Exploring 3D Printing: How Each Process Works and Where It Stops

• 1
  Key takeaways
• 2
  What actually happens inside an FDM, SLA or SLS build
• 3
  Where the tolerance budget goes on a printed part
• 4
  Build orientation changes strength, not just looks
• 5
  When 3D printing stops being the cheap option
• 6
  Design rules that survive contact with a real build
• 7
  Material behavior matters more than the material name
• 8
  From model to first article in six steps
• 9
  Comparing additive processes against CNC for functional parts

FDM pushes a thermoplastic filament through a heated nozzle and lays down a bead, usually 0.1–0.3 mm tall. The bead is round in section, so adjacent roads only touch along a line. That contact area is why a part printed in the XY plane is noticeably stronger than the same part loaded across the layer stack. On a 0.4 mm nozzle you can hold roughly a 0.8 mm wall as a practical minimum, and anything thinner prints as a wavy ribbon.

SLA and DLP cure liquid resin with a laser or a projector. Because the resin is cured rather than melted and re-solidified, surface texture comes out smooth with visible layer lines only at shallow angles. Minimum feature size is set by the laser spot or pixel pitch, not the layer height. The catch is green-part handling: the part needs washing and a full UV post-cure, and it keeps shrinking slightly during that cure.

SLS sinters polymer powder with an infrared source while a roller spreads a fresh layer. Unsintered powder surrounds the part and acts as support, so you can nest parts in a tall build and skip support-removal labor entirely. The trade-off is a grainy surface and internal porosity that shows up after dyeing. Powder reuse ratio matters more than most engineers expect: too much old powder and elongation drops.

Metal AM splits into powder bed fusion and directed energy deposition. Powder bed fusion builds in an inert chamber with a laser or electron beam, then the part is cut off the plate and stress-relieved. DED deposits metal through a nozzle and is closer to cladding than to printing, which suits repair and near-net preforms. Both routes usually finish with machining on every functional surface.

• 1
  Nozzle diameter sets the floor for FDM walls0.4 mm nozzle, 0.8 mm minimum wall, 0.1–0.3 mm layers.
• 2
  Spot size sets the floor for SLA featuresSmall holes print undersized and need a drill pass.
• 3
  Powder bed needs no support but traps powderBlind channels below 1 mm are hard to clear.
• 4
  Metal AM always ends with a machining operationInterfaces, bores and threads are cut, not printed.

Every additive process shrinks. FDM filament contracts as it cools, SLS powder sinters to a slightly smaller volume, and metal powder bed fusion pulls as the melt pool solidifies. Machine builders compensate with a global scale factor, but the correction is uniform while the shrinkage is not. A long thin part warps more at the ends than the middle, so the deviation is not a single number you can dial out.

Support removal is the second big contributor. On FDM, the support interface leaves a rougher surface than the upward-facing skin, and that roughness can be 2–3× the rest of the part. On SLA, support nubs leave small witness marks that need sanding before a cosmetic finish. Design for a flat base and self-supporting angles under 45° from vertical and you remove most of this error before it exists.

Layer height is often mistaken for accuracy. It is not. Layer height controls the stair-step appearance of sloped surfaces and the number of build hours, not the positional accuracy of a hole center. A 0.05 mm layer and a 0.2 mm layer on the same machine will hold the same linear tolerance, but the surface on the shallow face will look very different. Choose layer height for cosmetics and speed, then pick a process for accuracy.

For reference, our CNC work holds ±0.005 mm (±0.0002 in) and finishes down to Ra 0.2–0.8 μm when the drawing calls for it. That gap is not a criticism of printing. It is the reason printed parts and machined parts usually end up in the same assembly, doing different jobs.

A printed part has a grain, and the grain runs along the layer stack. In FDM, the bond between layers is a thermal weld, weaker than the filament itself. Load a bracket across the layers and it can fail at a fraction of the in-plane strength. Rotate the same part 90° in the slicer and the weak plane moves somewhere harmless. Orientation is a structural decision made in the build setup, often by someone who never saw the load case.

That is why print orientation belongs on the drawing or in the build notes. Mark the load direction, mark any sealing face, and mark which surfaces will be machined after printing. Without that, a supplier will orient the part for the shortest build time, which is usually the worst orientation for strength.

SLS behaves better because the powder bed holds the part and the polymer chains diffuse across layer boundaries. Anisotropy is still present, but the ratio is far smaller than FDM. SLA sits in the middle and depends heavily on post-cure: under-cured resin stays soft, over-cured resin gets brittle. There is a window, and it is process-specific.

For metal parts, orientation decides more than strength. It sets residual stress direction, decides which faces need support, and dictates how much stock you leave for the finishing cut. Leave 0.3–0.5 mm on any face that will be machined, and leave more on the top surface if the part is tall.

Additive wins on complexity, not on volume. One part with internal channels costs roughly the same to print as one simple block of the same bounding box, because the machine time is driven by height and layer count, not by how many pockets the part has. That is the whole economic argument, and it holds up to a point.

The crossover comes when you add finishing operations. Once a printed part needs facing, boring, tapping and a tight flatness callout, you have bought a printing run plus a machining run plus a setup. At that stage a machined part from 6061 or 17-4PH often lands cheaper and arrives with the tolerances already in the part rather than bolted on afterward.

Quantity pushes the same crossover earlier. Printing 200 identical small brackets means 200 separate builds or one very tall nested build, with per-part labor for depowdering and inspection. CNC machining amortizes a fixture across the whole run and holds ±0.005 mm on every piece. Above a few dozen units, the arithmetic usually flips.

The honest answer for most hardware teams is a split. Print the geometry that cannot be machined, machine the faces that must seal, thread or slide. Send both to one supplier and you skip the tolerance stack argument between two vendors.

Keep wall thickness uniform. A sudden jump from 2 mm to 6 mm gives the thick section more thermal mass, so it cools slower and pulls the thin section with it. If a boss has to be thick, hollow it or core it out from the back. This single rule prevents more warped parts than any slicer setting.

Design holes undersized and ream them. FDM and SLS both print holes smaller than nominal because the material shrinks inward and the slicer compensates with a circular approximation. A Ø6 H7 bore is a machining operation, not a printing one. Print it at Ø5.6 and drill it after, or design the hole as a slot if the fastener only needs clearance.

Leave a machining allowance where it matters. Sealing faces, bearing seats, thread bosses and any datum used for inspection should carry 0.3–0.5 mm of stock. Threads below M4 should be printed as pilot holes and tapped afterward; printed threads strip under torque and there is no reliable way to inspect them.

Finally, think about how the part will be held. A printed part with no flat face and no parallel surfaces is hard to fixture for the finishing cut. Add a temporary tab or a flat pad, machine the datums, then remove the tab. It costs almost nothing in the model and saves a setup on the floor.

Two suppliers can both say PA12 and give you different parts. Powder age, moisture content and machine calibration change elongation and surface finish more than the grade name does. When a printed part has to survive a drop test, ask for the build parameters, not just the resin or powder designation.

Moisture is the quiet failure mode for FDM. Wet filament outgasses in the nozzle and leaves voids and a rough surface. Nylon absorbs fast; dried and sealed filament prints visibly better. If a supplier stores filament in the open, expect variation between the first and the last part in a run.

For metal, the alloy family sets the post-processing route. Ti-6Al-4V needs stress relief before it is cut off the plate, or the part will bow during the wire EDM operation. 17-4PH responds to heat treatment, so the sequence of print, stress relief, machine and age has to be planned, not improvised.

We keep 6061, 6061-T6, 2024, 5052, 6063, 6082, 7075, 303, 304, 316L, 17-4PH, 4140, TC4 and Inconel in the machining queue, plus ABS, PC, POM, PA, PEEK and carbon-filled plastics for printed parts. Mixing both routes in one assembly is normal here.