Production additive manufacturing doesn't need more data. It needs a control plan.

A metrology publication recently ran a piece calling data-driven quality the missing infrastructure for production additive manufacturing. Diplomatic phrasing. The blunt version: most AM producers shipping into aerospace and automotive haven't written a proper PFMEA, haven't built an inspection routing that reflects how a printed part actually fails, and are passing off machine telemetry as a quality system. I qualify new manufacturing processes for aerospace production at Airbus. AM hits my gate every week. The pattern is always the same — brilliant machine capability claims, paperwork that belongs to a subtractive process, and nobody who can answer what happens when the thermal gradient goes sideways on layer 412.

The AM industry spent a decade perfecting the build and forgot the verification

Every new manufacturing process hits the same wall. Someone perfects the machine, publishes a white paper, and then discovers that building a part is roughly 30% of the work. The other 70% is proving — every single time, with evidence a regulator accepts — that what came out of the machine is what the drawing said should come out.

I lived through this at SNOP when we stood up a 900-person greenfield plant. The press lines were specified, commissioned, and capable within months. The quality system took another year. That gap is where most AM producers are sitting right now — except they're shipping flight-critical brackets while claiming the dashboard covers it.

The dashboards are impressive. Melt pool monitoring, layer-by-layer thermal imaging, in-situ sensors streaming from the build chamber. None of it replaces a control plan. It's observation without control. You can watch a process drift in real time and still not have the documented reaction plan, the inspection gate, or the traceability that an auditor — or a failure investigation — requires.

Why your machining PFMEA doesn't survive contact with a layer-by-layer process

The failure physics are different. This isn't a marginal point. It's the entire argument.

A machined bracket fails along predictable lines: surface finish degradation, tool wear, dimensional drift across a batch. Your PFMEA catches those because the process model is mature and the failure modes have been catalogued since the 1950s. Now consider the same bracket printed in Inconel. The failure modes include porosity clustering at overhang supports, keyholing at contour transitions, microstructural anisotropy between the build direction and the transverse plane, and residual stress distortion that doesn't appear until stress relief — or worse, until service. Anisotropy alone means the same geometry has different mechanical properties depending on which direction you load it. Try explaining that to a PFMEA template built for a CNC cell.

I've seen suppliers bring AM parts to the table with a failure mode analysis that was clearly copy-pasted from a machining process. Porosity rated as a severity 4. No detection method listed beyond final dimensional inspection. No special characteristic flagging on fatigue-critical surfaces. It reads like fiction. The real PFMEA work — sitting with the metallurgist, mapping what happens at each thermal cycle, linking failure modes to in-process and post-process detection — hasn't been done. It's tedious, unglamorous work, and nobody funded it because it doesn't make a demo video.

If your inspection routing doesn't reflect the physics of how the part was built, you're not doing quality control. You're doing wishful thinking with paperwork.

The inspection gap and why routing verification matters more than telemetry

This is where it gets uncomfortable for the monitoring crowd. CT scanning can find internal porosity, but at €150–400 per part on a production bracket, it's a qualification tool, not a routine inspection. Ultrasonic testing works on flat plates and struggles with the complex internal channels that are supposedly the entire point of AM. Dye penetrant catches surface-breaking defects and misses everything that matters below the skin.

The inspection strategy has to be designed into the process, not bolted on after. That's what Routing Verification KPIs are for. At Airbus I built a verification infrastructure that cut internal lead time by 97% — not by adding inspections, but by routing the right inspection to the right point, with documented evidence, where it controls the outcome. The same principle applies to AM. You don't inspect every layer. You build a routing that captures the critical checkpoints: powder lot verification, build parameter confirmation, post-build stress relief verification, HIP cycle confirmation if required, dimensional verification against the compensated model, and a final NDT strategy that matches the failure modes your PFMEA identified.

The telemetry people will tell you the melt pool data is the inspection. It isn't. It's a process signal. A signal becomes inspection when there's a documented decision rule, a reaction plan, and a traceable link to the part serial number. Most AM producers don't have that chain. They have a sensor and a hope.

AS9100 is coming for AM and the audit will be ugly

I evaluate process readiness against AS9100 routinely. The standard doesn't care that your process is innovative. It doesn't offer innovation credits. It asks whether you've identified your special characteristics, whether your control plan addresses them, whether your inspection routing is documented and followed, and whether your nonconformance process drives corrective action. AM producers who've been operating in a prototyping mindset — where every part is bespoke and the quality system is a phone call to the metallurgist — are walking into a compliance environment they are structurally unprepared for.

It's not just aerospace. The automotive side is moving too. Ford's recent recall of 741,000 vehicles is a reminder that the supply chain has zero tolerance for processes that look controlled but aren't. Anyone selling into either sector should understand that the quality infrastructure is the product. The part is just the deliverable.

AM scales when its quality infrastructure becomes as boring and systematic as a stamped bracket's. The technology is past the point where machine capability is the bottleneck. The bottleneck is that someone needs to sit down and write the control plan, map the inspection routing, and build the evidence chain — the same unglamorous work every other manufacturing process had to do before it earned its place on the production floor. Until AM does that, all the dashboards are just monitoring a process nobody brought under control.