Adding instead of taking away
Every process in manufacturing processes needed something made in advance for each new shape: a fixture and a program, a pattern, a die, a mould. Additive manufacturing (AM, popularly 3D printing) needs none. It builds a part directly from its digital model, one thin layer at a time, so a new shape costs no tooling and a complicated shape costs little more than a simple one.
That changes what is worth making. A hip-implant cup can have a porous outer layer that bone grows into. A fuel nozzle that used to be brazed together from many small parts can be printed as one piece with internal passages no drill could reach; GE Aviation's printed cobalt–chrome nozzle tip for the CFM LEAP engine, in service since 2016, is the best-known case. A hospital can print a model of a patient's heart from a CT scan to plan an operation. A start-up can try ten bracket designs in a week.
But AM is not magic. It is slow per kilogram, its parts are often weaker in one direction than another, metal parts need careful post-processing, and the physics of melting and solidifying thousands of tiny pools of metal creates defects that conventional parts do not have. This lesson goes far beyond the Start level's first look: the process families, design for AM, defects and anisotropy, time and cost, professional use, and safety.
The process families
The international terminology standard, ISO/ASTM 52900, sorts AM processes into seven categories by how each layer is formed.
| Category | How a layer is made | Typical materials | Strengths | Limits |
|---|---|---|---|---|
| Material extrusion | A heated nozzle draws lines of molten thermoplastic (or a paste) | PLA, PETG, ABS, nylon, fibre-filled polymers, high-temperature PEEK | Cheap machines and materials, easy to use | Visible layers, anisotropic, slow for fine detail |
| Vat photopolymerisation | Light (a laser, projector or LCD screen) cures liquid resin | Acrylate and epoxy resins, dental and casting resins | Fine detail and smooth surfaces | Resins can be brittle and degrade in sunlight; needs washing and post-curing |
| Powder-bed fusion | A laser or electron beam melts or sinters selected areas of a thin powder layer | Nylons; titanium, nickel, aluminium and steel alloys | Strong, dense, complex parts, no supports for polymers | Expensive machines, slow, residual stress and supports in metals |
| Directed energy deposition | A laser, electron beam or arc melts wire or blown powder as it is deposited | Titanium, nickel alloys, steels | Large parts, repair of worn parts, high deposition rate | Coarse features, needs machining |
| Binder jetting | A print head jets a liquid binder onto a powder bed; the part is then sintered | Steels, sand for casting moulds, ceramics | Fast, no supports, good for batches | Shrinkage and distortion in sintering |
| Material jetting | Print heads jet droplets of photopolymer (or wax), cured layer by layer | Photopolymers, several at once | Multi-material and full colour, fine finish | Costly materials, weaker parts |
| Sheet lamination | Sheets are cut and bonded or welded together | Paper, metal foils | Cheap, embedded components possible | Limited shapes, rarely used |
The lesson concentrates on the first five, which cover most industrial use.
The digital chain
A printed part starts as a CAD model, exported as a mesh (STL, or the richer 3MF format) or, increasingly, passed directly. A slicer (open-source examples include PrusaSlicer and Cura; industrial machines have their own build processors) chooses the orientation, adds supports where needed, cuts the model into layers and plans the path of the nozzle or laser in each, then writes a machine program. For a desktop extrusion printer that program is G-code, much like the lathe program of the last lesson. This fragment is not run here:
(not run here: the start of a slicer's G-code for a desktop extrusion printer)
M104 S240 ; set the nozzle to 240 degC
M140 S80 ; set the bed to 80 degC
G28 ; home all axes
M109 S240 ; wait for the nozzle temperature
G1 Z0.2 F600 ; move to the first layer height, 0.2 mm
G1 X60 Y20 E9.5 F1800 ; draw a line, extruding 9.5 mm of filamentAfter the build come the steps that the brochures leave out: removing supports and powder, washing and post-curing resin parts, and for metals stress-relief heat treatment, cutting parts off the build plate, often hot isostatic pressing (HIP) to close internal pores, machining of fits and threads, surface finishing, and inspection, increasingly by X-ray computed tomography.
Design for additive manufacturing
Designing for AM means designing for the layer. The same CAD model can print well or badly depending on how it is oriented and how its features meet the rules of the process.
- Overhangs and supports. Each layer must rest on the one below. Surfaces that lean outward more than a critical angle, commonly about 45° from vertical for extrusion and metal powder-bed fusion, sag or curl and need supports, which cost material and time and leave marks. Holes printed horizontally can be made teardrop- or diamond-shaped so they support themselves. Polymer powder-bed fusion and binder jetting need no supports, because the loose powder holds the part.
- Orientation. It sets the support volume, the surface finish (the stair-step effect below), the build height and so the time, and the direction of weakness.
- Minimum features. Walls thinner than about two extrusion widths, pins thinner than a millimetre or so, and gaps narrower than the process can resolve are not reproduced reliably; each machine and material has its own limits.
- Escape holes. Closed hollows trap powder or resin, so they need holes to let it out.
- Allowances. Surfaces that must fit (bearing seats, sealing faces, threads) are printed oversize and machined, so the part needs datum features a machine can locate on.
- Exploit the freedom. Consolidate assemblies into one part; replace solid sections with lattices; follow load paths found by topology optimisation; route cooling channels along the surface of an injection-mould insert (conformal cooling), which shortens the cooling time computed in the last lesson.
Computing it: print time and cost by material extrusion
A material-extrusion printer deposits polymer at a volumetric rate equal to the layer height times the extrusion width times the speed, so print time is roughly the deposited volume divided by that rate, plus time for travel moves and layer changes. The program estimates the time and cost of a PETG bracket whose solid CAD model has a volume of 40 cm³, printed with three perimeters and 20% infill, and shows how layer height trades time against surface finish.
import numpy as np
# Material extrusion: time and cost of a PETG bracket on a desktop printer.
V_model = 40.0 # volume of the solid CAD model, cm^3
fill = 0.45 # fraction actually deposited (3 perimeters, 20 % infill, top and bottom skins)
height = 50.0 # mm
w, speed = 0.45, 60.0 # extrusion width (mm) and average print speed (mm/s)
travel = 0.20 # extra time for travel moves, retractions, slowing at corners
t_layer_change = 1.5 # s per layer
rho, price = 1.27, 25.0 # PETG density (g/cm^3) and filament price (EUR/kg)
printer_rate = 0.35 # EUR per printing hour: purchase spread over its life, plus upkeep
power, kwh_price = 0.12, 0.30 # kW while printing, EUR per kWh
labour = 10 / 60 * 30.0 # 10 minutes of set-up and part removal at EUR 30 per hour
V_dep = V_model * fill * 1000 # deposited volume, mm^3
mass = V_model * fill * rho
print(f"deposited {V_dep / 1000:.0f} cm^3 = {mass:.0f} g of PETG")
print("layer (mm) flow (mm^3/s) layers time (h) cusp on a 30-deg slope (mm) cost (EUR)")
for t in [0.10, 0.15, 0.20, 0.25, 0.30]:
flow = t * w * speed # volumetric deposition rate
layers = int(round(height / t))
hours = (V_dep / flow * (1 + travel) + layers * t_layer_change) / 3600
cusp = t * np.cos(np.radians(30)) # stair-step height normal to a surface 30 deg from horizontal
cost = mass / 1000 * price + hours * (printer_rate + power * kwh_price) + labour
print(f"{t:8.2f} {flow:10.2f} {layers:6d} {hours:7.2f} {cusp:10.3f} {cost:6.2f}")
t = 0.20
hours = (V_dep / (t * w * speed) * (1 + travel) + round(height / t) * t_layer_change) / 3600
parts = {"filament": mass / 1000 * price, "printer time": hours * printer_rate,
"electricity": hours * power * kwh_price, "labour": labour}
total = sum(parts.values())
print(f"\nat 0.20 mm layers, cost breakdown of EUR {total:.2f}:")
for k, v in parts.items():
print(f" {k:13s} {v:5.2f} ({v / total:4.0%})")The bracket uses 18 cm³ (23 g) of plastic. At 0.20 mm layers it prints in about 1.2 hours; halving the layer height to 0.10 mm doubles the time to 2.4 hours and halves the stair-step cusps on a 30° slope, from 0.17 to 0.09 mm. The cost, about EUR 6, barely depends on the layer height, because 83% of it is the ten minutes of a person's time to set up the job and remove the part; the filament is only 9%.
Anisotropy and defects
Why printed parts are anisotropic
A material-extrusion part is a stack of welded lines. Within a line the polymer is continuous; between lines and between layers it is only as strong as the bond formed when hot polymer met cooler polymer, and there are small voids where round lines meet. The result is anisotropy: strength along the layers is usually much higher than across them, and a part loaded so that its layers are pulled apart can fail at a fraction of the in-plane strength. Metal powder-bed fusion parts are far denser and closer to isotropic, but they have a columnar grain structure growing along the build direction, and their properties, especially elongation and fatigue, can still differ by direction.
The program estimates the strength of a printed polymer loaded at an angle to its layers with the Tsai–Hill criterion, a classical failure criterion for anisotropic materials that you will also meet with composites. The strengths are illustrative; real values depend on the material, the printer and its settings, and must be measured on specimens printed the same way as the part.
import numpy as np
# Strength of a material-extrusion part loaded at an angle to its layers (Tsai-Hill criterion).
# Illustrative strengths for a printed polymer, MPa:
X = 50.0 # tension along the layers (in the plane of printing)
Z = 30.0 # tension across the layers (along the build direction): limited by layer bonding
S = 25.0 # shear along a layer interface
def strength(theta_deg):
"""Uniaxial strength when the load makes angle theta with the layer plane."""
c, s = np.cos(np.radians(theta_deg)), np.sin(np.radians(theta_deg))
# stresses in layer axes per unit applied stress: along layers c^2, across layers s^2, shear s c
inv = (c**2 / X)**2 - (c**2 * s**2) / X**2 + (s**2 / Z)**2 + (s * c / S)**2
return 1 / np.sqrt(inv)
print("load angle to layers strength (MPa) relative")
for th in [0, 15, 30, 45, 60, 75, 90]:
print(f"{th:12d} deg {strength(th):12.1f} {strength(th) / X:9.0%}")
# A bracket whose peak stress is 20 MPa: safety factor in three build orientations
sigma = 20.0
for name, th in [("printed flat (load along layers)", 0), ("printed on edge, load at 45 deg", 45),
("printed upright (load across layers)", 90)]:
print(f"{name:38s} safety factor {strength(th) / sigma:4.2f}")The strength falls from 50 MPa along the layers to 38.4 MPa at 45° and 30 MPa across them, 60% of the in-plane value. For a bracket with a peak stress of 20 MPa, that is the difference between a safety factor of 2.50 printed flat and 1.50 printed upright: the same file, the same printer, a different part.
Defects in metal powder-bed fusion
In laser powder-bed fusion, a laser of a few hundred watts, focused to a spot of order 0.1 mm, scans at around a metre per second, making a melt pool a fraction of a millimetre across that solidifies in well under a millisecond. Millions of such tracks make a part. A first measure of how much energy each volume of powder receives is the volumetric energy density
for laser power , scan speed , hatch spacing between tracks and layer thickness . Too little energy leaves lack-of-fusion pores, irregular gaps where tracks or layers did not fuse, often with unmelted powder inside. Too much makes the melt pool boil and drill down into a deep, unstable keyhole that collapses and traps round pores. Between lies a process window, found by experiment for each alloy and machine. Other defects include gas porosity from the powder, spatter that lands on the bed, rough surfaces with partly fused powder particles, and residual stress from the steep temperature gradients, which can distort a part, crack it, or tear it off its supports; that is why metal parts are stress-relieved before removal from the plate.
Computing it: a metal build's time and cost
A powder-bed build has two clocks. The laser must scan the volume of every part, at a build rate per laser; and the recoater must spread every layer, whose number depends only on the build height. The second clock does not care how many parts are on the plate, which is the key to the economics.
import numpy as np
# Laser powder-bed fusion of a titanium alloy (Ti-6Al-4V) bracket. All costs illustrative (EUR).
def energy_density(P, v, h, t):
"""Volumetric energy density P / (v h t) in J/mm^3: power W, speed mm/s, hatch and layer mm."""
return P / (v * h * t)
window = (40.0, 90.0) # an illustrative good window for this alloy and machine, J/mm^3
print("power (W) speed (mm/s) E (J/mm^3) verdict")
for P, v in [(200, 1500), (280, 1200), (370, 700)]:
E = energy_density(P, v, 0.14, 0.03)
verdict = ("too little: lack-of-fusion pores" if E < window[0] else
"too much: keyhole pores" if E > window[1] else "inside the window")
print(f"{P:8d} {v:10d} {E:9.1f} {verdict}")
# Build time: scanning the part's volume, plus spreading a new powder layer every 30 um.
P, v, h, t = 280.0, 1200.0, 0.14, 0.03
rate = v * h * t # mm^3/s melted by one laser
V_part = 35_000.0 # part plus supports, mm^3
height = 60.0 # mm of build height
recoat = 8.0 # s to spread each layer
contour = 1.2 # contours and skins scan more slowly: +20 %
layers = int(height / t)
print(f"\nbuild rate {rate:.2f} mm^3/s per laser = {rate * 3.6:.1f} cm^3/h; {layers} layers")
machine_rate = {1: 100.0, 4: 160.0} # EUR per hour for a one-laser and a four-laser machine
gas_energy = 5.0 # EUR per hour of argon and electricity
powder = 300.0 * 4.43e-3 * V_part / 1000 # EUR/kg x density (g/mm^3 -> kg) x volume
post = 150.0 # per part: depowder, stress relief, cut off, remove supports, machine
print("\nparts lasers build (h) machine+gas per part powder post total per part")
for n_parts, lasers in [(1, 1), (12, 1), (12, 4)]:
scan_h = n_parts * V_part / rate * contour / lasers / 3600
recoat_h = layers * recoat / 3600
build_h = scan_h + recoat_h
run = build_h * (machine_rate[lasers] + gas_energy) / n_parts
total = run + powder + post
print(f"{n_parts:5d} {lasers:6d} {build_h:9.1f} {run:20.0f} {powder:6.0f} {post:4.0f} {total:10.0f}")The three parameter sets give 31.7, 55.6 and 125.9 J/mm³: one below, one inside and one above the illustrative window. With the good set, one laser melts 5.04 mm³/s, about 18 cm³ an hour, so this 35 cm³ bracket takes 6.8 hours alone, of which 4.4 hours is spreading its 2000 layers. Built alone it costs about EUR 906; twelve on the same plate share the recoating and cost EUR 478 each; on a four-laser machine, EUR 353 each. Post-processing, at EUR 150 a part, is then the largest single cost, which is typical of metal AM.
Metal AM in aerospace and medicine
Aerospace adopted metal AM first, because it values light weight and part consolidation and makes parts in thousands rather than millions. Beyond the LEAP fuel nozzle, printed parts as of 2026 include brackets and ducts on airliners, turbine and combustor components, satellite structures and antenna supports, and many rocket-engine parts: Rocket Lab's Rutherford engines have printed main components, SpaceX prints the combustion chambers of its SuperDraco thrusters, and several launch companies in Europe, China, India and the United States print injectors and chambers. Directed energy deposition with wire is used for large titanium structures and for repairing worn turbine blades. Qualification is the hard part: a flight part needs a qualified material, a qualified machine and process (fixed parameters, powder specification, heat treatment), and a qualified part, with inspection. Standards for this come from ASTM's F42 committee and ISO/TC 261 working jointly, from SAE for aerospace materials, and from agencies' own requirements, such as NASA's standard for additively manufactured spaceflight hardware.
Medicine uses AM for what it does best: one-offs and porous structures. Titanium acetabular cups and spinal cages with printed porous lattices let bone grow in; patient-specific implants for the skull, jaw and pelvis are designed from CT scans; surgical guides help surgeons cut exactly where they planned; and vat photopolymerisation prints dental models, aligner moulds, crowns and hearing-aid shells by the million. Regulators such as the US FDA, the European system of notified bodies under the Medical Device Regulation, and China's NMPA review printed devices, with guidance on the extra controls that printing needs: powder reuse, build-to-build variation, cleaning of lattices, and verification of each patient-specific design.
Machine makers span many countries: EOS, Nikon SLM Solutions and Trumpf in Germany, Renishaw in the United Kingdom, Colibrium Additive (formerly GE Additive), 3D Systems and HP in the United States, Stratasys in Israel and the United States, and BLT and Farsoon in China, with desktop printers from firms in the Czech Republic, China, the Netherlands and elsewhere. The frontier, picked up in advanced manufacturing, includes in-situ monitoring of every melt pool with cameras and photodiodes, machine learning on that data, multi-laser and multi-material machines, and printing of copper for electric motors and of refractory metals for fusion and space.
Safety with printers, resins and powders
Additive manufacturing looks clean, but each family has hazards, and each machine maker's manual and each material's safety data sheet say how to manage them.
- Desktop extrusion printers. The nozzle (often over 200 °C) and bed are hot enough to burn, and the moving axes can pinch fingers. Melting plastics releases ultrafine particles and volatile organic compounds, more from some materials (such as ABS) than others; makers recommend using printers in well-ventilated rooms, preferably in an enclosure with filtration, not running them in a bedroom, and never leaving one running unattended for long periods without the safety features its maker specifies.
- Resin printers. Uncured resins can irritate and sensitise skin and eyes, so makers specify nitrile gloves and eye protection when handling them, good ventilation, and washing parts in the solvent they recommend, often isopropyl alcohol, which is flammable and must be kept away from ignition sources. Waste resin and solvent need proper disposal, not the sink.
- Metal powders. Fine metal powders, especially of aluminium and titanium, are combustible dusts that can burn or explode if dispersed in air near an ignition source, and they are harmful if breathed in. Industrial machines therefore build under inert argon or nitrogen, which itself creates an asphyxiation hazard in enclosed spaces, so rooms have oxygen monitoring. Powder is handled in controlled areas by trained staff with respiratory protection, earthing against static, suitable fire extinguishing means (never water on a metal fire) and wet or inerted filters; regulations on explosive atmospheres and on combustible metals apply.
- Lasers and electron beams in industrial machines are fully enclosed and interlocked.
For learners, the safe and rewarding path is clear: design in CAD, slice and inspect the toolpaths and supports on screen, and print with a desktop extrusion printer used exactly as its maker intends, in a ventilated space, with an adult or a supervisor in a school or makerspace. Resin printing belongs in a properly set-up room with protective equipment; metal powder processes belong only in industrial and research facilities.
A short history
Hideo Kodama at the Nagoya Municipal Industrial Research Institute in Japan published a layer-by-layer photopolymer method in 1981. Charles Hull patented stereolithography in the United States in 1986 and founded 3D Systems; Carl Deckard and Joseph Beaman at the University of Texas at Austin developed selective laser sintering in the late 1980s; Scott Crump filed a patent on fused deposition modelling in 1989 and founded Stratasys; and Emanuel Sachs's group at MIT developed binder jetting, which they called three-dimensional printing, in the early 1990s. Metal powder-bed fusion grew from work at the Fraunhofer Institute for Laser Technology in Aachen and elsewhere in the 1990s. In 2005 Adrian Bowyer at the University of Bath started the open-source RepRap project to build a printer that could print many of its own parts; when key extrusion patents expired around 2009, RepRap's descendants made desktop printers cheap enough for schools and homes.
Common misconceptions
Misconception
"3D printing will replace conventional manufacturing." For large quantities of simple parts, moulding, casting, pressing and machining are far faster and cheaper. AM wins on geometry, individuality and small numbers, and increasingly works alongside them, for example printing mould inserts with conformal cooling.
Misconception
"A printed part is as strong as the material's datasheet says." Datasheet values come from specimens printed and tested in particular conditions. Orientation, settings and defects can change strength, and especially fatigue strength, a great deal; critical parts are designed with values measured for the actual process.
Misconception
"Complexity is free." Geometric complexity costs little, but support removal, depowdering of fine lattices, inspection of hidden features and post-machining all cost money, and some internal features cannot be inspected at all without CT.
Misconception
"Metal parts come off the printer finished." They come off attached to a plate, full of residual stress and supports, with rough surfaces. Heat treatment, HIP, support removal, machining and inspection are part of the process and often most of its cost.
Exercises
1. Additive manufacturing
Warm-upHumanA dental laboratory needs dozens of models of patients' teeth every day, with fine detail and smooth surfaces, to form clear aligners on. Which process family suits it best?
2. Additive manufacturing
CoreHumanWhich of these processes can build overhanging features without separate support structures, because the part is surrounded by unfused material as it is built? Choose all that apply.
3. Additive manufacturing
Warm-upHumanA material-extrusion printer must deposit 30 cm³ of plastic with 0.20 mm layers, 0.50 mm extrusion width and an average speed of 50 mm/s. Ignoring travel moves and layer changes, how many minutes of printing is that?
4. Additive manufacturing
CoreHumanA laser powder-bed fusion machine runs at 300 W, a scan speed of 1000 mm/s, a hatch spacing of 0.10 mm and a layer thickness of 0.04 mm. What is the volumetric energy density, in J/mm³?
5. Additive manufacturing
CoreHumanA metal build is 90 mm tall with 30 µm layers, and spreading each powder layer takes 9 s. How many minutes of the build are spent recoating, however many parts are on the plate?
6. Additive manufacturing
CoreHumanA flat hook, 6 mm thick, will carry a load that bends it in its own plane, like a coat hook on a wall. It is to be printed by material extrusion. Which orientation is likely to make it strongest?
7. Additive manufacturing
CoreComputationalWrite print_time_hours(volume, layer, width, speed, height, travel=0.2, layer_change=1.5) that returns the estimated print time in hours for a material-extrusion print, using the lesson's model:
- deposition time =
volume(mm³) / (layer×width×speed), all in mm and mm/s, - multiplied by
1 + travelfor travel moves, - plus
layer_changeseconds for each of theround(height / layer)layers.
Raise ValueError if any of volume, layer, width, speed or height is not positive.
8. Additive manufacturing
StretchComputationalWrite tsai_hill_strength(theta, X, Z, S) that returns the uniaxial strength of a layered (transversely isotropic) material loaded at angle theta (in degrees) to its layer plane, by the Tsai–Hill criterion used in the lesson:
where X is the strength along the layers, Z across them and S the interlayer shear strength.
Further reading and resources
- ASTM InternationalIts Committee F42 develops additive-manufacturing standards jointly with ISO/TC 261, including ISO/ASTM 52900 on terminology and process categories.
- ISOISO/TC 261 is the international technical committee for additive manufacturing.
- America MakesThe US national institute for additive manufacturing, with public road maps and project reports.
- Additive manufacturing of metallic components – Process, structure and properties · T. DebRoy and others · 2018A long, widely cited review of metal AM physics, microstructure, defects and properties (Progress in Materials Science 92).
- RepRapThe open-source self-replicating printer project that began the desktop 3D-printing movement.
- PrusaSlicerAn open-source slicer for material-extrusion and resin printers.
- Ultimaker CuraAnother widely used open-source slicer.
- U.S. Food and Drug AdministrationSearch for its guidance on technical considerations for additive manufactured medical devices.
- Health and Safety Executive (UK)Guidance on combustible dusts, explosive atmospheres and fume control, relevant to metal powders and printing.
- Additive Manufacturing Technologies · Ian Gibson, David Rosen, Brent Stucker and Mahyar KhorasaniA standard textbook covering every process family, design for AM and post-processing.