Garbage In, Garbage Out
Interdisciplinary spaghetti. That’s kinda what additive manufacturing is like. It’s all over the place and yet all the noodles are locked together. There are so many fundamental sciences that go into making a part with laser powder bed fusion. Looking at a fully described list of inputs is nauseating. It makes my head spin. My tiny portion of the additive manufacturing lasagna covers the intersection of materials science, software engineering, machine learning, and fundamental physics.
A few weeks back I had a breakthrough in my research and development of “clean” melt pool micrographs of inconel 718. Above is a cross-sectional view of a laser melt on a solutionized inconel plate. Very pretty colors. This image was one of the primary catalyst that made me want to start this blog. Look at that CRISP melt pool boundary. It invokes the same feeling that a really clean line art tattoo does. One of these days I’ll finally get a metallography inspired tattoo...
The fundamentals of my current role involve generating very clean melt pool boundary images so that other members of my team can measure their morphology and make adjustments to the processing inputs. In the process-structure-properties relationship chain, my work is in the structure and properties regime. My work has pretty consequential down-stream effects because our goal of process improvement ultimately depends on the quality of data coming out of the lab. Below is a good example that demonstrates that:
The top image was immersion etched in Kalling’s II, and the bottom is the same sample re-polished and electrochemically etched in phosphoric acid (14%) with 5 volts for 5 seconds. Notice how the top image produced with Kalling’s shows very detailed grain structure compared to the phosphoric etched image. Critically, though, note how easily resolved the overlapping melt pool boundaries are. Even in this low resolution image and low magnification, boundaries are easily distinguished. This is important for manual human measurements, but even more important if we ever want to use image recognition software to make our measurements for us!
So, yea, distinguishable melt pool boundaries are really important for us. Even more importantly for utilizing image recognition software, is that my etching results are repeatable across many many samples. These type of algorithms require extensive training sets of manually measured images and data. If the training set is bad (like the kalling’s etch), then we have a classic case of garbage in, garbage out.
I’m far from fully mastering inconel 718 preparation, but it’s awesome to have the opportunity to focus on mastery of one material. When I’m not juggling one material system today/another tomorrow, significant brain space and effort is freed up to dial in procedures. Learning how to drive contrast between melt pools and underlying matrix material has been a really fulfilling challenge. This is not something I’ve found good literature on and have mostly resigned to figuring things out on my own. I’m currently in the “lot’s of happy little accidents” phase. Hopefully I can pull a Bob Ross and turn those accidents into meaningful imagery.
Etched with Kalling’s
Electrochemically etched
