Capacity planning for additive manufacturing selects printer families and unit counts against forecasted part demand, constrained by per-printer build volume, material compatibility, and the soft constraint that almost any 3D printer can in principle run almost any STL file [S1]. The classical capacity-planning definition — equipment needed to meet future demand — applies, but with an extra combinatorial layer: the planner must also decide which technologies (FDM, SLA, SLS, MJF, DMLS, PolyJet) carry the load [S1][S2].
Short-term execution sits on top of that long-term choice. A high-volume print farm is typically run as a three-step operation: prepare (CAD → STL, slicing, machine setup), produce (unattended layer-by-layer build), and collect (post-processing and dispatch) [S5]. Manual schedulers struggle to oversee whole fleets, which is the gap automated batching and placement software close [S3].
What capacity planning actually decides in an AM context
The decision is twofold: (a) which printer families to procure, and (b) how many units of each, given a demand mix of part geometries, sizes, and materials [S1]. Because a single FDM or SLS machine is optimised for certain geometries yet remains technically able to print most STL files, the technology-assignment problem becomes combinatorial, not just a bill-of-materials exercise [S1]. Adding to that, demand for parts from new product development is intrinsically hard to forecast, which is why multi-criteria tools such as fuzzy AHP appear in the literature on 3D-printer selection [S1].
For distributed networks — including social-manufacturing job shops — planners also have to score each candidate printer against part projection area, expressed as PAC = Σ(part projection area) / printer build area, and against cycle time plus the rasterization plus placement cost of a given nest [S4]. The Figure 4 Modular product page from 3D Systems is a concrete reference point: a scalable cell rated for up to 10,000 parts per month with capacity expandable to 24 print engines in a single workflow [S6].
Throughput drivers: fill density, batching, and orientation
Throughput in a 3D printing factory is dominated by two variables: how many parts fit in one build, and how often each build is started. A 2015 Winter Simulation Conference paper modelled Shapeways' shop floor and showed a two-stage procedure — bin-packing with due-date and size-mixture constraints, then 3D tray placement via third-party nesting software — increased printer throughput by approximately 10% over manual batching [S3]. The same paper frames batch composition as an extension of the classical one-dimensional bin packing problem with added lateness and size-mixture terms [S3].
Production planning at the part level decomposes into three explicit stages. First, a printability check is run and orientation is set per object. Second, parts enter a buffer until enough cumulative volume exists to justify starting a build; below that threshold, printers are intentionally left idle to avoid under-filled trays [S3]. Third, parts are assigned to a batch — a group to be run on one specific printer — and the batch is nested to maximise density before slicing [S3][S4].
Decision criteria: technology, build envelope, material, post-processing
Most job-shop and bureau service menus today expose the same core technology set, and PCBWay's public capability list is a representative example: FDM, SLA/DLP, SLS, SLM (often grouped with DMLS for metals), MJF, PolyJet, plus vacuum casting for short runs [S2]. Each binds a different cost-throughput-material triangle, and selecting among them is the practical shape of capacity planning at the technology level [S1][S2].
A useful four-criteria comparison for assigning a part to a process family reads: (1) build envelope versus largest part dimension — SLA and SLS typically outsize desktop FDM on Z-height; (2) material — FDM is thermoplastic filament, SLA is photopolymer resin, SLS uses polymer powder, MJF uses fumed powder, DMLS/SLM uses metal powder, PolyJet uses photopolymer jets [S2]; (3) surface finish and tolerance — SLA and PolyJet deliver the smoothest as-printed surfaces, FDM the roughest; (4) throughput per build — SLS, MJF, and DMLS pack the build chamber densest because no support structures are needed for most geometries, which is exactly the lever automated nesting tries to exploit [S3][S4].
Who this is for, and who should not bother
Formal capacity planning pays back in three settings: bureaus running 50+ machines, in-house factories serving new-product-introduction pipelines with unpredictable part mixes, and distributed social-manufacturing networks that route jobs across many small printers [S3][S4]. The 10% throughput gain documented at Shapeways came from automating batching across a multi-printer marketplace, not from tuning a single machine [S3].
It is overkill for a single-printer prototyping desk, a maker space with one or two FDM units, or any shop whose entire weekly output fits inside a single build tray. In those cases, capacity planning collapses to "do we need a second printer," and a one-page build-volume versus part-mix table is sufficient. As a side note for process engineers who also specify pressure transmitter and flow meter instrumentation on the same plant, the planning discipline is similar: pick a device family by envelope and material first, then count units, then schedule.
Failure modes and constraints planners keep hitting
The recurring failure mode in real 3D printing factories is under-filled builds. A buffer that is too small starves the printers; a planner who is too aggressive at starting batches wastes material and machine hours on sparse trays [S3]. The second failure mode is technology mis-assignment: a part routed to FDM that would have nested far better in SLS, or a metal part queued on a polymer machine [S1][S2]. The third is orientation lock-in — once a part is oriented and supported in software, changing the orientation mid-batch invalidates the slice, so planners tend to commit early and live with the packing result [S3][S4].
A practical guardrail used in the literature is the PAC ratio itself: keep cumulative part projection area close to but not exceeding the printer's build area, and re-bin only when the buffer crosses the threshold needed to fully populate a build [S3][S4]. This is the same discipline engineers apply when sizing a pressure sensor array against pipe diameter — match envelope to the largest unit first, then count.
Standards, sourcing, and the modular-cell escape hatch
There is no single ISO or ASTM standard that dictates how many 3D printers a factory should own; capacity planning lives in operations-research literature and in OEM datasheets rather than in regulated standards [S1]. What suppliers do publish are throughput envelopes: a 24-engine modular Figure 4 cell is rated to 10,000 parts per month with automated job management, automated material delivery, and centralised post-processing [S6]. That figure is a useful sanity check when sizing a greenfield bureau — one modular cell of that class can replace a hand-configured farm of 8-12 standalone SLA-style machines once queue and post-processing are factored in [S6].
For related plant-spec reading, 3D scanner selection follows a similar envelope-plus-throughput logic, and the BOM-and-integration view in industrial robot cost breakdowns mirrors the modular-cell TCO argument. Track two signals next: OEM disclosures of parts-per-month per engine (currently a single-figure rating on modular cells such as Figure 4 [S6]) and any peer-reviewed updates to fuzzy-AHP selection weights for new-material printers [S1].