Back in March 2024, I watched a batch of 14 metal parts come off our Velo3D Sapphire. The top surfaces looked fine. The internal channels looked clear. Then the CMM report came back with three dimensions out of tolerance, and one part showed a crack near an unsupported overhang. (Not a subtle crack. The kind you can catch with a fingernail.)
That build cost us about $3,200 in direct scrap. It also cost us a week, a customer's trust, and the comfortable assumption that if the machine is good, the parts will be good.
I'm a manufacturing engineer. I've been handling custom manufacturing orders for eight years, and I've personally made and documented 14 significant mistakes totaling roughly $180,000 in wasted budget. This one hurt more than most because it was preventable. Now I maintain a checklist so the junior engineers on our team don't repeat it.
Why we bought a Velo3D in the first place
Our shop in Dubai is not a glamorous R&D lab. Before the Velo3D arrived, we were easy to categorize: CNC precision turning parts in one cell, laser welding machines in another, and a lot of experience with clients who demand traceability. We still run that work today. CNC and welding pay the bills.
But over the last few years, customers started asking for geometry that a lathe or a weld fixture couldn't produce. Internal cooling channels. Lattice structures. Lightweight aerospace brackets with organic shapes. We looked at several metal AM options, and Velo3D kept coming up.
The phrase that first got our attention was "Velo3D SpaceX supplier." It showed up in procurement news, not just in a glossy brochure. Then we found the same name tied to Anduril, another defense company that doesn't accept loose quality. That "Velo3D supplier to SpaceX and Anduril" context gave us confidence that the system had been through serious qualification work.
The support-free idea also seemed like a game-changer. We had been told, again and again, that metal 3D printing requires supports for any overhang beyond about 45 degrees. Velo3D's system is designed to reduce or remove supports for many geometries. That was exactly what we needed for the internal-channel parts we wanted to quote.
The decision wasn't a no-brainer. The machine cost more than a mid-range CNC lathe. But our leadership saw it as a way to win work that couldn't be quoted by shops with only CNC precision turning parts and laser welding machines. The same way people search for laser welding machines Dubai when they need a local supplier, our customers were searching for a shop that could handle complex metal AM.
The question that should have been simple
Before the machine arrived, we worked through the obvious questions. Build volume. Material compatibility. Powder handling. Post-processing. Lead times for consumables.
The search phrase "what format does 3d printer use" looks basic. In practice, it was the question that exposed our biggest gap.
Everyone at our shop assumed the answer was STL. We use STEP files for CNC programming, but we thought the 3D printer would take STL the same way the plastic office printers do. That is an easy mistake to make.
STL describes the surface of a part as a cloud of triangles. It's okay for a prototype. In metal AM, it can create faceting errors on curved surfaces, lose exact edge tolerances, and separate the design from the build parameters. The format that matters in industrial metal AM is the prepared build file: sliced, parameterized, and holding layer-by-layer instructions. Many systems use 3MF or a proprietary format at that stage. Velo3D's Flow software takes CAD geometry and generates those instructions. (As of our last software update in late 2024, at least.)
The important point is that the original design data needs to be clean and complete before it reaches the slicer. We didn't take it seriously enough.
The build that went sideways
In early 2024, a customer in the medical device space asked us to quote a component with an internal cooling channel. The channel had a spiral path and a sharp corner that no drill could reach. It should have been a perfect job for the Velo3D.
I converted the design to STL in a hurry. I assumed the file was fine because it looked fine on my screen. I skipped the full build review in Flow. I also assumed the support-free capability meant we didn't need to worry about the sharp corner.
All three assumptions were wrong.
The first sign of trouble showed up in the layer-by-layer monitoring feed. A small section of the cooling channel shifted about halfway through the build. The machine kept going, which it is designed to do. But the shift changed the wall thickness.
The second problem was a dimensional mismatch on the mounting face. The STL faceting created a slightly different angle than the original STEP model. On a CNC part, the programming software would have used the exact model. In the printer prep, we lost some of that exactness.
The third problem was the crack. It wasn't visible right after the build. It appeared after heat treatment, near an area where the geometry created a hidden stress concentration. "Support-free" doesn't mean "physics-free." Thermal stress still has to go somewhere.
By the time we had a complete build report, we had 11 usable parts, two parts with dimensional drift, and one scrap part. The usable parts still needed post-machining, so the total loss was higher than raw material cost.
I don't blame the machine. The 3D printer didn't care about my deadline. It cared about melt pool stability, layer adhesion, thermal history, and oxygen levels. I was so focused on the hardware that I ignored the process around it.
What the failure taught us
If I could redo that decision, I'd spend one day validating the file and running a test coupon before committing to a 14-part batch. I didn't because the order was already late and the machine was available. That is exactly when mistakes happen.
Now every metal AM job goes through a pre-build checklist before we press start:
- Confirm the file format. We prefer STEP or native CAD into Flow, not a low-resolution STL.
- Run the full design review in Flow. Check overhangs, hidden surfaces, and minimum wall thickness.
- Review the support strategy. If the software says no supports needed, we still ask where the thermal stress will go.
- Do a single-build first article on any new geometry before a production batch.
- Compare the post-build inspection report to the customer's requirements, not just to the CAD model.
We have caught 11 potential errors using this checklist in the past nine months. Every one of those would have been a late delivery and an unhappy customer.
"The first article isn't optional. A machine can be excellent and still produce scrap if the file prep is wrong."
Quality is the brand
I used to think quality was a technical topic. It's not. It's a brand topic.
A customer sees a perfect first article and thinks these people know what they're doing. A customer sees a scrapped batch and starts doubting every future quote. The same machine can produce both outcomes depending on how carefully we prepare.
Most buyers focus on print speed and build volume. They completely miss file preparation, build reports, and thermal validation. That's not their job. It's our job. The minute we stopped treating the 3D printer like a magic box and started treating it like a process, the parts got better.
I also want to be clear about what metal AM didn't do for us. It didn't replace our CNC precision turning parts business. It didn't make our laser welding machines obsolete. We still machine critical surfaces after printing. We still weld components when that's the best way to make them. AM is another tool, not a replacement for everything else.
The reason Velo3D's name appears in the same sentence as SpaceX and Anduril is not because those companies wanted a shiny machine. It's because they could trace the process, prove the material properties, and trust the quality system. That trust is exactly what we are trying to build in our own shop, one build report at a time.