Walk onto a machine shop floor and ask where a part’s history lives, and you get a coherent answer: work orders in the MES, tool offsets and NC programs in the CAM/PLM layer, inspection results routed back through the genealogy record. Ask the same question about a metal additive build cell, and you usually get a shrug. The printer is a black box. It has a job number, a start time, an end time, and a pass/fail. Everything that happened inside — the laser power trace, the powder lot, the recoater blade wear, the melt-pool camera data — sits in a proprietary build-processor export that nobody upstream ever looks at unless something goes wrong.
That gap was tolerable when metal AM was mostly prototyping and low-rate bridge production. It is not tolerable where aerospace and medical device manufacturers are now qualifying AM production lines under process specs that expect the same auditability a machined or forged part gets. A regulator or customer quality engineer doing a nonconformance investigation on a printed titanium bracket wants to trace it the way they’d trace a five-axis milled one: lot, machine, parameters, operator, in-process signals, disposition. Most current MES-to-AM integrations can’t do that without someone manually pulling files out of the OEM’s proprietary software.
Why additive breaks the usual MES assumptions
Conventional MES genealogy was built around discrete, mostly deterministic processes: a router step consumes a lot, a machine executes a program, a result is recorded. Metal AM violates several of those assumptions at once. A single build plate can contain dozens of distinct part numbers sharing one thermal history. Powder is not consumed once — it’s blended, sieved, and reused across builds, so genealogy is closer to a continuous process industry model (think blending or extrusion) than to discrete-part machining. And the process itself throws off enormous volumes of high-frequency sensor data — melt-pool imaging, pyrometry, laser power and position — that no MES was ever designed to ingest, let alone contextualize.
This is the core reason so many teams default to treating the printer as a work center with a binary outcome. It’s not laziness; it’s that the natural home for build-file-level and sensor-level data is genuinely different from the natural home for order-level and lot-level data. The mistake is stopping there instead of building the bridge.
A decision matrix: where should each data type actually live
The practical question isn’t “MES or PLM” as an either/or. It’s which system is the system of record for each category of data, and where the other system needs a reference, not a copy.
Build parameters and machine build files
These belong in the AM build-processor / PLM layer as the system of record. Laser power profiles, scan strategy, layer thickness, and the sliced build file are process-design artifacts tied to the part’s qualified process — they’re conceptually closer to an NC program than to a work order. The MES should reference the build file version and parameter set ID, not attempt to store or re-derive them. Trying to make MES the master of build parameters means duplicating PLM’s job poorly and creating version-control conflicts the moment a process engineer tweaks a scan strategy.
Powder lot genealogy
This is MES territory, full stop, because it’s fundamentally a materials-consumption and lot-traceability problem — the same category of data MES already owns for chemicals, alloys, and consumables in any ISA-95-aligned deployment. Powder blend ratios, virgin-versus-reused percentages, sieve cycles, and lot-to-lot chemistry results should flow into the MES’s genealogy tables exactly like any other raw material lot, linked to the specific build job that consumed them. The complication is that AM powder genealogy is many-to-many — one powder lot feeds many builds, one build can draw from blended lots — so your MES material-consumption model needs to support blend genealogy, not just simple lot-to-order consumption. If your MES can’t model a blended lot with fractional contributions from multiple parent lots, that’s a real gap to flag before you scale.
In-situ melt-pool and process monitoring data
This is the hardest case, and it’s where most organizations get the split wrong in both directions. The raw sensor stream — often high-bandwidth optical and thermal data captured per layer — does not belong in MES. No shop-floor execution system is built to store or analyze that volume natively, and trying forces MES into a role it wasn’t architected for. That data should live in a dedicated AM process-monitoring or data-lake layer, ideally with OPC UA or a similar structured interface exposing it to other systems.
What does belong in MES is the derived quality signal: pass/fail flags, anomaly counts, deviation summaries against the qualified process envelope, tied to specific build layers or part locations on the plate. The MES doesn’t need the melt-pool video; it needs to know that layer 340 threw an anomaly flag correlated to a specific part’s location on the plate, and it needs that flag to become part of that part’s genealogy record and, where relevant, trigger a nonconformance workflow.
Who owns the handoff
This is less a technology question than an organizational one, and it’s usually where integration projects stall. In practice, the AM process engineering group owns the build-processor and parameter side; quality engineering owns the genealogy and disposition logic; plant IT or MES admins own the data pipe between them. The failure mode to watch for is nobody owning the mapping between build-file coordinates and MES part/lot identifiers — without that mapping, a melt-pool anomaly on the build plate can’t be automatically tied to the one part it affects, and someone ends up manually cross-referencing plate maps against traveler records during every audit.
What to actually do in 2026
Don’t try to make MES swallow the AM build-processor’s job. Instead, treat the AM cell like any other complex process asset with an OPC UA or MQTT interface: define the handful of derived signals that matter for traceability, agree on identifiers that let a build-plate coordinate resolve to an MES lot number, and get powder blend genealogy modeled correctly before volumes scale past what a spreadsheet can track. The shops that get audited smoothly on qualified AM production won’t be the ones with the fanciest melt-pool cameras — they’ll be the ones that solved the identifier mapping problem early and didn’t wait for an audit finding to notice it was missing.
This article was written with the assistance of artificial intelligence. While we aim for accuracy, the information may be incomplete, out of date, or incorrect, and should be independently verified before you rely on it for any decision. It is provided for general information only and does not constitute professional advice.
