Staring is caring…

I can still remember the first time I saw something go really wrong in an Additive Manufacturing process with my own eyes.

Of course, I had already seen plenty of parts with lines in them, distortions, cracks and all manner of weird and wonderful defects. Usually these were accompanied by outlandish theories about what had gone wrong, generally involving mechanisms which didn’t stand up to much scrutiny, normally couched in metaphysical jargon which made it abundantly clear that no-one really had a grasp for what was going on in these processes.

I gradually started to notice a trend to these theories. Laser people always thought it was something to do with the laser, powder people always thought it had something to do with the powder and gas people always thought it was to do with the gas flow. Of course if you get any of these things seriously wrong they will result in defects, but I started to wonder if there wasn’t a much more fundamental variable which we were constantly and intentionally changing but which most people didn’t seem to consider - the geometry of what you are building. Of course this was something that the 'real' operators, who spent hours standing by the machine watching their builds running, knew already.

The first time I saw a support failure during the build process it was eye-opening. There I was, head pressed against the warm glass of the M270 watching another layer of powder spreading cleanly across the surface, when suddenly

“bang”

A huge hole (well at least 50mm across and maybe 5mm deep) appeared in the otherwise immaculate powder bed, and I could see the 3-dimensional geometry of a section of the part from the side in a way which, well, you really shouldn’t in this kind of process. The laser started skipping merrily around as if nothing had happened, and while I was still trying to understand what the hell had caused this crater in the powder bed, along came the re-coater and with a nice big dose of powder it completely filled in the hole.

The only thing which indicated something different was a shiny sliver of the part poking up out of the powder, and within a couple more layers it was no longer visible, and everything looked perfectly normal again. I noted down the layer-height on my notepad, watched the process for about another 30 minutes until I got bored, and then went back to the office.

When the build was extracted, we found a support had failed close to the baseplate, resulting in the part suddenly distorting as the residual stress was released, flinging the powder up into the gas flow and leaving a segment of the part sticking up ever so slightly higher than it had been before the support failure. There was a witness line in the part which corresponded pretty well to the layer height I had noted down. Suddenly the mechanism for these mysterious shift lines we occasionally saw started to fall into place for me.

This was back in the days when all this stuff was research so we had the luxury of time to investigate, ponder and even run experimental builds to try and mimic what had happened. We learnt a lot from that chance observation.

I could tell a similar story for almost every type of defect I have seen, and in my experience the mechanism behind each of them could only be adequately understood by seeing them happen, and then dissecting the part afterwards to understand what had happened below the surface of the powder bed where you couldn’t see what was going on at the time.

The majority of defects I have personally witnessed involved something you could see happen through the window if you were lucky enough to be there at the precise moment when it happened. A support failure, a crack in a part, a sudden distortion due to an island merging event, local overheating due to insufficient support on a shallow angled geometry, a progressive distortion due to a shallow unsupported segment, (followed often by a re-coater crash with the offending segment) or a short feeding of powder due to a very large scan area in line with the re-coater which was not adequately compensated for by the rather crude algorithms used to adapt the powder dosing. All these things (or their consequences) can be seen with your eyes as they happen - if you are looking.

Seeing these things happen in real time through the window of the machine gives an appreciation which you simply cannot get any other way for just how extreme the forces at play are in these processes.

The residual stresses in a part that you can hold comfortably in your hand can bend a 35mm thick Titanium build plate by several mm so that you can spin it on the table. That should give you some idea of the amount of elastic potential energy stored in that part during the build process, which needs to be resisted by the supports, and which can only be relaxed through heat treatment.

It is these forces which tend to wreak havoc in the build process, and which support structures are primarily there to resist. If you get it wrong and they fail during the build process, you have got yourself a scrap part and probably a scrap build, or at the very least a lot of expensive investigations in order to prove the parts are fit for release.

So, what can you do to be sure that parts have built correctly and aren’t riddled with defects?

Well firstly you can see obvious things like support failures and shift lines in a visual inspection – and you should understand their significance and the multitude of other internal defects which they might be an indicator of. Other Non-Destructive Test (NDT) methods such as Computed Tomography (CT) and Dye Penetrant Inspection (DPI) are also good way to find defects, but they also have their limitations (not to mention the expense).

There is no magic bullet, but in my opinion there is still no substitute for staring through the window of the machine during the process. These days we prefer to let the in-built camera systems do this for us. We have camera systems on all of our machines, and after years of manually inspecting those thousands of images per build, we have now developed an algorithm which does that for us, identifies suspicious layer images and automatically categorises and flags them for human inspection.

Thanks to the fact that we document all of our inspections in quite some detail, 3D scan every part to check for distortion and often use CT-scanning as an NDT method, we have gathered a lot of reference data to correlate back with those images to understand the consequences of what we see during the process, and how they manifest themselves as defects (or don’t) in the final part.

This enables us to identify much earlier in the production process if anything has gone awry and helps us to better understand the root cause of the defects which we do occasionally encounter. And of course, understanding the root cause is the way in which we identify corrective actions to improve our processes and avoid similar defects in the future.

We are not yet perfect, but we are constantly improving and I think that should be true of everyone striving to apply this technology in safety critical applications. Staring through the window of the machine is still an important part of that.

Oh, and before some salesperson spoils the mood by commenting that ‘their technology/process/company is great and they never have these issues’ – just remember that their denial is simply evidence that they probably aren’t inspecting properly, aren’t learning from their mistakes and hence aren’t improving… avoid them like the plague (or Covid-19, to be more topical).

Happy staring everyone!