In metal LPBF, the build processor translates layered geometry and process settings into scan instructions executable by the machine. When this software changes, the manufacturing instruction may change even if CAD and nominal parameters do not. This guide covers acceptance, regression and change control for the processor and its job file; it does not replace full qualification of a part or material process.

1. Define the digital configuration

The baseline should include CAD, mesh and support revisions, placement, build-preparation and processor versions, machine model and firmware, material, layer thickness, parameter set, calibration state and export options. A filename is insufficient: assign unique identifiers and checksums to input and machine job files.

NIST's AM data and metadata review notes that slices and scan paths commonly enter proprietary vendor formats and no generally interoperable neutral build-job format exists. A job file is therefore not necessarily portable across vendors or even configurations. Local provenance must be preserved.

2. Separate changes by risk

A documentation or interface correction may be low risk when proven not to affect export. Any change that can alter slicing, contours, hatch, vector order, laser allocation, beam diameter, power, speed, skywriting, ghost vectors or file format is high risk. A new parallel-processing algorithm is high risk even when the named function appears unchanged.

Supplier release notes are a starting point, not evidence. Local risk depends on critical part characteristics, qualification agreements and whether output differences can be inspected. If machine files are encrypted, obtain a supplier comparison report or use an approved viewer or diagnostic interface.

3. Build a representative regression pack

One cube cannot cover the processor. Include small and large jobs, many parts, thin walls, islands, changing cross-sections, contour-to-hatch transitions, multi-laser overlap zones, supports and a case near maximum layer count. Include known-invalid input to test warnings and stop behaviour.

Measure total runtime, peak memory, output size and determinism. Repeated execution with the same input and configuration should return identical output or documented equivalent instructions. Speed is useful only when there is no silent loss of function, missing layer or unexplained vector change.

4. Compare geometry and scan strategy

At layer level, compare filled area, contours, supports, vector count and bounding boxes. Look for vanished thin walls, opened inclusions, self-intersections, inverted normals, shifted supports and duplicated vectors. On multi-laser machines compare zone boundaries and load allocation; with variable beams, record which vector types receive another diameter.

Thermal strategy cannot be judged by file size. Vector order, rotation, stripe width, islands, delays and auxiliary heating collectively create the thermal history. NIST LPBF datasets illustrate why build commands, scan strategy, material, in-situ signals and post-build measurements belong in one context.

5. Connect digital differences to physical tests

Digital agreement is necessary but not always sufficient. For high-risk changes, build geometry coupons in critical orientations, density or CT specimens, surface coupons and material-property specimens. Cover every laser and overlap region on multi-laser equipment. Control machine state, powder, platform position, heat treatment and measurement method.

Set acceptance limits before testing. Similar images or means are insufficient; examine variation, extremes and trends. If the release enables a new thermal or exposure strategy, partial or full material and part requalification may be required.

6. Keep performance separate from conformity

Processor speed is an operational KPI, not a quality KPI. Measure operator waiting, total preparation lead time, machine-utilization effects and rerun time after failure. Speed creates value only when it releases a real bottleneck. Saving ten minutes before one six-hour build differs from a centre generating one hundred large jobs per day.

Do not combine processing and printing time. A change in estimated build duration may indicate a real strategy change. If it falls, determine whether laser allocation, vector density or delays changed; if it rises, establish whether stability or quality was the reason.

7. Version lock, release and rollback

Use an approved processor only from a verified installer under controlled licensing and access. The work instruction should name version, compatible machine and firmware state, approved parameters and file-transfer method. Retain regression inputs, outputs, measurements, deviations and approval.

The rollback plan needs the previous installer, licence and configuration, plus confirmation of whether new files are backward compatible. A security defect or urgent patch may justify an accelerated path, but document temporary risk, time limit and subsequent full regression.

Decision gate

Accept a release only when configuration and provenance are traceable; the regression pack passes; changed scan strategies are understood and approved; physical coupons meet limits; performance benefit is representative; and rollback works. If any gate is missing, keep the release in an isolated test environment.

Sources and scope

NIST work on AM data acquisition and digital provenance supplies the traceability framework. Materialise's Next-Generation Build Processor documentation provides current function examples, while the Additive Industries release supplies a concrete regression case. A supplier benchmark is not a universal acceptance limit; each facility must create representative local evidence.