian optimization, an algorithm that mathematically hunts for the most promising combinations of settings, to find the perfect laser parameters.
Within just two experimental iterations, the algorithm identified a narrow processing window that effectively eliminated structural tearing. The optimized parameters produced a near-perfect build with a crack density of just 0.01 millimeters per square millimeter, the lowest value ever reported for this alloy without chemically altering the material.
The resulting metal is exceptionally robust. Tested at 900 degrees Celsius, a standard operating environment for gas turbines, the 3D-printed alloy demonstrated an ultimate tensile strength of 937 MPa, a 23% improvement over conventional cast versions of the metal. Yield strength increased by 35% to 809 MPa, all while maintaining a 10.8% elongation to ensure the component bends before it shatters.
Freezing the flaws
There is a good reason why these cracks form in the first place. When molten metal cools slowly, heavy elements like carbon and hafnium have time to drift and pool at the edges of the growing metal crystals, much like debris being pushed to the edges of a freezing puddle. These concentrated borders create weak, brittle zones that easily snap under thermal stress.
The machine learning model bypassed this by identifying a specific combination of a lower laser energy input and a faster printing speed, which forced the molten metal to cool more rapidly. By freezing the melt pool faster, the algorithm effectively trapped these elements evenly in place, preventing the formation of weak boundaries.
Furthermore, the optimized laser path encouraged the metal crystals to grow in a unified direction. Instead of forming a jumbled microscopic structure with highly mismatched boundaries, the grains aligned more closely, behaving like a neatly stacked brick wall that naturally resists tearing. Finally, the modified heat distribution drastically reduced the built-in thermal tension, or residual stress, that normally warps and cracks printed parts.
Who could benefit?
Although the algorithm successfully cured the cracking problem, the physical validation has so far been performed on laboratory-scale 10-millimeter test cubes and small tensile specimens. The mathematical framework itself, however, is fully mature and proven to work.
The immediate next step for the field is to scale this exact algorithmic blueprint to print full-size and geometrically complex turbine blades. This predictive framework could also be applied to other crack-sensitive alloys critical to the energy and aerospace sectors, offering an efficient route away from costly trial-and-error and paving the way for wider industrial adoption.
International Journal of Extreme Manufacturing (IJEM, IF: 21.3 ) is dedicated to publishing the best advanced manufacturing research with extreme dimensions to address both the fundamental scientific challenges and significant engineering needs.
Visit our webpage , like us on Facebook , and follow us on Twitter and LinkedIn .
International Journal of Extreme Manufacturing
Machine learning assisted laser powder bed fusion process optimization of CM247LC: crack mitigation and strength–ductility enhancement
8-Apr-2026