The printer is capable. The process isn't.

The most expensive "process improvement" I ever audited took a weekend to make. A process engineer at an additive job shop nudged a scan strategy on a Friday, printed on the Monday and cut 14% from the build cycle. The parts looked superb. Eleven weeks later the porosity he had relocated – not removed – from the core to an unscanned corner of the plate reached a customer as a machined fitting, and the 8D that followed burned roughly sixty times the hours he saved. Nobody approved the change. Nobody had to. The machine accepted the number, and in that shop the machine's acceptance was the entire approval process.

So when the trade press fills with stories of process improvement in aerospace additive – faster iteration, self-tuning parameters, melt-pool cameras – I read them in a different register. Inside a qualified aerospace quality system, "process improvement" is not a compliment. It is a change to a frozen process, and change has a pathway, a price and a paper trail. The industry keeps selling machine capability as process capability, and the distance between those two words is exactly where parts fail.

The machine is not the process

An installation- and operationally qualified printer tells you the equipment works. It says nothing about what actually decides conformance: which powder lot fed the build and how often that powder has been reused; where on the plate the part sat, because no two positions share a thermal history; which operator ran the shift; whether the sieve, the filter and the argon stayed inside their own control limits. Resolution in microns and sensor counts are machine terms. Capability is proven across lots, shifts and powder generations, in Cpk, with evidence retained build by build.

I built a greenfield QA/QC department for more than 900 people at SNOP, and the first principle we installed was that capability is designed into a system, never inspected into a part. Automotive had taught me that earlier. A press is capable; the process is the coil lot, the die temperature, the setter and the control plan wrapped around all three. Additive does not escape that logic. It concentrates an entire process chain inside one chamber and hides it from anyone not deliberately looking.

Improvement is change, and change is requalification

AS9100/EN 9100 and IATF 16949 agree on this more than their partisans admit: a production process change requires review, validation and, where the contract demands it, customer approval before the first serial part. A VDA 6.3 auditor will ask exactly one question about your new parameter set – who approved it, and where is the evidence? The PFMEA revision, control plan revision, first article inspection and coupon data behind that answer are not bureaucracy. They are how a good idea stops being a guess.

Here is the honest ledger of a "free" cycle-time gain:

• witness coupons across three builds, CT and metallography – €8,000–12,000
• two requalification builds with engineering support – €6,000
• PFMEA, control plan and FAI delta documentation – some 60 engineering hours
• customer approval lead time – six to twelve weeks of frozen serial flow

Set that against a saving of, say, €900 per build and the gain can still be worth taking. That is the point. It becomes a business decision backed by evidence, not a side effect that surfaces later as an escape.

I paid for this lesson in a cheaper industry. Much of the failure cost we hunted at Witte Automotive through QRQC, A3 and the Q-Wall was not design error. It was friendly, local, unvalidated improvement: a weld current "optimised" on nights, a drying step skipped because the oven was full. The cost only came down when every change, however small, passed real change control and stood on the wall where the whole plant could see it.

An unapproved improvement is a nonconformity with good intentions.

How to improve without breaking the qualification

Additive can improve quickly and honestly. In a governed pipeline, not a weekend sprint.

Development parameter sets live on development machines, or in quarantined builds that are marked and never touch serial flow. Witness coupons sit on every plate, in fixed positions, with SPC on the results – tensile strength drifts before geometry does. Adoption demands statistics, not a golden build: capability demonstrated across powder lots, plate positions and shifts, on a defined run, with the data retained. And parameter changes are handled at design-change rigour – change note, PFMEA and control plan revision, FAI delta, customer notification where the contract requires it – with the parameters locked in the controller, so a change leaves a log entry instead of a mystery.

In-situ monitoring and machine learning earn their place in that pipeline as evidence, not approval. Melt-pool data retained per part makes requalification cheaper, because you can show the process window held instead of re-proving it from zero. But once parameters self-tune, change control runs at machine speed and governance has to keep pace. I build autonomous systems in my own practice, and the first design rule there is the same one: autonomy without an audit trail is a defect, not a feature. That is the real race in this industry – not the printing.

Additive earned its seat in aerospace serial production on the strength of what the machines can do; it keeps that seat on the strength of what governs them. The printers are converging anyway. The governance is not. You do not manage the printer as the product. You manage the process as the product and let the printer execute it. Until the headlines say that, read every claimed "process improvement" as exactly what it is: a change proposal, awaiting its evidence.