The email came in at 9:47 on a Wednesday. Subject line: "Final Inspection Approval – Velo3D Bracket Batch." I opened the attachment, zoomed into the photos, and felt my stomach drop.
If you've ever approved a part based on a photo, then held it in your hands and realized you were wrong, you know the feeling.
The Setup
We run a Velo3D Sapphire metal 3D printer as one of our production workhorses. It's not cheap, and it's not for simple parts. We use it for aluminum brackets, thin-wall housings, and structural components that have to pass AS9100-style requirements.
By the time the Velo3D SpaceX contract 2025 conversations started showing up in our customer audits, we already knew the stakes. Any company that supplies metal additive parts to aerospace primes has to be ready for someone to pull your process records apart and ask hard questions.
I'm the quality/compliance manager for the shop. I review roughly 200 unique part numbers a year. In Q1 2024, I rejected about 4% of first deliveries because of surface finish, dimensional drift, or incomplete paperwork. This batch was one of them.
The Turn
The order looked straightforward: 50 aluminum brackets in AlSi10Mg. The customer spec called for a critical mating surface at Ra 1.6 µm or better. The first articles came off the building plate and measured well. Then they went to a local machine shop for support removal and finish machining.
They came back looking fine. For about ten seconds.
I put a surface roughness tester on the mating face. Readings were all over the place: 1.8, 2.2, 3.4 µm Ra. Not one part was consistently below the 1.6 µm limit.
According to the machine shop, they hadn't changed anything. Then I looked at their tool receipt. They'd swapped to a new end mill from a different supplier: a China carbide end mill for aluminum, purchased to save about $12 per tool.
Country of origin doesn't automatically mean bad. But this particular tool didn't have the same edge geometry as the one we'd qualified. It cut okay for the first 24 parts, then the coating wore unevenly and the finish started tearing instead of shearing.
From the outside, the batch looked like a normal production run. The reality was that the surface quality was degrading with every part after the first few.
The shop manager told me, "It's within industry standard." I told him our contract didn't say "industry standard." It said 1.6 µm Ra.
We rejected the batch.
The Hidden Problem
Here's where the story changes. While we were preparing the rejection report, I noticed faint discoloration near a thin-wall feature on one of the first-article parts. It looked like someone had run a laser marker over the area—or near it.
We use a fiber laser marker for part identification. I've had more than one conversation with our technicians about fiber vs. IR laser options for aluminum. The surface absorption is different, the heat input is different, and on a thin wall, that heat can distort the part. The marking pass itself was within limits, but a software update two weeks earlier had reset the power table. The spot size was slightly larger than the qualified setting. It was enough to cause micro-deformation on two thin walls.
That same week, I ran a refresher on laser safety. The training deck included a slide titled "CO2 Laser on Hands: Before and After." The before photo looked like a normal hand. The after photo showed deep tissue damage from a split second of exposure. I've never forgotten it. It's why I take laser calibration seriously—not just for safety, but for process control. Every laser in the facility, from the Sapphire's build laser to the marking laser on the bench, is a variable you're responsible for.
So the end mill wasn't the whole story. It was the visible problem. The hidden problem was that we hadn't updated our verification protocol after the marking software changed.
Why Metal 3D Printing Makes This Harder
Metal additive manufacturing makes classic quality problems harder because the material itself is a variable. With a Velo3D Sapphire metal 3D printer, you're building parts layer by layer, and the thermal history can be different from one corner of the build plate to another. Defects in metal additive manufacturing aren't always visible. Porosity can sit just below a surface that measures perfectly with a micrometer.
That's why our first-article inspection includes coupon testing, surface roughness, and a review of the build log. We check powder lot, laser power, layer thickness, and oxygen levels. But we almost missed the marking laser issue because it was outside the print envelope.
Quality control in additive manufacturing starts before the print, not after.
The Cost
The redo cost us $22,000. The replacement parts took two weeks longer than the original schedule. We ate the machining rework, the re-inspection time, and the overtime. The customer didn't charge us a penalty, but they did notice the original report.
Here's the part that surprised me: after I hit "reject" in the system, I kept second-guessing. What if my roughness tester was miscalibrated? What if the customer would have accepted the parts? What if I was being too strict?
The two weeks until the replacement batch arrived were stressful. Then the replacement batch passed first time, and the customer's feedback score for that program went up. That settled it.
People think paying more for quality is a cost. It's not. The $12 tool saved by the machine shop caused more than $22,000 in rework. The real expense wasn't the machining—it was the lost schedule, the shaken trust, and the time we spent explaining why a batch of parts looked acceptable but wasn't.
What I'd Do Differently
After that rejection, we made three changes:
- Tooling changes require written approval before they hit the floor. No more "we just switched a drill bit" surprises.
- Laser marking parameters are checked against a signed baseline after every software update. If the settings don't match the approved log, the part doesn't move.
- Roughness is no longer measured only on first articles. We check every fifth part, and the data goes into the inspection report.
According to AS9100 requirements, process changes that affect product quality have to be documented and approved. We had that standard in the contract, but we weren't applying it to our own tooling supplier. Now we do.
The customer doesn't see the internal debate. They see the inspection report, the surface finish, and the email that says "ready for release." That perception is the company. Quality isn't a department—it's the last impression you leave before the part ships.
If you're running a Velo3D Sapphire metal 3D printer, or any system making parts that matter to someone, don't let a photo be the final inspection. Get a gauge on the part. Question the variables. And when something feels off, even after a first pass, trust the numbers over your hope.
It's easier to explain a rejected batch than a failed part on an aircraft.