Johns Hopkins APL Looks at Machine Learning for LPBF

Researchers at The Johns Hopkins University Applied Physics Laboratory (APL) have developed a new approach that integrates 3D printing with machine learning, specifically for laser powder bed fusion (LPBF). Essentially, the model allows them to create simulations useful for verifying the production of materials created using LPBF. Machine learning is part of a branch of artificial intelligence and can be applied in various fields, from medical to aerospace.
The technique devised by APL makes it possible to predict what microstructure will be formed on the printing surface thanks to measurements made on the single layer of powder. To do this, researchers used computational modeling and simulation. These predictions of the object that will be created allow for early intervention in case of errors. This thus not only saves time, materials and costs, but also could exponentially increase the output of materials manufactured through LPBF technology.

This image shows the microstructure prediction process, which is achieved by analyzing the impact of cooling rate and temperature gradient on grain orientation and size.
This study is part of a larger body of work being conducted at The Johns Hopkins University Applied Physics Laboratory focused on using artificial intelligence to accelerate the discovery of new materials for extreme environments. Morgan Trexler, the liaison for APL’s Science of Extreme and Multifunctional Materials program in the Exploratory Research and Development Mission Area, commented, “We anticipate that this new approach will be extremely impactful in helping design and understand material formation during additive manufacturing processes, and this fits into our overarching strategy focused on accelerating materials development for national security.”
Why Use Machine Learning for LPBF
In LPBF, layers of metal powder are fused by a high-power laser to create three-dimensional objects layer by layer. This technology is useful for producing strong metal parts as well as those with complex geometries. However, because the powders are different from each other and thus have their own characteristics, the processing conditions can undergo many variations, starting with the laser settings and ending with the interactions between the various powder particles. As a result, the properties of the objects being printed can vary greatly.
To realize this application, the team of researchers led by Li Ma, senior engineer at APL, used a computational fluid dynamics (CFD) model to accurately measure temperature changes and cooling rates during the printing process as they relate to grain orientation and size. Computational fluid dynamics (CFD) is a discipline that employs computer simulations to predict the behavior of material flows, based on the laws of conservation of mass, momentum and energy. This innovative approach makes it possible not only to predict the microstructure of the part before printing, but also to estimate the mechanical properties of the material and the physical performance of the finished part.
Ali Ramazani developed the first phase-field microstructural formation model by integrating the results obtained from the CFD model. This approach improved the accuracy and validity of the simulations. However, its contribution, while significant, was not sufficient to solve all the data collection issues. Indeed, the production of a single component using LPBF technology involves millions of interactions between powder and laser, requiring an enormous amount of computational time and thus making the simulation of each small section extremely complex.

In this image, the results of the APL probabilistic diffusion field model were compared with the simulation results. The APL model accurately detects the microstructure formation and grain growth observed in the simulated results.
The breakthrough came from Hudson Liu, an intern in the APL Student Program to Inspire, Relate, and Enrich (ASPIRE) and a high school student at the Gilman School in Baltimore. He theorized a machine learning model that would greatly reduce the need to run expensive simulations. Liu integrated several pre-existing machine learning models to develop what the team calls a “diffusion probabilistic field model.” This model generates images based on the cooling rate and thermal gradient of LPBF printing, quantifying temperature changes based on parameters such as the distance between the laser impact point and the surrounding solid metal.
Hudson Liu said, “The key benefit of using a model is its speed. Our model can approximate in seconds or minutes what would take hours in a simulation,” adding, “This allows researchers to quickly explore a wide range of parameters and at much lower cost.” The model was validated through microscopic analyzes of LPBF material. Training of the Machine Learning program required more than 400 simulations performed at APL.
Future Outlook
The research team is already working to train new models using video data, which will allow microstructures to be predicted in both 3D and 2D. In time, it will then be possible to make predictions of the microstructures of larger components and to analyze the results of more laser passes.
But that’s not all. The project, which came about through internal funding at the university, has been the subject of interest from NASA’s Space Technology Research Institute (STRI), which would like to implement it for its own services. Li Ma said, “NASA wants validated models that can help them predict what will happen in a build, and how the subsequent part will perform, without expensive experimentation. So this approach is valuable, particularly when you think about doing additive manufacturing on the Moon or in space, where experimentation becomes so expensive that it’s effectively impossible.” It will therefore come as no surprise if in the near future we hear about the applications of APL implemented not only in NASA space travel, but also in other areas.
What do you think about the use of Machine Learning with LPBF 3D printing? Let us know in a comment below or on our LinkedIn, Facebook, and Twitter pages! Don’t forget to sign up for our free weekly Newsletter here, the latest 3D printing news straight to your inbox! You can also find all our videos on our YouTube channel.
*Cover Image: the image shows the microstructure of the precipitation-reinforced nickel-based superalloy used to validate the APL model and its cooling rate prediction (photo credits: ASM International 2024)