3D Printing

The short answer At room temperature, as-built SLM parts land close to wrought: SLM 316L prints at about 570 MPa tensile against 485 to 515 MPa for annealed wrought bar, and SLM Ti6Al4V reaches about 1,200 MPa against roughly 950 MPa wrought. Yield strength is often higher and elongation lower. Fatigue is the gap: as-built surfaces sit at Ra 6.3 to 12.5 micrometres. The comparison, alloy by alloy The headline numbers are better than most engineers expect, and the reason is the process itself. Laser melting solidifies metal at cooling rates between roughly 10 to the third and 10 to the sixth kelvin per second, which is fast enough to produce a much finer grain and cellular structure than casting or even forging. Finer structure means higher yield strength. What it also means is residual stress locked into the part, a rougher surface than any machined equivalent, and anisotropy, because the properties in the build direction are not the same as the properties across it. Transverse tensile values for common alloys, as-built against wrought. Heat treatment trades yield strength for ductility. AlloySLM as-builtSLM after heat treatmentWrought or machined reference 316L stainless570 MPa tensile, 470 MPa yield, 40 percent elongation570 MPa tensile, yield drops to about 380 MPa, 35 percent elongationAnnealed bar: 485 to 515 MPa tensile, 170 to 205 MPa yield Ti6Al4V1,200 MPa tensile, 1,050 MPa yield, 8 percent elongationAbove 930 MPa tensile, above 860 MPa yield, above 10 percent elongationAnnealed wrought: about 950 MPa tensile, about 880 MPa yield AlSi10Mg360 MPa tensile, 240 MPa yield, 6 percent elongationAbove 267 MPa tensile, above 200 MPa yield, about 10 percent elongationCast A360: about 320 MPa tensile Inconel 718960 MPa tensile, 600 MPa yield, 30 percent elongationAnnealed and aged: above 1,240 MPa tensile, above 940 MPa yieldWrought, aged: about 1,100 MPa and up Maraging steel1,100 MPa tensile, 1,000 MPa yield, 8 percent elongationAged: 1,950 MPa tensile, 1,900 MPa yield, 2 percent elongationWrought maraging, aged: 1,900 to 2,000 MPa Two rules fall out of the table. First, as-built prints match or beat the wrought reference on tensile strength for stainless, titanium and aluminium alloys, because of the fine solidification structure. Second, once you heat treat for ductility, the advantage narrows and in some alloys reverses: heat treated Ti6Al4V at above 930 MPa is below the 1,200 MPa of the as-built print, and 316L loses most of its yield advantage. Why as-built yield can beat wrought, and what heat treatment gives back As-built 316L commonly reports yield strength in the 440 to 470 MPa range against 170 to 205 MPa for annealed wrought bar. That is not a measurement error. The fine cellular subgrain structure produced by rapid solidification pins dislocation movement in much the same way that cold work would, and the residual stress in the part adds to the effect. It is also the reason the part can distort: the same residual stress that raises yield strength is released unevenly if a thin section is machined away or if the part is heated without support. Stress relief therefore does two things at once. It removes the internal stresses that cause distortion during subsequent machining and heat treatment, and it gives back ductility. A typical profile is exactly what the table shows: tensile strength essentially unchanged, yield strength reduced toward the wrought value, elongation improved. For a part that has to bend rather than crack in service, that trade is usually the right one. The practical consequence for a drawing is that the mechanical property callout has to name the condition as well as the alloy, because as-built and heat treated are two different materials with the same name. Fatigue is the real difference Static strength is not what breaks most metal parts in service. Fatigue is, and it is where an untreated printed part is weakest. The mechanism is geometric rather than metallurgical: an as-built SLM surface has a roughness of roughly Ra 6.3 to 12.5 micrometres, and every peak of that surface is a stress concentration sitting exactly where the load is highest. A machined surface at Ra 0.8 to 1.6 micrometres removes those notches. Between the two, published comparisons consistently show as-built printed parts losing a substantial share of their fatigue life relative to polished wrought material, even where tensile strength is equal or better. The same part, the same alloy, four different fatigue behaviours. Three further factors move fatigue life up or down, and none of them is visible in a tensile test. Internal porosity, which acts as a crack initiation site, is reduced by hot isostatic pressing. Build orientation matters, because layers loaded in tension across the layer boundaries behave differently from layers loaded within the plane. And residual stress changes the mean stress the part sees. The practical conclusion is that fatigue life for a printed part has to be established by testing the actual geometry and surface condition, not read from a material datasheet. What each post-processing step buys Post-processing is not cosmetic housekeeping. Each step changes a specific failure mode, and the sequence matters. Four steps, four different problems solved. Adding 20 to 40 percent to production time and cost is typical. Support removal and stress relief. Supports are cut away and the part is thermally relieved, usually in a vacuum furnace. This is the step that stops the part moving later. Hot isostatic pressing, where fatigue or sealing matters. Combined heat and pressure close internal porosity and raise density, which improves fatigue behaviour and makes a pressure-tight part possible. It does not fix a rough surface. Machining of functional faces. Sealing faces, bearing bores, threads, spigots and locating features are machined after printing rather than printed to size, because that is the only way to hold the tolerance and the surface finish together on the feature that has to seal or locate. Peening or polishing for fatigue. Shot peening puts the surface into compression, and vibratory polishing removes the peaks. Both improve fatigue life substantially, and neither restores as much as machining the surface to a controlled finish. Where SLM beats machining outright There are geometries where the comparison is not close, because machining cannot produce the feature at all. Conformal cooling channels that follow the contour of a mould insert instead of being drilled in straight lines. Lattice cores that carry a bending load with a fraction of the mass of a solid section. Internal flow passages with no line of sight to the outside. Multi-part assemblies consolidated into a single printed body, which removes joints, fasteners, leak paths and the tolerance stack between them. In these cases the system-level answer can favour printing even when the material is marginally weaker. A lattice-cored bracket that carries the same load at a fraction of the mass is a better bracket than a solid machined one, and a mould insert with conformal cooling can cut cycle time enough to pay for its own printing. The comparison to make is between finished parts in service, not between material datasheets. Where machining still wins, and the rules that decide the outcome For most parts that fit inside the build envelope and do not need internal geometry, machining a wrought billet remains the better answer, for reasons that are structural rather than sentimental. Isotropy and certified stock. A wrought bar comes with a heat number, a mill certificate and properties that are the same in every direction. A printed part has directional properties and needs its own qualification route. Sealing, wearing and sliding surfaces. Faces that seal, bores that carry a bearing, threads that are loaded and fits that slide need a machined surface, whether or not the blank was printed. Large, flat, fine surfaces. A face that has to be flat over a long length and fine in finish is cheaper and more reliable to machine from solid than to print and then machine. Volume economics and size. At production volume, casting or forging plus machining wins on unit cost, and the build envelope limits both the size and the number of parts per build. Design rules are not optional. Minimum wall thickness is roughly 0.8 to 1.0 mm in aluminium and stainless steel and 1.0 to 1.5 mm in titanium, wall height to thickness should stay under about 8 to 1, and thin unsupported walls risk incomplete fusion and porosity. Orientation, support placement and powder reuse all change the result, and reused powder has to be controlled rather than simply topped up. Fatigue-critical parts need a programme. If fatigue governs the design and the geometry cannot be machined afterwards, the right answer is usually a qualification programme with test coupons in the actual orientation and surface condition, not a datasheet comparison. Choosing between them in one pass Five questions separate the two routes quickly. Does the part need an internal channel, a lattice or a consolidated assembly that machining cannot reach? If yes, print it and machine the functional faces afterwards. Is fatigue the governing failure mode? If yes, budget for machining, peening or HIP and test the finished geometry. Are the critical features sealing, wearing or locating surfaces? If yes, they will be machined regardless, so compare the two routes on the same finished part. Is annual volume high enough for tooling to amortise? If yes, casting, forging or stamping plus machining will beat printing on unit cost. Does the part need a mill certificate for its material lot? If yes, that requirement points to wrought stock and therefore to machining. In practice the two processes are complements rather than competitors, and the strongest answer for a hard part is often both: print the geometry that cannot be machined, then machine the surfaces that have to be precise. See SLM metal 3D printing for the additive route, CNC machining for the subtractive one and surface finishing for the post-processing steps that decide fatigue life. Send a part for a print or machine comparison Scope and sources. Mechanical property values were compiled in 2026 from the direct metal laser sintering material data published by Protolabs for 316L, AlSi10Mg, Ti6Al4V, Inconel 718 and maraging steel in the as-built and heat treated conditions, and from a survey of as-printed SLM alloy properties covering the same alloys plus 17-4 PH and CoCrMo, together with the cooling-rate and anisotropy discussion. Surface roughness ranges, minimum wall thickness, aspect ratio guidance, density and the 20 to 40 percent post-processing cost add-on come from an SLM process guide. Wrought reference values for annealed 316L, Ti6Al4V and cast A360 are the standard published minimums for those product forms. These are typical published ranges from general-purpose machines and parameter sets; they are not a specification for any particular build, because properties depend on the machine, the parameters, the orientation, the powder lot and the post-processing route. Confirm properties by testing coupons in the same orientation and condition as the production part. Nothing on this page states or implies a certification held by any supplier, and no fatigue life should be assumed from the numbers above without testing the finished geometry.

The short answer For plastic prototypes, sintered nylon beats everything else: SLS or MJF PA12 prints at roughly 45 to 50 MPa tensile with 15 to 20 percent elongation and near-isotropic strength, while FDM parts carry only 50 to 75 percent of their strength across layers. For metal prototypes, SLM is the strongest route, matching wrought values. The ranking, on one scale Strength is not a property of the printer. It is a property of the polymer that ends up in the part, of how well one layer bonded to the next, and of the direction the load travels through them. Putting the materials that bureaus actually stock on a single scale makes the comparison a comparison of materials rather than of machine names. Ranked by tensile strength alone, the filled filaments win. Ranked by survival, they lose. Two readings of that chart matter more than the rest. The strongest printed plastic is not nylon, it is carbon-fibre or glass-filled nylon at 70 to 110 MPa. But the second reading is the one that decides real parts: plain PA12 reaches its 45 to 50 MPa with 15 to 20 percent elongation, while the filled grades reach twice that strength with only 2 to 5 percent elongation. A stiff glass-filled bracket resists deflection and then shatters; a plain nylon bracket bends, absorbs the impact, and stays in one piece. Tensile strength on its own is the wrong number to design a prototype against, which is why the working metric for a functional prototype is usually the product of strength and elongation, or simply the measured fatigue life. Material and processTensileElongation at breakLoss across layers SLS or MJF nylon PA1245 to 50 MPa15 to 20 percentNone, essentially isotropic SLS nylon PA11About 45 MPaUp to 200 percentNone, essentially isotropic SLS glass-filled PA1270 to 85 MPa2 to 5 percentNone, essentially isotropic SLA tough resin55 to 65 MPa10 to 20 percentNone, cured as one solid SLA standard resin50 to 65 MPa4 to 6 percentNone, cured as one solid FDM carbon-fibre nylon80 to 110 MPa2 to 4 percent35 to 50 percent FDM polycarbonateAbout 60 MPa5 to 8 percent30 to 45 percent FDM ABS35 to 50 MPa5 to 10 percent25 to 40 percent The table also shows why the answer depends on the load path rather than on the material list. A part loaded entirely in the plane of the bed loses nothing to layer bonding no matter which process made it. A part loaded across the layers loses a quarter to half of its strength if it was extruded, and nothing at all if it was sintered or cured in a vat. Elongation is the column that predicts whether a prototype survives its first drop test. Why layer direction decides the answer Fused deposition builds a part by pressing a molten bead against a bead that has already cooled. The bond between those beads is a weld line, and a weld line is never as strong as the material around it. Published guidance puts interlayer bond strength at roughly 50 to 75 percent of bulk strength, and measured tensile values in the build direction typically land at 50 to 70 percent of the in-plane value. That single fact is responsible for most prototype failures that arrive with a datasheet-attached excuse. Sintered nylon behaves differently because there is no seam to speak of. The laser fuses powder inside a bed held near the melting point, the surrounding powder supports the part, and the result is a continuous structure with the same properties whichever way it is loaded. Resin parts behave the same way for a different reason: the part is cured from a liquid, so it is one solid rather than a stack of welds. This is why a printed bracket that has to take a bending load is usually quoted in SLS nylon even though FDM polycarbonate has a higher nominal tensile figure. If the geometry forces FDM, the standard fix is to orient the part so the layer planes sit across the load rather than along it, and to accept that the axis which cannot be reoriented is the weak one. Metal is a different answer Once a prototype has to be metal, the ranking changes completely and only one additive family is competitive. Selective laser melting fuses metal powder into a part that is 99 percent dense or better, and as-built tensile values land near or above the wrought equivalent: roughly 540 MPa for 316L stainless, 900 to 1,100 MPa for Ti6Al4V, 300 to 420 MPa for AlSi10Mg and around 1,000 MPa for Inconel 718. Heat treatment then trades yield strength for ductility, and post-processing decides fatigue life, because an as-built surface at Ra 6 to 12 micrometres is a row of crack starters. For a metal prototype, the strength question and the process question are therefore separate. Printing wins when the geometry cannot be machined - conformal channels, lattice cores, consolidations. Machining wins when the material has to come with a mill certificate and the properties have to be the same in every direction. The full alloy-by-alloy comparison is set out in our answer on SLM metal 3D printing strength, so this page stays on the plastic question of which technology is strongest. Four moves that make a printed prototype stronger Strength is not fixed at the moment the process is chosen. It can be moved, and the cheapest move is the first one. Orientation is free, material and post-processing are not. Orient the load. Rotate the part in the build so the primary tensile stress runs within the layer plane. Sharpe corners cost 50 to 80 percent of fatigue life on their own, so add 2 to 5 mm radii while the model is still open. Add material where it fails. Ribs give stiffness without weight, and gradual section changes spread stress. Thick solid walls are the expensive way to buy the same result. Change the grade, not the process. Moving from plain PA12 to glass-filled PA12 raises stiffness by roughly 50 percent and tensile strength to 70 to 85 MPa, but it cuts elongation to 2 to 5 percent. On a part that sees impact, that is a downgrade dressed as an upgrade. Post-process for the failure mode. Heat treatment relieves the residual stress that makes a part distort after machining. Hot isostatic pressing closes internal porosity and raises fatigue life. Vapour smoothing or bead blasting removes the surface peaks that start cracks. Limits: when a printed prototype is the wrong answer No printed plastic is a production material. A prototype in PA12 tells you how a PA12 part behaves, not how the eventual injection-moulded ABS or die-cast aluminium part will behave. For a test that has to predict production, machine the prototype from the production material instead. Fatigue is not in the datasheet. Sintered PA12 reports a fatigue limit near 10 MPa at one million cycles, which is a fifth of its tensile strength. Any cyclically loaded prototype needs its own test, in the real orientation and the real surface condition. Powder parts are slightly porous. SLS nylon is not fluid-tight without sealing, and reused powder changes crystallinity and therefore properties. If the part has to hold pressure, plan for sealing or a different process. Resin changes outdoors. Standard photopolymers discolour and embrittle under ultraviolet exposure within weeks unless they are coated. A prototype that lives in daylight needs a resin specified for it, or a finish over it. Size is a constraint, not a footnote. Build envelopes cap both part size and the number of parts per build, and a part that exceeds the envelope has to be split and joined, which changes its strength. No certification is implied here. Which quality system applies to a given build is confirmed per programme and per factory in writing, before an order is released. What to send for a strength-led prototype Send the CAD model, the loads and the direction they act in, the failure mode you are worried about, and the quantity of prototypes you need. Those four items decide the process before price enters the conversation: a part with a bending load and an impact requirement points to SLS nylon, the same geometry with an in-plane load at half the cost points to FDM, and a part with a sealing face points to printing followed by machining. See SLS 3D printing for sintered nylon, FDM 3D printing for filament parts and SLA resin printing for fine detail, then send a model for a free engineering review. Send a model with its loads and get a process recommendation Scope and sources. Material property ranges were compiled in 2026 from published 3D printing material comparison tables by 3DPrintMap (PLA about 50 MPa, PA12 SLS about 50 MPa, glass-filled nylon about 70 MPa, PC about 60 MPa, carbon-fibre nylon 80 to 110 MPa) and from a mechanical property guide for printed materials (SLS PA12 at 45 to 50 MPa and 1,700 MPa modulus, MJF PA12 at about 48 MPa and 18 to 20 percent elongation, tough resin at about 55 MPa). The layer-direction loss figures (interlayer bond strength 50 to 75 percent of bulk, in-plane versus Z-axis values at 50 to 70 percent) and the fatigue limit of about 10 MPa for SLS PA12 at one million cycles, together with the PA11 elongation near 200 percent, come from a study of material selection under dynamic loads and from a rapid prototyping engineering guide. Surface roughness and post-processing effects on fatigue follow the SOMI Custom Parts capability set and the SLM figures already published on this site. These are typical published ranges for general-purpose machines and parameter sets, not a specification for any particular build, because strength depends on the machine, the parameters, the orientation, the powder lot and the post-processing route. Confirm properties by testing coupons printed in the same orientation and condition as the production part. Nothing on this page states or implies a certification held by any supplier.

The short answer SLA cures liquid resin with ultraviolet light and holds the finest detail, down to 25 micron layers and Ra 1 to 5 micrometres. SLS sinters nylon powder with no supports, giving the strongest and most near-isotropic parts. FDM extrudes filament and is the cheapest, with 100 to 300 micron layers and visible layer lines. Three processes, three mechanisms The three names are abbreviations for three completely different physical events, and every difference in the table further down follows from that. SLA stands for stereolithography. A build platform sits one layer below the surface of a vat of liquid photopolymer, and an ultraviolet light source traces the cross-section of the part, curing the resin where it touches. The platform drops, fresh resin flows across, and the next layer is traced. Because the part is cured from a liquid as one solid, there are no weld lines and no direction of weakness. The trade-off is that it needs support structures, which leave contact marks that have to be sanded off. SLS stands for selective laser sintering. A roller spreads a thin layer of nylon powder, a laser fuses the cross-section into it, the bed drops, and the roller spreads the next layer. The unfused powder around the part supports it, so no support geometry is needed at all, which is why SLS can produce internal channels, undercuts, living hinges and interlocking assemblies that the other two processes cannot build in one piece. Reused powder is blended back in at a controlled refresh ratio, because the chemistry of the powder changes as it is heated repeatedly. FDM stands for fused deposition modelling, also called material extrusion. A filament of thermoplastic is fed through a heated nozzle which draws the cross-section bead by bead. It is the only one of the three that builds with real engineering thermoplastics rather than resin or nylon powder, which is why the material list runs to ABS, polycarbonate, ASA, PETG and carbon-filled nylon. It is also the only one where the bond between layers is a genuine weak point. Head to head on the numbers that decide Nine attributes, three processes. The tolerance row and the support row settle most real decisions. AttributeSLA resinSLS nylonFDM filament MechanismUltraviolet light cures liquid resinLaser sinters nylon powderNozzle extrudes molten filament Layer height25 to 100 microns80 to 150 microns100 to 300 microns Dimensional tolerancePlus or minus 0.05 to 0.15 mmPlus or minus 0.1 to 0.3 mmPlus or minus 0.2 to 0.5 mm Surface as builtRa 1 to 5 micron, smoothRa 6 to 12 micron, matteRa 10 to 30 micron, layer lines Smallest feature0.1 to 0.3 mm0.2 to 0.6 mm0.5 to 0.8 mm SupportsYes, contact marks to sandNone, powder supports the partYes, on overhangs Strength characterBrittle in impact, very preciseTough, near isotropicTough, weaker across layers Typical part costUSD 15 to 60USD 30 to 150USD 4 to 24 Best single useDetail, appearance, patternsFunctional nylon partsCheap, fast or large parts Read down the tolerance column and the decision for a mating feature is already made: only SLA comes close to the fit a machined part holds, and even it stops an order of magnitude short of CNC. Read across the support row and the decision for a part with an internal channel is made too, because SLS is the only one of the three that can print it without geometry that must be removed afterwards. Where the precision actually sits Layer height is the number buyers ask about first and the number that explains the least, because it sets surface finish and vertical detail rather than dimensional accuracy. A 25 micron layer produces a smoother flank and a crisper curve; it does not make the part more accurate in X and Y, and it roughly doubles the build time for the same height of part. Layer height trades time for finish, while tolerance is set by the machine, the calibration and the material's behaviour after the build. A finer layer buys finish and detail, not accuracy, and it is paid for in machine time. What each one costs Price follows material cost, machine cost and how much labour the post-processing takes. FDM uses cheap commodity filament and needs only manual support removal, so it is the cheapest. SLS uses a machine that costs as much as a house and consumes nylon powder at industrial prices, so it is the most expensive of the three per part, but it also needs the least post-processing because there is nothing to remove. SLA sits between the two on price and, in practice, is the one where the post-processing step is most likely to be underestimated. Price bands for a small prototype. Volume, geometry and finish move the number more than the process label does. Choosing in one question The three are not interchangeable, and the choice is usually settled by a single requirement rather than by a weighted score. Does the part have to look finished? Choose SLA. It is the only one of the three that produces a smooth surface and fine features straight off the machine, and it is the standard route for appearance models, dental and medical models, and master patterns for casting. Does the part have to work, and work in every direction? Choose SLS. Sintered nylon is the strongest of the three, it is isotropic, and the free powder support allows snap fits, hinges and internal geometry that the other two cannot produce. Does the part have to be cheap, fast or large? Choose FDM. Commodity filament, the largest build volumes and the shortest lead time make it the right answer for concept models, jigs, fixtures and large enclosures where layer lines do not matter. When two answers apply, the usual resolution is to print early iterations in FDM, move appearance parts to SLA, and switch the parts that will be loaded to SLS or to machining. Running the same geometry through more than one process is also how tolerance gets settled empirically: print it, measure the features that matter, and correct the model rather than arguing with the datasheet. Limits: where each one fails SLA parts are brittle. Standard resins break at 4 to 6 percent elongation, so thin snap fits and clips crack. Tough and engineering resins improve this to 10 to 20 percent, but they cost more and still behave like a photopolymer rather than like nylon. SLA parts change in daylight. Photopolymers degrade under prolonged ultraviolet exposure, discolouring and embrittling within weeks outdoors. Anything that lives in sunlight needs a specific resin grade or a protective finish. SLS surfaces are matte and grainy. As-built roughness sits at Ra 6 to 12 micrometres, which is fine for function and wrong for a customer-facing appearance part. Dyeing and blasting change the look, not the texture, and the parts are slightly porous unless they are sealed. FDM is weakest where the load crosses the layers. In-plane strength is good, build-direction strength is 50 to 70 percent of it, and layer lines are always visible. It is the wrong choice for fine detail, for fluid-tight parts and for anything judged on appearance. None of the three is a production process. They produce prototype quantities. Above a few hundred parts the economics move to moulding or casting, and the design rules change with them, so a printed part is validated for the process that printed it and not for the one that will make the production units. No certification is implied here. Which quality system applies to a given build is confirmed per programme and per factory in writing before an order is released. How to get a quoted comparison Send the model, the feature that matters most and the quantity, and the same geometry can be quoted across all three processes so the choice is made on your part rather than on a table. See SLA 3D printing for fine detail and smooth surfaces, SLS 3D printing for durable nylon parts and FDM 3D printing for economical large prototypes, then send a file for a free engineering review. Send a model for a three-process quote Scope and sources. Capability ranges were compiled in 2026 from the published tolerance and surface-roughness tables of Modo Rapid (FDM layer 100 to 300 microns, tolerance plus or minus 0.2 to 0.5 mm, Ra 10 to 30 microns; SLA layer 25 to 100 microns, tolerance plus or minus 0.05 to 0.15 mm, Ra 1 to 5 microns; SLS layer 80 to 150 microns, tolerance plus or minus 0.1 to 0.3 mm, Ra 6 to 12 microns) and from the SLA, SLS, FDM and SLM quick comparison published by Zorapid, which also gives minimum feature sizes and the 50 to 75 percent interlayer bond-strength figure quoted by several sources. Cost bands per prototype part come from a three-process cost comparison reporting USD 4 to 24 for FDM, USD 15 to 60 for SLA and USD 30 to 150 for SLS. The 50 to 70 percent build-direction strength loss, the matte as-built SLS texture, resin ultraviolet degradation and the powder refresh-ratio effect come from a rapid prototyping engineering guide and from a process decision guide. These are typical published ranges from general-purpose machines; building orientation, geometry, material grade and post-processing all move them. Nothing on this page states or implies a certification held by any supplier.

The short answer Print when the question is shape, and machine when the question is performance. 3D printing delivers a first article in one to three days for USD 50 to 500, holding about ±0.1 to 0.3 mm. CNC holds ±0.01 to 0.05 mm in production material but takes three to seven days and costs USD 80 to 2,000. The rule: shape versus performance Almost every prototype decision reduces to one question. Is this iteration trying to establish whether the design is right, or whether it works? A shape question - does the housing feel correct in the hand, do the two halves close, does the cable exit in the right place - is answered by the fastest process that can hold the geometry, and that is printing. A performance question - does the bracket bend under the specified load, does the seal hold at temperature, does the thread strip at the specified torque - is only answered by a part made from the production material by a method that gives production properties, and that is machining from billet. The reason the distinction matters is that a printed prototype answers a performance question wrongly, and expensively. A test that fails on a printed part is ambiguous: it may have failed because the design is wrong, or because the material and the layer direction were wrong. A test that fails on a machined part is unambiguous. That clarity is what the extra cost of machining buys, and it is usually worth paying once per programme rather than on every iteration. Stage by stage: what to print and what to machine Four stages, and only one of them needs a machined part. StageQuestion being asked3D printingCNC machining Concept and formDoes it look and feel right?Best fit, hours to a partPossible but costly and slow Fit and assemblyDo the parts mate?Good to about 0.3 mmAccurate to about 0.02 mm Functional testDoes it survive the load?Wrong material, wrong layersSame properties as production Thermal and chemicalDoes it hold up in service?Rarely representativeReal polymer or real metal Sealing and threadsWill it seal and hold torque?Usually needs machining afterMachined as standard Pre-production samplingIs the design frozen?Unit cost stays flatUnit cost falls with quantity Read the middle column to decide. The last two columns only tell you which process answers it. In practice the pipeline runs in one direction. Print the first five to ten iterations at USD 50 to 500 each and change geometry freely, because there is no setup and no programming to redo. Then machine the one or two designs that are close to final, at USD 200 to 2,000 each, and use those for the tests that have to predict production. Attempting the same tests on a printed part is cheaper and less informative, and a programme that skips the machined stage usually discovers the difference during tooling, which is the most expensive place to find it. The tolerance question settles most cases Where a prototype contains a press fit, a bearing bore, a sealing face or a threaded joint, the decision is already made. Additive processes in plastic hold roughly plus or minus 0.1 to 0.3 mm as a practical standard, which is adequate for form and fit checking and insufficient for anything that seals or rotates. Machining holds plus or minus 0.01 to 0.05 mm and an as-machined surface of Ra 0.8 to 3.2 micrometres, which is production-representative straight off the machine. This is the chart that decides it. Anything that seals or rotates lives to the left of 0.05 mm. Where the money crosses over The cost argument is not printing versus machining in general. It is printing versus machining at a particular quantity, in a particular material, with a particular geometry. Quantity of one design3D printingCNC machiningWhat usually wins 1 partUSD 30 to 80USD 80 to 200Printing, on cost and speed 5 partsUSD 30 to 80 eachUSD 40 to 120 eachPrinting, on cost 20 partsUSD 25 to 60 eachUSD 30 to 80 eachClose, decided by tolerance 100 partsUSD 25 to 60 eachUSD 20 to 50 eachMachining, setup amortised 10 iteration roundsTotal USD 150 to 500USD 800 to 2,000 each roundPrinting, by an order of magnitude Two structural facts sit behind that table. CNC carries a programming and fixturing cost of roughly USD 500 to 3,000 that has to be paid once per design, so the first part is expensive and the hundredth is not. Printing carries almost no setup, so the first part and the hundredth cost about the same, which is a strength at low volume and a weakness at high volume. Published crossover points cluster between roughly 20 and 100 parts for plastics, and as low as 5 to 20 parts for metal, where printing stays expensive per part and machining gets cheap quickly. Below the crossover, print; above it, machine; and if the quantity is somewhere in the middle and the part has a sealing face, machine. Limits: what printing cannot validate Material properties are not transferable. A printed part behaves like the printed material, not like the eventual moulded or cast one. Dimensional and functional conclusions can be drawn; a strength, creep or temperature conclusion cannot. Anisotropy is a hidden variable in the test result. Printed properties depend on build orientation, so two identical models printed in two orientations can fail at different loads. Machined parts are uniform in every direction. Printed parts are not certified stock. Wrought bar arrives with a heat number and a mill certificate. When a regulated programme needs traceability, that requirement points to machining. For machined parts from controlled stock, this is met by default. Machining is not free either. Thin walls that vibrate, deep pockets that need long tools, hard materials that need special cutters and features that require five setups all raise the price of the machined route sharply, occasionally past the point where printing plus a machined finish is cheaper. Neither route covers the production step. A validated prototype does not commit you to a process for production units. Above a few hundred parts the decision moves again, to moulding, casting or stamping, and the design rules move with it. No certification is implied here. Which quality system applies to a given build or a given machining order is confirmed per programme and per factory in writing, before an order is released. How to start Send the model, the volume you expect in the first year, the material you intend to use in production, and the tests the prototype has to survive. Those four items decide the route before price is discussed, and they also decide whether the answer is a printed part, a machined part or the hybrid version where the geometry is printed and the sealing faces, bores and threads are machined afterwards with 0.5 to 2 mm of allowance left on them. See CNC machining for production-representative prototypes, SLS 3D printing for functional nylon iterations and plastic injection molding for the volume step after the prototype is approved. Send a model for a print or machine recommendation Scope and sources. Cost, tolerance and lead-time ranges were compiled in 2026 from a comparison of 3D printing and CNC machining by quantity (USD 30 to 80 for a printed part against USD 80 to 200 for a machined one at quantity one, crossing over around 20 to 50 parts, and printing at Ra 6.3 to 32 microns against Ra 0.8 to 3.2 microns for machining), from a 2026 precision parts comparison (programming and setup of USD 500 to 3,000 on the CNC side, a break-even zone of 50 to 100 parts, and printed accuracy of plus or minus 0.08 to 0.25 mm), and from a additive versus subtractive guide (first articles in one to three days for printing against five to fifteen days for machining including programming, prototype prices of USD 50 to 500 against USD 200 to 2,000, and the 0.5 to 2 mm machining allowance used in hybrid workflows). Cost-per-part figures for the printed route are order-of-magnitude planning figures for simple plastic parts and move with size, material and finish. Nothing on this page is a quotation, and nothing here states or implies a certification held by any supplier; quality systems are confirmed per programme and per factory in writing before production.

The short answer Seven additive families exist, but five carry almost all professional prototype work: material extrusion, vat photopolymerisation, polymer and metal powder bed fusion, and jetting. They differ by how material is deposited, by the accuracy each holds - roughly ±0.05 mm at best, about ±1 mm at worst - and by what the finished part can do. Seven families, five that matter Additive manufacturing is not one technology. The international process categories divide it into seven families, grouped by the physical event that joins one layer to the next, and every capability difference a buyer sees is a consequence of that grouping. For custom prototype work, five of the seven are commercially routine and two are specialised. Seven families, one page. Accuracy ranges by a factor of twenty across them. FamilyMaterialsAccuracyWhere it is used Material extrusion (FDM)ABS, PLA, PETG, polycarbonate, nylon, filled filamentsPlus or minus 0.2 to 0.5 mmConcept models, jigs and fixtures, large enclosures Vat photopolymerisation (SLA, DLP, MSLA)Standard, tough, flexible, castable and dental resinsPlus or minus 0.05 to 0.15 mmFine detail, appearance models, investment casting patterns Powder bed fusion, polymer (SLS, MJF)Nylon PA12 and PA11, glass-filled nylon, TPU powderPlus or minus 0.1 to 0.3 mmFunctional nylon parts, complex assemblies, low-volume production Powder bed fusion, metal (SLM, DMLS, EBM)Stainless steel, aluminium, titanium, Inconel, cobalt chromePlus or minus 0.05 to 0.1 mmMetal functional parts, conformal cooling, consolidated assemblies Material jetting (PolyJet, MJF detail agents)Photopolymers, wax-like supportsAbout plus or minus 0.1 mmMulti-material and multi-colour parts, medical and dental models Binder jettingMetal powder, sand, ceramicAbout plus or minus 0.2 mmBatch metal parts and sand casting moulds Directed energy deposition (DED, WAAM)Metal powder or wireAbout plus or minus 1 mmLarge metal structures, cladding and repair The two families missing from most prototype quotations are the two that do not suit single parts: binder jetting, which becomes economic in batches because the whole build is sintered together afterwards, and directed energy deposition, which trades accuracy for deposition rate on parts measured in metres. Both are real production routes and poor prototype routes. What separates them on the floor Three variables explain almost every difference a buyer notices. The first is what the machine does with material: melting a filament, curing a liquid, fusing a powder bed, or gluing then sintering. The second is whether supports are needed, because a support is a labour step and a contact mark that has to be removed. The third is the energy source and its control, because that is what sets both the accuracy and the material properties a part can reach. Powder bed fusion is the family that removes supports from the equation, since the surrounding powder holds the part. That is why it dominates functional prototype work: internal channels, undercuts and interlocking assemblies that need support in the other families can be printed in one piece. Vat photopolymerisation removes seams instead, curing the part from liquid into one solid, which is why it holds the finest detail and the smoothest surface of any family while remaining brittle in impact. Material extrusion is the only family that builds with genuine engineering thermoplastics, and the only one where the bond between layers is a measurable weak point, at 50 to 75 percent of bulk strength. One scale, seven families. A part that needs better than 0.1 mm eliminates four of them immediately. How a prototype is routed to a family Bureaus do not start from the process list. They start from the geometry and work outwards, and the order of the four steps below is deliberate: the first two eliminate families on engineering grounds, and only then does price enter. Geometry first, price last. Reversing the order is how projects buy the wrong process twice. Read the geometry. Internal channels, lattice cores, undercuts, thin walls and part size each eliminate families before anything else is discussed. A part with an internal channel that must be free of support material is a powder bed part; a part 1.5 metres long is a material extrusion part or a directed energy deposition part. Name the requirement. Detail, strength, heat resistance, chemical resistance, electrical properties, appearance or unit volume - one of these usually dominates, and it points at a family rather than a machine. Match the family. The output of the first two steps is a shortlist of two families with an accuracy figure and a material attached to each, which is the point at which the decision becomes a comparison rather than a choice. Print and finish. Post-processing is part of the process, not an extra: support removal and ultraviolet cure for resin, depowdering and bead blasting for nylon, stress relief, support removal and often machining for metal. A route that does not name its finishing steps has not been costed properly. Where each family stops working Material extrusion: stops at fine detail, at fluid-tight parts and at loads crossing the layer direction. It is the cheapest and the fastest, and it is the wrong answer for a customer-facing appearance part. Vat photopolymerisation: stops at impact loading, at sustained ultraviolet exposure and at temperatures above roughly 60 to 90 degrees Celsius depending on the resin grade. Resins also shrink and can distort on thin, high-aspect geometry. Polymer powder bed fusion: stops at surface appearance and at fluid-tightness. As-built texture is matte and grainy, the parts are slightly porous, and reused powder changes crystallinity unless the refresh ratio is controlled. Metal powder bed fusion: stops at the build envelope, at geometry that needs no internal detail, and at cost. It also needs a post-processing route, because sealing faces, bores and threads are machined afterwards rather than printed to size. Material and binder jetting: stop at structural loading. Jetting gives colour, multi-material and smooth surfaces rather than strength, and binder jetted metal depends on a sintering step that shrinks the part and leaves it porous unless it is infiltrated. Directed energy deposition: stops at accuracy and finish. It builds large metal forms at a high deposition rate, which makes it a near-net-shape process that expects to be machined afterwards. None of them is a production process at volume. Above a few hundred parts the economics move to moulding, casting or stamping, and a prototype validated for one of these families is not validated for the process that will make the production units. Choosing a supplier by what it runs The practical test for a prototyping supplier is not the length of its machine list but whether the process that suits your part is under the same roof as the processes around it. A part that is printed and then needs a sealing face machined, a surface blasted, a thread cut or a dye applied should not have to travel between three companies, because the tolerance stack and the accountability both get lost in the post. Ask which families are run in house, which are subcontracted, what the accuracy figure is for each, and how the finishing route is costed. Send the model, the governing requirement and the quantity, and the process shortlist can be produced against your part rather than against a comparison chart. See SLA 3D printing for fine-detail resin parts, FDM 3D printing for economical large parts and SLM metal 3D printing for dense metal components, then send a file for a free engineering review. Send a model for a process shortlist Scope and sources. The seven-family grouping and the accuracy ranges were compiled in 2026 from an industry comparison of additive technology routes, which lists material extrusion at plus or minus 0.2 mm or coarser, vat photopolymerisation at plus or minus 0.05 to 0.1 mm, polymer powder bed fusion at plus or minus 0.1 to 0.2 mm, metal powder bed fusion at plus or minus 0.1 to 0.2 mm, material jetting at about plus or minus 0.1 mm, directed energy deposition at about plus or minus 1 mm and binder jetting at about plus or minus 0.2 mm, and from the SLA, SLS, FDM and SLM comparison published by Zorapid for the support requirement, minimum feature size and post-processing sequences of the four most common families. Surface roughness, interlayer bond strength at 50 to 75 percent of bulk strength, powder refresh effects and the material jetting and binder jetting limitations come from a rapid prototyping engineering guide and from a guide to resin and powder based 3D printing. These are typical published ranges for general-purpose industrial machines; geometry, orientation, material grade and post-processing move them, and the figures are planning context rather than a specification or a quotation. Nothing on this page states or implies a certification held by any supplier.

The short answer SLA and SLS are the two leading resin-and-powder additive processes. SLA cures liquid photopolymer with ultraviolet light, one thin layer at a time, and holds the finest detail, down to 25 micron layers. SLS sinters nylon powder with a laser and needs no supports, so it produces the strongest, most near-isotropic plastic parts. SLA: liquid resin cured by light SLA stands for stereolithography, and it was the first additive process to be commercialised, in the 1980s. The mechanism is photopolymerisation: a build platform sits one layer below the surface of a vat of liquid resin, and an ultraviolet light source draws the cross-section of the part across that surface, hardening the resin only where the light touches it. The platform then drops by one layer height, fresh resin flows across, and the next cross-section is drawn on top of the last. Two consequences follow directly. Because the part is built from a liquid and cured into one continuous solid, there is no seam between layers and no direction in which the material is weaker. And because the light source can be focused very finely, the process resolves features that no other method resolves: layers of 25 to 100 microns, features down to about 0.1 to 0.3 mm, and an as-built surface of Ra 1 to 5 micrometres, which is smoother than most machined finishes. The whole process is four repeating steps, then two that happen outside the machine. The cost of that precision is paid in three places. Supports are needed, because the part hangs from the platform and overhangs have nowhere else to stand, and removing them leaves contact marks that have to be sanded. Post-processing is mandatory, because uncured resin has to be washed off in solvent and the part then needs a final ultraviolet cure to reach its stated properties. And the material itself is brittle: a standard resin elongates only 4 to 6 percent before it breaks, which is why a resin part that looked perfect can crack the first time it is dropped. SLS: nylon powder fused by a laser SLS stands for selective laser sintering. Instead of a vat of liquid, the machine holds a bed of fine nylon powder. A roller spreads a layer roughly 80 to 150 microns thick across the bed, a laser traces the cross-section of the part and fuses those powder particles together, the bed drops, and the roller spreads the next layer. Nothing is glued and nothing is cured in a liquid; the particles are sintered into a continuous solid. The defining advantage is not the laser, it is the powder. The unfused powder around the part supports it, so the process needs no support structures at all. That single fact opens geometry the other processes cannot reach: internal channels, undercuts, living hinges, snap fits and interlocking assemblies printed in one piece, with nothing to cut away afterwards and no contact marks to sand. The bed is also why SLS is the only common additive process that nests many parts in a single build rather than printing them one at a time. The material is usually nylon, either PA12 or PA11, and that is what gives SLS its reputation. PA12 prints at roughly 45 to 50 MPa tensile with 15 to 20 percent elongation and a heat deflection temperature in the 90 to 170 degrees Celsius range depending on the grade; PA11 trades some strength for elongation that runs to 200 percent, which is why it is chosen for parts that flex repeatedly rather than parts that carry a steady load. Glass-filled grades push tensile strength to 70 to 85 MPa at the cost of elongation, which falls to 2 to 5 percent. The two side by side on the mechanics Same light source, different physics. The support row and the failure-mode row decide most parts. PointSLA, stereolithographySLS, selective laser sintering What is fusedLiquid photopolymer resinNylon powder, PA12 or PA11 Energy sourceUltraviolet lightLaser, carbon dioxide or fibre Layer height25 to 100 microns80 to 150 microns SupportsRequired, and they leave marksNone, the surrounding powder holds the part Post-processingSolvent wash, ultraviolet cure, support removalDepowdering and bead blasting Surface as builtSmooth, Ra 1 to 5 micronMatte, Ra 6 to 12 micron Tensile strength50 to 65 MPa standard, 55 to 65 MPa tough grade45 to 50 MPa for PA12, 70 to 85 MPa glass filled Failure modeBrittle under impact, degrades in ultraviolet lightSlightly porous, matte texture Typical useFine detail, dental and medical models, casting patternsFunctional nylon components, complex assemblies One row of that table is worth repeating because it catches buyers out: an SLS part is the stronger part, and it is not the smoother one. If the part has to be photographed or handled by a customer, SLA wins on appearance regardless of the strength argument. If the part has to flex, take an impact or carry a snap fit, SLS wins regardless of the appearance argument. Why SLS nylon is the tougher plastic Tensile strength alone does not predict which of these two parts survives a drop test, and the reason is elongation. A material that stretches 15 to 20 percent before it breaks absorbs energy that a material stretching 4 to 6 percent simply does not, so a nylon part deforms and recovers where a resin part cracks. The chart below makes the gap visible on one scale. Elongation, not tensile strength, is the number that predicts whether the prototype survives its first drop test. This is also why an SLS part is usually the right choice for anything that moves. Living hinges, snap fits, clips and press-fit features all rely on the material deforming without cracking, which is a property of nylon's elongation rather than of its strength. A resin part can be made to work in the same role with a tough or engineering resin grade, but the result is still a photopolymer under load, and it still changes outdoors. What neither one does well Neither is a production process. Both make prototype and low-volume quantities. Above a few hundred identical parts the economics move to injection moulding, casting or machining, and a part validated for one of these processes is not validated for the process that makes the production units. Neither holds machining tolerances. SLA holds about plus or minus 0.05 to 0.15 mm and SLS about plus or minus 0.1 to 0.3 mm, against plus or minus 0.01 to 0.05 mm for machining. Sealing faces, bearing bores and threads are machined afterwards, or the feature is redesigned. SLA parts change in daylight. Photopolymers discolour and embrittle under prolonged ultraviolet exposure unless the resin is chosen for outdoor service or the part is coated. SLS parts are porous and matte. As-built nylon is not fluid-tight without sealing, and its Ra 6 to 12 micrometres texture is wrong for a visible cosmetic surface. Neither produces a certified material lot. A printed part is not wrought stock and does not arrive with a mill certificate. Where a programme needs that traceability, it is a requirement on the process and it is confirmed per programme and per factory in writing before production. Neither replaces the design rules. Minimum wall thickness is roughly 0.6 to 0.8 mm for SLA and 0.7 to 1.0 mm for SLS, thin unsupported features distort, and powder parts must be designed so that trapped powder can be emptied out. How to choose, and what to send Two questions settle it. Does the part have to look finished, hold fine detail or serve as a casting pattern? That is SLA. Does the part have to flex, take an impact or contain internal geometry that cannot be supported? That is SLS. If the answer is both, the usual route is to split the part: print the visual and detail features in resin, and the loaded or moving features in nylon, then assemble and check fit before any tooling is committed. Send the model, the surface and feature requirement, the loads and the quantity, and the process recommendation comes back with an accuracy figure, a material and a finishing route attached to it. See SLA 3D printing for fine-detail resin parts, SLS 3D printing for durable nylon components and surface finishing for the post-processing that decides how either part finally looks. Send a model for an SLA or SLS recommendation Scope and sources. Mechanism, layer height and surface roughness descriptions were compiled in 2026 from the published capability tables of Modo Rapid (SLA layer 25 to 100 microns, tolerance plus or minus 0.05 to 0.15 mm, as-built Ra 1 to 5 microns, minimum wall 0.6 to 0.8 mm; SLS layer 80 to 150 microns, tolerance plus or minus 0.1 to 0.3 mm, as-built Ra 6 to 12 microns, minimum wall 0.8 to 1.0 mm) and from the SLA, SLS, FDM and SLM comparison published by Zorapid. Material properties come from a 3D printing materials comparison (SLA standard resin about 50 to 65 MPa and 4 to 6 percent elongation, tough resin about 55 to 65 MPa and 10 to 20 percent, PA12 SLS about 50 MPa, glass-filled nylon about 70 MPa) and from a study of loaded printed components that reports PA12 at 50 MPa with 20 percent elongation and PA11 at 45 MPa with elongation up to 200 percent. Machine and material cost context comes from a comparison of SLA and SLS in tooling. These are typical published ranges for general-purpose industrial machines; geometry, orientation, grade and post-processing all move them, and none of the figures is a specification for a particular build. Nothing on this page states or implies a certification held by any supplier.

The short answer 3D printing is used because it turns a design into a physical part in hours without tooling, so iteration is cheap and the geometry is unconstrained. It pays off under a few hundred parts, for designs still moving and for shapes machining cannot reach; above that, moulding or machining wins. Six reasons a part gets printed, and the figure behind each Nobody prints a part because printing is interesting. A project is printed because one of six pressures is present, and each pressure carries a number that can be checked before anything is committed. Read the chart below as a set of tests rather than as a list of benefits: if none of the third column applies to the part in front of you, the reason to print is probably not there yet. Six reasons, one line each. The first two are economic and auditable; the last four are engineering and supply reasons. Reason to printWhat it buysThe figure that decides Iteration speedA design change costs nothingFirst part in 2 to 24 hours No toolingNo mould and no minimum orderBreak-even 300 to 2,000 units Design freedomChannels, lattices, one-piece bodiesShapes a cutting tool cannot reach One-off economicsUnit cost flat from 1 to 500 partsRoughly the same price at any quantity CustomisationEvery part can be differentNo per-variant tooling penalty Supply continuityBridge stock and spare partsHours instead of weeks The first two rows are the ones a finance team can audit, and between them they account for most production decisions. The next four are the reasons a printed part often stays in service long after the prototype phase has ended. Reason one: speed, which in practice means iteration The number that matters is not how fast one part prints, it is how long a complete design loop takes. A printed part is available within 2 to 24 hours on a fused-deposition machine, and published bureau timings put resin printing at 1 to 3 days, SLS nylon at 2 to 3 days, industrial FDM at 2 to 4 days and metal powder-bed printing at 3 to 5 days. A change is an edited model and a new slice, so an iteration loop of one to five days is normal rather than exceptional. Set that against the alternative for a moulded plastic part. Injection moulding tooling takes 2 to 4 weeks in aluminium and 4 to 8 weeks in hardened steel, and it costs roughly USD 1,500 to 4,000 for a simple single-cavity aluminium tool and USD 8,000 to 40,000 for a production-grade steel tool. Every design change after that costs USD 1,000 to 5,000 or more, because the change is cut into steel. Three revisions before a design settles turns a fortnight of thinking into two months of waiting, and it is that deferral — not the price of the plastic — that printing removes. The loop is the product. A part that costs more per unit but arrives in a day wins when the design is still moving. This is also the reason the technology is used inside production rather than only before it. Jigs, fixtures, soft jaws and end-of-arm tooling are printed because the line cannot wait six weeks for a machined fixture, and because the fixture is usually revised twice in its first month of use. Reason two: there is no tool to pay for Printing removes a fixed cost and keeps a higher variable cost; moulding does the opposite. The comparison collapses into one line: break-even quantity equals tooling cost divided by the difference between the printed and the moulded unit cost. With an USD 8,000 tool, a printed part at USD 4 and a moulded part at USD 0.50, the crossover sits near 2,300 units. A 120 by 80 by 40 mm enclosure quoted with a USD 12,000 tool, printed at USD 18 to 25 each against USD 2 to 4 moulded, crosses over at roughly 600 to 750 units. Simple parts cross at 300 to 500; complex parts with expensive multi-cavity tools cross at 1,500 to 2,000. Below the bar, printing is cheaper in total. Above it, the tool has been paid for and moulding wins on unit price. Three costs sit outside that arithmetic and usually decide real projects: the working capital tied up in a minimum order of 500 or more parts, the inventory risk if demand falls short, and the revenue deferred by a six to ten week tooling wait. Counting the deferred launch revenue moves the practical crossover to the right of the chart above, and it is the reason that bridge production — printing parts while the mould is being cut — is now a standard launch strategy rather than a compromise. Reason three: geometry no cutting tool can reach The third reason is not economic at all. A machined manifold with cross-drilled internal passages needs plugs and O-rings, and every joint is a potential leak; the same manifold printed as a single body has no joints. Conformal cooling channels that follow the curve of a mould insert, lattice blocks that remove mass without removing stiffness, and housings that replace four bolted parts with one are all shapes a cutter cannot reach, or cannot reach at a price worth paying. The reason has a limit worth stating early: printing buys geometry, not precision. As-built metal surfaces land near Ra 8 to 15 micrometres, internal channels keep that roughness so flow resistance is higher than a honed bore, fine-pitch printed threads are unreliable, and a 6 mm H7 bore arrives undersized and has to be reamed. Design the printed part for the geometry gain, then machine the faces that carry a dimension. Reasons four to six: one-offs, customisation and supply Because the unit cost does not fall with volume, printing is a poor choice for a mass-produced part and an excellent one for everything else. One part and five hundred parts cost about the same each, so a variant, a spare or a one-off carries no tooling penalty. That is what turns printing into a service channel rather than only a prototype channel: spare parts for equipment whose tooling no longer exists, surgical guides and dental aligners that differ for every patient, and fixtures that exist in a handful of copies. The supply-chain reason is the newest and the fastest growing. A printed part produced while the production tool is still being cut lets a product launch before the mould is finished, and a distributed print network lets a repair part be produced near the point of demand rather than shipped from a central warehouse. Airbus now produces more than 25,000 flight-critical printed parts a year across three aircraft platforms, with a published 43 percent weight reduction and lead-time cuts of up to 85 percent on those parts, and a state-funded shared-printer network in Michigan had processed more than 51,000 print jobs by mid-2025. When printing is the wrong reason The material is not the production material. A part printed in nylon tells you how nylon behaves, not how the eventual moulded ABS or cast aluminium part will behave. For a test that has to predict production, machine the prototype from production stock instead. Tolerance is an order of magnitude away. Resin printing holds about plus or minus 0.05 to 0.15 mm, SLS about plus or minus 0.1 to 0.3 mm and FDM about plus or minus 0.2 to 0.5 mm, against plus or minus 0.01 to 0.05 mm for machining. Mating features and sealing faces are machined afterwards or redesigned. The surface has to be an appearance surface. Layer lines and as-built roughness are visible. A part judged by a customer needs sanding, vapour smoothing or a coat, and each of those is a cost the comparison above did not include. The load crosses the layers. Extruded parts carry only about 50 to 75 percent of their bulk strength across the layer boundary, so a printed bracket can be strong in the plane of the bed and weak through it. The unit cost never falls. Printing has no scale economy. At 10,000 units a moulded part may cost USD 0.30 while the printed equivalent still costs USD 3, and no amount of volume closes that gap. Traceability is the requirement. A printed part is not wrought stock and does not arrive with a mill certificate. Where a programme needs that paper, it is a requirement on the process, and it is confirmed per programme and per factory in writing before production. What to send if printability decides the project Send the CAD model as a STEP file if you have one, a toleranced drawing, the quantity and annual volume, the feature that matters most and the environment the part will see. Those five items settle the process before price is discussed: a bracket with a bending or impact load points to sintered nylon, the same geometry with an in-plane load and a tight budget points to filament printing, a fine-detail or appearance part points to resin, and a part that has to be metal points to powder-bed printing followed by machining. See FDM 3D printing for economical filament parts, SLS 3D printing for durable nylon components and SLA resin 3D printing for fine detail, then send a model for a free engineering review. Send a model and get a process recommendation Scope and sources. Process mechanics and layer-height ranges were compiled in 2026 from a 3D printing process guide by Premsa Industries (FDM layers 0.1 to 0.3 mm, infill 15 to 40 percent) and from the seven-category ISO/ASTM 52900 comparison by AS Prototypes (metal powder bed 20 to 50 microns, vat photopolymerisation 25 to 100 microns, material extrusion 100 to 300 microns, SLA tolerance plus or minus 0.05 to 0.1 mm, SLS tolerance plus or minus 0.2 to 0.3 mm, DMLS tolerance plus or minus 0.1 to 0.2 mm, SLS PA12 at 45 to 50 MPa, FDM polycarbonate at 55 to 70 MPa, DMLS Ti6Al4V at 1,050 to 1,150 MPa, bureau lead times of 1 to 5 days). Surface, thread and bore limits come from a machining-side process explainer (as-printed metal Ra 8 to 15 microns, wall thickness below about 1 mm becomes a process variable, H7 bores require reaming). Cost and break-even figures come from a 2026 tooling and per-part cost guide (aluminium tool USD 1,500 to 4,000 in 2 to 4 weeks, steel tool USD 8,000 to 40,000 in 4 to 8 weeks, FDM part USD 1 to 10, moulded part USD 0.10 to 2.00, worked break-even of about 2,286 units) and from a small-batch comparison (enclosure tooling USD 12,000, printed USD 18 to 25, moulded USD 2 to 4, break-even 600 to 750 units, design change USD 1,000 to 5,000). The Airbus figure of more than 25,000 flight-critical printed parts a year with a 43 percent weight reduction and up to 85 percent lead-time reduction, and the Michigan shared-printer network figure of more than 51,000 print jobs by mid-2025, are reported in a 2025 additive manufacturing review. These are typical published ranges for general-purpose machines and parameter sets, not a specification for any particular build; orientation, material grade, geometry and post-processing all move them. Nothing on this page states or implies a certification held by any supplier.

The short answer 3D printing, or additive manufacturing, builds a part by adding material one thin layer at a time from a digital model instead of cutting it out of a block. Layers run about 20 to 300 micrometres, the part is finished in hours, and the bond between layers is its weakest plane. From a digital model to a solid part: four stages The word printing is misleading, because nothing is printed onto paper. A solid model is cut into horizontal slices by software, and a machine then deposits, fuses or cures material one slice at a time until the slices stack into the finished shape. The four stages below are the same whether the machine costs USD 300 or USD 800,000. The printer is only the third of four stages, and it is often the least expensive one to change. Model. The part is designed in CAD and exported as a mesh (STL, 3MF) or as exact geometry (STEP). What matters is that the model is a closed solid with no gaps, no loose faces and the right units — a file exported in inches and read in millimetres comes out 25.4 times the intended size. Slice. Slicing software cuts the model into layers and calculates the path for each one. This is where layer height, wall count, infill, top and bottom skins, support strategy and orientation are set, and those five settings decide most of the part's strength, finish and build time. Build. The machine adds material layer by layer: a heated nozzle lays down molten thermoplastic, or a light source cures liquid resin, or a laser fuses powder. Each layer bonds to the one below it, and that bond is the part's weak plane. Finish. Supports are removed, powder or resin is cleaned off, and the part may be cured, sanded, smoothed, dyed, heat-treated or machined on critical faces. Post-processing is where as-built accuracy is corrected and where appearance is decided. Additive against subtractive, and why the difference matters CNC machining starts with a block or a bar and removes material until only the part is left; the removed material becomes chips. Additive manufacturing does the reverse. That single difference explains most of the strengths and most of the limits in one step, so it is worth putting the two side by side. Two rows decide most projects: the geometry row and the tolerance row, and they point in opposite directions. Point of differenceAdditive: 3D printingSubtractive: CNC machining How the part is madeMaterial added layer by layerMaterial removed from solid stock Material usedThe part and its supports onlyThe part, plus chips and offcuts Tooling neededNone: no mould and no fixtureFixtures and cutting tools Geometry reachInternal channels, lattices, one-piece bodiesA cutting tool must reach the face Typical tolerancePlus or minus 0.05 to 0.5 mmPlus or minus 0.01 to 0.05 mm Surface as builtRa 0.8 to 25 micronRa 0.4 to 3.2 micron Weak pointThe bond between layersNone introduced by the process Best quantity bandOne to a few hundred partsOne to hundreds of thousands Read the tolerance row first. Nothing printed off a standard industrial machine comes close to a machined fit, and the gap is roughly one order of magnitude. Then read the geometry row, which is where printing wins outright: a cutting tool has to reach every surface it produces, so internal passages, lattice cores and organic shapes are either impossible to machine or ruinously expensive to machine. What the machine is actually doing, family by family Seven process families are recognised under the ISO/ASTM 52900 framework, and five of them do almost all professional work. They differ in how material is joined, and the joining method sets the material list, the layer thickness and the achievable accuracy. Material extrusion A filament of thermoplastic is pushed through a heated nozzle that draws the cross-section bead by bead. Layers run 100 to 300 micrometres and tolerance is about plus or minus 0.2 to 0.5 mm. It is the only family that builds with genuine engineering thermoplastics such as ABS, polycarbonate and filled nylon, and it is the cheapest route to a large part. Its single weakness is the bond between beads. Vat photopolymerisation A platform sits just below the surface of a vat of liquid photopolymer, and light traces each cross-section and cures it. Layers run 25 to 100 micrometres and tolerance reaches plus or minus 0.05 to 0.1 mm, the tightest of the polymer routes, with as-built roughness as low as Ra 0.8 to 2.5 micrometres. Resin parts are precise and smooth but brittle, and they degrade under prolonged ultraviolet exposure. Powder bed fusion The best-known example is selective laser sintering of polymer powder at 80 to 150 micrometre layers; its metal equivalents fuse titanium, stainless steel, aluminium or nickel alloys at 20 to 60 micrometre layers. Unfused powder supports the part, so internal channels, hinges and interlocking assemblies can be built in one piece with no support marks. Sintered nylon PA12 reaches about 45 to 50 MPa tensile, and fully dense metal parts match wrought values. Material and binder jetting Droplets of photopolymer or binder are jetted onto a layer of powder or onto a build plate, and the part is cured or sintered afterwards. Tolerances are typically plus or minus 0.1 to 0.5 mm. The family is used for multi-colour and multi-durometer models, and for sand moulds and batch metal parts. Directed energy deposition Metal powder or wire is blown into a moving melt pool created by a laser, electron beam or arc, building layers from 500 to 2,000 micrometres. Accuracy is loose at plus or minus 0.5 to 1.5 mm, but the deposition rate is far higher than a powder bed, so the family is used for large structures, feature addition onto forgings and repair of worn parts. Where the accuracy comes from, and where it is lost Buyers usually ask about layer height first and it is the setting that explains the least. Layer height sets vertical resolution and surface finish; it does not make a part more accurate in the horizontal plane, and halving it roughly doubles build time. Real accuracy is a stack of five decisions, and the table below lists them in the order they are usually made. Only one of the five levers is on the machine. The other four are on the drawing. Post-processing is the lever most often left out of a quotation. A printed bore is undersized and has to be reamed; a printed sealing face is rough and has to be milled; a printed part carries residual stress that can move it after the build unless it is annealed. If a drawing has a real tolerance on it, assume a machining pass is part of the route. What 3D printing does not do It does not hold machining tolerances. Additive routes hold plus or minus 0.05 to 0.5 mm depending on the family, against plus or minus 0.01 to 0.05 mm on a machined part. Features that mate, seal or locate are machined afterwards. It is not isotropic. Extruded and jetted parts are weaker across the layer boundary, typically 50 to 75 percent of bulk strength. Powder-bed and vat-cured parts are close to isotropic, but the layer direction still affects surface and support marks. It does not scale economically. There is no tool to amortise, so the unit price stays roughly flat. Above a few hundred to a couple of thousand parts, moulding or casting wins on unit cost. It does not replace the design rules. Thin walls below about 1 mm become a process variable rather than a drawing callout, fine threads print unreliably, internal channels must be designed so that trapped powder can be emptied, and unsupported overhangs distort. It does not arrive with material certification. A printed part is not wrought stock and does not carry a mill certificate. Where a programme requires that traceability, it is a requirement on the process, confirmed per programme and per factory in writing before production. It is not one process. Choosing between extrusion, resin, polymer powder and metal powder changes the material, the tolerance, the surface and the price by more than most design decisions do. What to send for a printed part Send a STEP file if the model came from CAD, a toleranced drawing, the quantity, the feature that has to be accurate and the environment the part will work in. Those five items decide the family before price is quoted: fine detail and a smooth surface point to resin, a part that flexes or takes an impact points to sintered nylon, a large or inexpensive part points to filament extrusion, and anything that has to be metal points to powder-bed printing followed by machining. See SLA resin 3D printing for fine detail, SLS 3D printing for durable nylon parts and FDM 3D printing for economical large parts, then send a model for a free engineering review. Send a model and get the right print process Scope and sources. The four-stage build sequence, layer ranges and design levers were compiled in 2026 from a 3D printing process guide by Premsa Industries (FDM layers 0.1 to 0.3 mm, infill 15 to 40 percent, model export rules and the inches-versus-millimetres error) and from the ISO/ASTM 52900 seven-category comparison by AS Prototypes, which gives the layer, tolerance and material figures quoted by family (metal powder bed 20 to 50 microns at plus or minus 0.1 to 0.2 mm, vat photopolymerisation 25 to 100 microns at plus or minus 0.05 to 0.1 mm, material extrusion 100 to 300 microns at plus or minus 0.2 mm, binder jetting 50 to 100 microns at plus or minus 0.2 to 0.5 mm, directed energy deposition 500 to 2,000 microns at plus or minus 0.5 to 1.5 mm, SLS PA12 at 45 to 50 MPa, FDM polycarbonate at 55 to 70 MPa, DMLS Ti6Al4V at 1,050 to 1,150 MPa, as-built Ra 0.8 to 25 microns) and from a summary of polymer, resin and metal printing families. Practical limits — as-printed metal roughness around Ra 8 to 15 microns, wall thickness below about 1 mm behaving as a process variable, fine threads and H7 bores requiring a secondary machining pass, and residual stress moving a part after the build — come from a machining-side process explainer. The 50 to 75 percent interlayer bond-strength range is the figure quoted across the process guides above. These are typical published values for general-purpose industrial machines; orientation, material grade, geometry, parameter set and post-processing all move them, and none of the figures is a specification for a particular build. Nothing on this page states or implies a certification held by any supplier.