A reproducible operating method for a metal additive manufacturing cell is not a checklist that lives in a binder. It is a running record that ties every input, every machine observation, every exception and every named owner to the specific build that produced parts on the shop floor. When a cell runs powder-bed fusion or directed energy work, small deviations in powder handling, laser parameters or post-build handling are easy to introduce and difficult to explain weeks later. The practical answer is a step-by-step log built on the same ideas that NIST applies to measurement science for additive manufacturing at NIST’s additive manufacturing program, combined with a lightweight provenance structure drawn from the W3C’s guidance at the PROV Model Primer. The result is a document another engineer can read and re-execute.
The method starts by fixing the boundary of a single build rather than mixing several jobs into one narrative. A build record covers one machine run, identified by the job name, machine identifier, date, operator and the revision of the build file that was loaded. This boundary matters because later analysis of part quality depends on knowing exactly which powder lot, which parameter set and which software release were in play. NIST’s work on additive manufacturing emphasizes that measurements and standards are needed to expand industrial use, and that emphasis applies directly here: without a scoped record, measurements taken later cannot be attributed to a cause. The record should name the owner—typically the process engineer or cell lead—who is responsible for the content, not just the person who typed it. Ownership makes the log usable when questions arise during review.
Before the build begins, the log must list inputs with enough precision to be reordered or reproduced. That includes the metal powder lot number, the sieve or conditioning history, the build plate material and identification, the machine software version, the parameter file revision and the support strategy used for the geometry. Where the site tracks classes of equipment or material states, those classifications belong next to the raw values; for example, a powder reuse category or a machine calibration class can be recorded through internal references such as material and process type records. The input section is not a marketing datasheet. It is the minimum set of facts that another operator would need to set up the same job on the same or a comparable machine. If a photograph of the powder container label or the loaded plate is part of normal cell practice, the log should state that the image exists and where it is filed, rather than assuming the image itself is the record.
Once the build is underway, the operator records observations at defined points rather than only at the end. Useful checkpoints are after powder spreading begins, after the first layer is fused, at a set interval such as every fifty layers, and immediately after the build completes. At each checkpoint the log notes what was seen: layer appearance, any recoater anomaly, melt pool monitoring output available on the cell, alarms displayed and the action taken. This is where the PROV approach helps. The primer describes entities, activities, agents and the usage and generation of data over time, so the log can separate the activity of printing from the entities consumed and produced. A practical entry might say that activity “layer fusion” used entity “powder lot X” and generated entity “as-built stack” under the responsibility of the operator named earlier. The detail prevents the later assumption that no news meant no events.
Exceptions are the reason the log exists. If a build is paused because a filter warning appears, the record must give the time, the displayed message, the pause duration, the decision made and the person who authorized resuming. If a support structure is found detached after depowdering, that observation belongs in the same record even though it occurred outside the machine. Exceptions should not be summarized into a single “minor issues” line; each one needs its own timestamped entry so the sequence is clear. The PROV notion of derivation and revision is useful here: if the build plan was revised after a pause, the log should show that a new revision of the instruction was derived from the earlier plan and who approved it. This makes the difference between a controlled interruption and an undocumented change visible to later readers.
A build does not end at machine extraction. The log continues through stress relief, plate removal, part identification and transfer to inspection or finishing. Each handover names the receiving activity, the parts or plate as entities, and the new owner responsible for the next stage. Where inspection follows, the measurement export should reference the build identity so results are not orphaned from their source; this is the practical counterpart to earlier operational notes on keeping build identity, measurement export and revision together. Internal links such as inspection and measurement type records can sit beside the handover entry when the site already classifies those steps. If parts are routed to a different cell, the log records the transfer rather than closing the file, because provenance across the web of activities is exactly what the W3C primer describes when it explains agents and responsibility influencing an object after generation.
The last step is a review pass by the named owner before the record is considered complete. The reviewer asks a narrow question: could a competent engineer unfamiliar with this specific run repeat the process or explain its deviations using only this log and the referenced files? If the answer is no, the missing input, unclear observation or unassigned owner is added. Reviews should be periodic for a cell, not only triggered by a failure, so the practice stays current. Over time the accumulated records give the organization evidence about which inputs correlate with rework or scrap, without requiring invented statistics or guaranteed outcomes. They also support standards adoption because the same structure—inputs, activities, observations, exceptions, owners—can be extended to other workcells; a classification such as equipment and cell type records helps keep those extensions consistent. The value is not in the format itself but in the discipline of writing what actually happened, at the time it happened, under a name that can be asked about later.