One part, many ways to make it
Take a small aluminium housing, the size of your fist, with a bore for a bearing and four mounting holes. A prototype workshop would machine it from a solid bar in an afternoon. A small firm making a few hundred might sand-cast it and machine the bore. A car-parts supplier making half a million a year would die-cast it, in a steel die costing as much as a house, and might replace it with a glass-filled plastic moulding if the loads allow. Each answer is right for its quantity, and each changes the shape: a die-cast part has thin, even walls and draft angles; a machined part has square pockets the size of the cutter.
How things are made sorted processes into four families. This lesson takes the main ones apart: the numbers that govern machining, the heat flow that limits casting and moulding, the mechanics of forming and joining, a cost model for choosing between them, and the statistics that tell you whether a process can hold the tolerances the drawing asks for. Additive manufacturing has the next lesson to itself.
Machining: speeds, feeds and tool life
Machining removes material as chips with a hard cutting edge. On a lathe the work turns and a single-point tool moves along it, making round parts; on a milling machine a rotating multi-edged cutter moves across the work, making flat faces, slots and pockets; drilling, boring, reaming and tapping make and finish holes; grinding uses abrasive grains for fine finishes and hard materials. Almost all production machining is now CNC, and a machining centre changes its own tools from a magazine.
Definition— cutting speed, feed and material removal rate
The cutting speed is the speed of the work surface past the edge, (with the work diameter in turning, the cutter diameter in milling). In turning, the feed is the tool's advance per revolution and the depth of cut the radial depth; the material removal rate is . In milling, the feed per tooth gives the table feed for a cutter with teeth, and . The cutting power is roughly , where the specific cutting energy is a property of the work material: of order 0.5–1 J/mm³ for aluminium alloys and 2–5 J/mm³ for carbon and alloy steels.
Cutting speed is chosen for the tool, not the machine. Tool makers publish recommended speeds and feeds for each tool grade and work material. Faster cutting removes metal sooner but wears the edge out much faster, by abrasion, diffusion and chemical attack at temperatures that reach several hundred degrees at the edge. F. W. Taylor, in years of experiments at Midvale and Bethlehem Steel in Pennsylvania published in 1907, found the relation
where is the tool life in minutes and the exponent is about 0.1–0.15 for high-speed steel and 0.2–0.3 for carbide. With , doubling the speed cuts tool life sixteen-fold. Between "slow and cheap on tools" and "fast and cheap on machine time" lies a speed of minimum cost per part.
import numpy as np
# Turning one pass along a steel shaft with a carbide insert.
D, L = 60.0, 200.0 # workpiece diameter and length of cut, mm
vc = 180.0 # cutting speed, m/min
f, ap = 0.25, 2.0 # feed (mm/rev) and depth of cut (mm)
u = 2.5 # specific cutting energy of a medium-carbon steel, J/mm^3 (W s/mm^3)
n = 1000 * vc / (np.pi * D) # spindle speed, rev/min
vf = f * n # feed rate, mm/min
t = L / vf # time for the pass, min
MRR = vc * 1000 * f * ap # material removal rate, mm^3/min
print(f"turning: spindle {n:.0f} rev/min, feed rate {vf:.0f} mm/min, pass time {t * 60:.0f} s")
print(f" material removal rate {MRR / 1000:.0f} cm^3/min, cutting power {u * MRR / 60 / 1000:.2f} kW")
# Face milling: a 63 mm cutter with 5 inserts takes a 40 mm wide, 1.5 mm deep cut.
Dc, z, fz, vc_m = 63.0, 5, 0.12, 200.0
ae, ap_m = 40.0, 1.5
n_m = 1000 * vc_m / (np.pi * Dc)
vf_m = fz * z * n_m
print(f"milling: spindle {n_m:.0f} rev/min, table feed {vf_m:.0f} mm/min, "
f"MRR {ae * ap_m * vf_m / 1000:.1f} cm^3/min, power {u * ae * ap_m * vf_m / 60 / 1000:.2f} kW")
# Tool life (Taylor): v T^k = C. Faster cutting wears the tool out much sooner.
k, C = 0.25, 450.0 # illustrative constants for carbide on this steel (v in m/min, T in min)
def tool_life(v):
return (C / v) ** (1 / k)
# Cost per part for one finishing pass at speed v (costs in EUR, illustrative)
Cm = 1.5 # machine and operator, EUR per minute
Ct = 8.0 # cost of one cutting edge, EUR
t_ch, t_load = 2.0, 1.0 # minutes to change an edge, and to load and unload a part
def cost(v):
tm = np.pi * D * L / (1000 * v * f) # cutting time per part, min
return Cm * (tm + t_load) + (Cm * t_ch + Ct) * tm / tool_life(v)
print("\n v (m/min) tool life (min) parts per edge cost per part (EUR)")
for v in [100, 150, 200, 250, 300, 400]:
tm = np.pi * D * L / (1000 * v * f)
print(f" {v:8d} {tool_life(v):12.1f} {tool_life(v) / tm:12.1f} {cost(v):10.2f}")
vs = np.linspace(60, 450, 3901)
best = vs[np.argmin(cost(vs))]
T_opt = (1 / k - 1) * (t_ch + Ct / Cm)
print(f"cheapest speed found by scanning: {best:.0f} m/min (tool life {tool_life(best):.0f} min)")
print(f"minimum-cost formula: T_opt = (1/k - 1)(t_ch + Ct/Cm) = {T_opt:.1f} min -> v = {C / T_opt**k:.0f} m/min")At 180 m/min the 60 mm shaft turns at 955 rev/min and one 200 mm pass takes 50 s, removing 90 cm³ of steel a minute with 3.75 kW at the edge. The face mill removes 36.4 cm³/min with 1.52 kW. The cost table falls from EUR 3.80 per part at 100 m/min (where the edge would last 410 minutes) to a minimum of about EUR 2.95 near 200 m/min, and rises again as edges wear out in minutes. The scan and the classical minimum-cost formula agree: the cheapest speed is 208 m/min, where each edge lasts 22 minutes.
A CNC machine is programmed in G-code, a line-by-line language of moves and machine functions, usually written by CAM software from the CAD model: FreeCAD's CAM workbench, or commercial packages such as Mastercam, Siemens NX CAM, hyperMILL, Fusion and many others. Here is the turning pass above as a program for a lathe controller. It is not run here, and dialects differ between controller makers:
(not run here: G-code for a CNC lathe, shown for reading)
G21 G18 G90 G99 (millimetres, XZ plane, absolute positions, feed per revolution)
G50 S3000 (never exceed 3000 rev/min)
G96 S180 M03 (constant cutting speed 180 m/min, spindle on)
G00 X62.0 Z2.0 (rapid move to just clear of the work)
G01 X56.0 F0.25 (feed in: diameter 56 mm, 2 mm depth of cut)
G01 Z-200.0 F0.25 (turn 200 mm along the axis at 0.25 mm/rev)
G00 X62.0 (retract)
G00 Z2.0
M05 (spindle off)Casting, forming and sheet metal
Casting
Casting pours molten metal into a mould cavity shaped like the part. Sand casting packs sand around a pattern and is cheap in tooling, suitable for anything from a valve body to a ship's propeller. Investment casting coats a wax pattern in ceramic, melts the wax out and pours into the shell, giving fine detail and good finish; it is how turbine blades and many medical implants are made. High-pressure die casting injects aluminium, magnesium or zinc into a steel die in a fraction of a second, producing thin-walled housings by the hundred thousand.
The physics is heat flow. Metal shrinks as it solidifies, so the last part to freeze must be fed with liquid from a reservoir, a riser, or it will contain a shrinkage cavity. Nicolae Chvorinov, studying castings in Czechoslovakia in 1940, found that solidification time grows with the square of the casting's modulus, its volume-to-surface ratio: , where depends on the metal and the mould. A riser works only if it freezes after the casting it feeds.
Forming and forging
Bulk forming reshapes solid metal by plastic deformation. Rolling makes plate, sheet and rails; extrusion pushes billets through a die to make long profiles, like aluminium window frames; forging squeezes or hammers a billet, usually hot, between dies. Forging makes the strongest and toughest parts (crankshafts, connecting rods, aircraft landing-gear and turbine discs) because it closes internal porosity and aligns the metal's grain flow along the part's outline. Cold working at room temperature gives better finish and accuracy and strengthens the metal by work hardening, the process explained in elasticity, plasticity and strengthening, at the price of higher forces.
Sheet metal
Sheet-metal parts, from a car's door to a computer's case, are blanked (sheared to an outline), pierced, bent and drawn (stretched into a cup or panel) in presses. Bending must allow for springback: the elastic part of the strain recovers when the punch lifts, so dies overbend. The blank for a bent part is calculated from the length of its neutral axis, which in a bend lies a fraction (typically 0.3–0.5) of the thickness from the inside surface. Laser and punch-press cutting with CNC press brakes make sheet metal one of the cheapest ways to produce accurate enclosures in small batches as well as large ones.
Welding, joining and injection moulding
Welding and joining
Fusion welding melts the edges of two parts, usually with filler metal, so they solidify as one. Arc welding (manual metal arc, MIG/MAG with a wire fed through the torch, TIG with a tungsten electrode) is the workhorse of steel structures, ships and pipelines; resistance spot welding joins car-body panels, several thousand welds per car, in a fraction of a second each; laser and electron-beam welding make deep, narrow welds with little distortion. Every fusion weld has a heat-affected zone whose microstructure, and so strength and toughness, differ from the parent metal, and welds shrink as they cool, distorting the parts and leaving residual stresses. Welds are also where most fatigue cracks in steel structures begin, which is why the fatigue rules for welded joints in design codes are so conservative.
Solid-state methods avoid melting. Friction stir welding, invented at TWI in Cambridge in 1991, stirs the softened metals together with a spinning tool and joins aluminium alloys that are hard to fusion-weld; it is used for rocket fuel tanks, rail carriages and ship panels. Brazing and soldering use a filler that melts at a lower temperature than the parts, adhesive bonding joins dissimilar materials and spreads load over large areas, and mechanical fasteners (bolts, rivets, clinches) allow disassembly and repair.
Injection moulding
Most plastic parts are injection moulded: plastic pellets are melted by a heated screw, injected into a closed steel mould under high pressure, held under pressure to pack in material as it shrinks, cooled until rigid, and ejected. A cycle takes from a few seconds to a minute, and one mould may make millions of parts. The cycle is dominated by cooling, which is again heat conduction: for a wall of thickness cooled from both faces, solving the heat equation of heat transfer and keeping the first term gives
with the plastic's thermal diffusivity. The design rules follow from the physics: keep walls thin and uniform (thick sections cool slowly and sink), add ribs for stiffness instead of thickness, give every wall a draft angle so the part releases, and round the corners so the melt flows.
import numpy as np
# 1. Injection moulding: the part must cool until its centre is below the ejection temperature.
# For a wall of thickness s cooled from both faces (first term of the slab conduction solution):
# t_c = s^2 / (pi^2 alpha) * ln( (4/pi) (T_melt - T_mould) / (T_eject - T_mould) )
alpha = 0.09 # thermal diffusivity of ABS, mm^2/s
T_melt, T_mould, T_eject = 240.0, 60.0, 95.0
print("wall (mm) cooling time (s)")
for s in [1.0, 1.5, 2.0, 3.0, 4.0]:
tc = s**2 / (np.pi**2 * alpha) * np.log(4 / np.pi * (T_melt - T_mould) / (T_eject - T_mould))
print(f"{s:7.1f} {tc:8.1f}")
# 2. Sand casting: Chvorinov's rule, solidification time t = B (V/A)^2.
B = 2.5 # mould constant, s/mm^2 (illustrative: measured for each metal and mould)
def modulus_box(a, b, c):
return a * b * c / (2 * (a * b + b * c + a * c))
M_plate = modulus_box(200, 100, 20) # a 200 x 100 x 20 mm aluminium plate
side = 400_000 ** (1 / 3) # a cube with the same volume, 400 cm^3
M_cube = modulus_box(side, side, side)
print(f"\nplate 200x100x20 mm: V/A = {M_plate:.2f} mm, freezes in about {B * M_plate**2:.0f} s")
print(f"cube of the same volume ({side:.1f} mm side): V/A = {M_cube:.2f} mm, about {B * M_cube**2:.0f} s")
# A cylindrical riser (height = diameter, V/A = D/6) must freeze after the casting it feeds:
# rule of thumb, riser modulus at least 1.2 x casting modulus.
D_riser = 6 * 1.2 * M_plate
print(f"riser: D = H >= {D_riser:.1f} mm, freezes in about {B * (D_riser / 6)**2:.0f} s")A 2 mm wall of ABS needs about 8.5 s to cool; a 4 mm wall needs 33.8 s, four times as long, because the time grows with the square of the thickness. Making a part's walls 2 mm instead of 4 mm cuts the cooling time, the largest part of the cycle, by three quarters, and the machine cost per part with it. The aluminium plate's modulus is 7.69 mm and it freezes in about 148 s; a cube of the same volume, with less surface, takes 377 s. A riser for the plate must be at least 55.4 mm in diameter and height to stay liquid longer (213 s).
Choosing a process: cost per part
Start with what the part needs. Shape rules out many processes (a hollow, re-entrant shape cannot be pressed from sheet); material rules out others (titanium is not die-cast; high-strength steel is not injection moulded); size, tolerance and surface finish narrow the field further. Michael Ashby's process-selection charts, which plot processes against mass, section thickness, tolerance and economic batch size, make this screening quick. Among the survivors, cost decides. A useful model for the cost of each part is
where is the number of parts and how many parts one die or mould can make before it is worn out. Tooling dominates small batches; material and cycle time dominate large ones. The program applies it to the aluminium housing from the motivation, with illustrative figures of a realistic size, and to a glass-filled polyamide moulding of the same part, in case the plastic meets the requirements.
import numpy as np
# Cost per part of a small housing (about 0.25 kg finished), made five ways.
# All figures are illustrative (EUR), of a realistic size, not quotes from any supplier.
# tooling: patterns, dies, moulds, fixtures and programming, paid once per tool
# tool_life: parts one tool can make before it must be replaced
# material, process, finishing: cost per part of material (with scrap), of the forming
# or cutting step (machine-hour rate x cycle time), and of secondary machining
processes = {
# tooling tool_life material process finishing
"CNC from aluminium bar": ( 400, 10**9, 3.00, 21.60, 0.00),
"sand casting": ( 1500, 5_000, 1.20, 4.00, 7.20),
"investment casting": ( 6000, 50_000, 1.50, 9.00, 2.40),
"high-pressure die cast": (40000, 100_000, 0.80, 2.00, 2.70),
"injection moulded PA-GF":(25000, 500_000, 0.40, 0.70, 0.00),
}
def cost_per_part(name, n):
tooling, life, mat, proc, fin = processes[name]
tools_needed = int(np.ceil(n / life)) # a worn-out die must be replaced
return tooling * tools_needed / n + mat + proc + fin
quantities = [10, 100, 1_000, 10_000, 100_000, 1_000_000]
print("parts made " + "".join(f"{q:>11,}" for q in quantities))
for name in processes:
print(f"{name:24s}" + "".join(f"{cost_per_part(name, q):11.2f}" for q in quantities))
print("cheapest " + "".join(f"{min(processes, key=lambda p: cost_per_part(p, q)).split()[0]:>11s}" for q in quantities))
# Where do the metal processes cross over? Try every quantity up to 20 000, then a coarser grid.
metals = ["CNC from aluminium bar", "sand casting", "investment casting", "high-pressure die cast"]
grid = np.concatenate([np.arange(1, 20_001), np.unique(np.round(np.logspace(4.31, 6, 300)).astype(int))])
winner = [min(metals, key=lambda p: cost_per_part(p, q)) for q in grid]
print("\nif the part must be aluminium:")
print(f" from {grid[0]:>7,} parts: {winner[0]}")
for i in range(1, len(grid)):
if winner[i] != winner[i - 1]:
print(f" from {grid[i]:>7,} parts: {winner[i]}")The numbers tell the familiar story. Ten housings cost EUR 64.60 each by machining and over EUR 4000 each by die casting. If the part must be aluminium, machining wins up to 90 parts, sand casting from 91 to 5362, and high-pressure die casting from 5363 on, levelling off at EUR 5.90. Investment casting never wins on cost here; it earns its place when the part needs fine detail, thin sections or alloys that cannot be die-cast. And if a glass-filled polyamide moulding meets the requirements, it costs EUR 3.60 at ten thousand parts and EUR 1.15 at a million, which is why so many housings that used to be metal are now plastic.
Metrology and quality control
A tolerance on a drawing is a promise; metrology, the science of measurement, checks whether the factory keeps it. Workshops measure with calipers, micrometers, bore gauges, height gauges on surface plates, and gauge blocks as references; quality labs use coordinate measuring machines (CMMs) that probe hundreds of points and compare them with the CAD model, optical and laser scanners, and profilometers for surface roughness. Every measurement has an uncertainty, and a measuring system must be checked (a gauge repeatability and reproducibility study) before its results are trusted: a rule of thumb asks for an instrument at least ten times finer than the tolerance it checks. Measurements are traceable, through a chain of calibrations, to national metrology institutes and to the SI metre.
Measuring every part tells you which parts are bad. Measuring a sample regularly tells you whether the process is good, which is cheaper and catches problems before parts go wrong. Walter Shewhart at Bell Telephone Laboratories introduced control charts in the 1920s; W. Edwards Deming took statistical quality control to Japanese industry after 1950, where it became part of the Toyota Production System and of the quality movement that followed.
Definition— process capability indices
For a stable process with mean and standard deviation , and specification limits LSL and USL,
compares the tolerance with the process spread; also penalises a process that is off centre. is a common requirement, and 1.67 or more for safety-critical features.
import math
import numpy as np
# A turned bearing seat: 25.000 +/- 0.020 mm. Measure 25 subgroups of 5 consecutive parts.
LSL, USL = 24.980, 25.020
rng = np.random.default_rng(11)
x = rng.normal(25.006, 0.005, size=(25, 5)) # the process runs slightly large
x[20:] += 0.009 # from subgroup 21 on, the tool has worn
def phi(z):
"""Standard normal cumulative distribution function."""
return 0.5 * (1 + math.erf(z / math.sqrt(2)))
def capability(mu, sigma):
cp = (USL - LSL) / (6 * sigma)
cpk = min(USL - mu, mu - LSL) / (3 * sigma)
ppm = 1e6 * (phi((LSL - mu) / sigma) + 1 - phi((USL - mu) / sigma))
return cp, cpk, ppm
stable = x[:20] # the first 20 subgroups, before the wear
mu = stable.mean()
sigma = stable.std(ddof=1, axis=1).mean() / 0.9400 # mean subgroup s / c4 (n = 5): within-subgroup sigma
cp, cpk, ppm = capability(mu, sigma)
print(f"process mean {mu:.4f} mm, sigma {sigma * 1000:.2f} um")
print(f"Cp = {cp:.2f}, Cpk = {cpk:.2f}, expected out of tolerance {ppm:.0f} ppm")
cp0, cpk0, ppm0 = capability(25.000, sigma)
print(f"if centred on 25.000: Cp = Cpk = {cpk0:.2f}, {ppm0:.2f} ppm")
# X-bar control chart: limits from the stable period, then watch every subgroup.
means = x.mean(axis=1)
ucl, lcl = mu + 3 * sigma / math.sqrt(5), mu - 3 * sigma / math.sqrt(5)
print(f"\nX-bar chart: centre {mu:.4f}, limits {lcl:.4f} to {ucl:.4f} mm")
for i, m in enumerate(means, start=1):
pos = int(round((m - 24.995) / 0.025 * 50))
flag = " <- out of control" if (m > ucl or m < lcl) else ""
print(f"{i:3d} {m:.4f} " + " " * pos + "o" + flag)
print("\n Cpk out of tolerance (centred process, both tails)")
for c in [0.67, 1.00, 1.33, 1.67, 2.00]:
print(f"{c:5.2f} {2e6 * (1 - phi(3 * c)):10.3f} ppm")
print(f"one tail at 4.5 sigma (Cpk = 1.5): {1e6 * (1 - phi(4.5)):.1f} ppm")The process is precise ( µm, ) but runs 6 µm large, so and about 1000 parts per million would be out of tolerance. Re-centring it, a simple tool offset, raises to 1.49 and cuts the rejects to about 8 per million. The control chart, with limits computed from the first twenty subgroups, flags every subgroup from 21 on, as soon as the tool wear shifts the mean, long before many parts are out of tolerance. The last table explains the "six sigma" target of the 1980s: a process whose tolerance is but which may drift by has and produces about 3.4 defects per million.
Safety in workshops and factories
Manufacturing concentrates energy: spinning chucks and cutters, presses delivering hundreds of tonnes, molten metal at over 600 °C (aluminium) to over 1500 °C (steel), welding arcs bright with ultraviolet, robots that move faster than people can react. Workplaces control these hazards in a fixed order of preference: eliminate the hazard if possible, engineer it away (guards, interlocks that stop a machine when a door opens, light curtains and two-hand controls on presses, fume extraction, enclosed robot cells), use procedures and training (lock-out before maintenance, permits for hot work), and finally personal protective equipment (eye protection, hearing protection, aluminised clothing in foundries, welding helmets and gloves). Safety regulators such as the UK's HSE and the US OSHA publish guidance for each process, and machines sold in many countries must meet machinery-safety standards before they can be put on the market.
These processes belong in properly equipped workshops and factories, run by trained people. What a learner can safely do is design for them: model a part in CAD, generate and simulate its tool paths in a CAM program such as FreeCAD's CAM workbench (which shows the material being removed on screen), check a moulded design's wall thickness and draft, and then, in a school or makerspace workshop with supervision, make simple parts with hand tools or a 3D printer used as its maker intends.
How it is done in industry
In a modern plant the processes in this lesson are linked by data. CAM programs come from the same model as the drawing; CMMs report measurements straight into statistical process control software; machine tools report spindle loads and vibration so that worn tools are changed before they break. ISO 9001 quality systems require that processes are controlled and capability is demonstrated before production starts, and car and aerospace makers add formal approval procedures for every new part and supplier. Lean manufacturing, which grew out of the Toyota Production System, attacks waste in waiting, transport and inventory as hard as statistical methods attack variation. The frontier lesson on advanced manufacturing picks up where this one ends: hybrid machines that add and subtract material, digital twins of whole production lines, and machine learning on process data.
Common misconceptions
Misconception
"Machining is always the most accurate process." Machining is accurate, but a good die casting or moulding can hold tight tolerances on many features at once, part after part. The usual practice is to cast or mould near net shape and machine only the features that need it.
Misconception
"Cutting faster is always more productive." Past the minimum-cost speed, tool changes and broken edges cost more than the time saved; Taylor's law makes tool life collapse quickly with speed.
Misconception
"If every part is inspected, quality is assured." Inspection finds defects after they are made, and inspectors miss some. Controlling the process, so that it rarely makes a bad part, is cheaper and more reliable.
Misconception
"A high Cp means few rejects." Only if the process is centred. A process with running near one limit can reject thousands per million; is the index that counts.
Exercises
1. Manufacturing processes
Warm-upHumanA 40 mm diameter steel bar is to be turned at a cutting speed of 150 m/min. What spindle speed should the lathe run at, to the nearest rev/min?
2. Manufacturing processes
Warm-upHumanA turning operation runs at a cutting speed of 120 m/min with a feed of 0.20 mm/rev and a depth of cut of 3 mm. What is the material removal rate, in cm³/min?
3. Manufacturing processes
CoreHumanA carbide tool follows Taylor's law . The cutting speed is raised by 20%. What happens to the tool life?
4. Manufacturing processes
CoreHumanA shaft diameter is specified as mm. The stable process has mean 10.02 mm and standard deviation 0.010 mm. What are and ?
5. Manufacturing processes
CoreHumanAn injection-moulded part with 1.5 mm walls needs 6 s of cooling before it can be ejected. A redesign thickens the walls to 3 mm, with the same plastic and temperatures. About how many seconds of cooling will it need?
6. Manufacturing processes
CoreHumanProcess A has tooling costing EUR 2000 and a running cost of EUR 12 per part. Process B has tooling costing EUR 30 000 and a running cost of EUR 5 per part. Ignoring tool wear, at how many parts do the two cost the same in total?
7. Manufacturing processes
CoreComputationalWrite cost_per_part(proc, n) and cheapest(processes, n).
procis a tuple(tooling, tool_life, variable): a tooling cost paid once per tool, the number of parts one tool can make, and the running cost per part.cost_per_partreturnstooling * ceil(n / tool_life) / n + variable.processesis a dict from names to such tuples;cheapestreturns the name of the process with the lowest cost per part fornparts.
The tests use the lesson's housing, with each process's material, process and finishing costs added into variable.
8. Manufacturing processes
CoreComputationalWrite capability(data, lsl, usl) that returns (cp, cpk) for a list of measurements, estimating the mean and the standard deviation from the data (use the sample standard deviation, dividing by ). Raise ValueError if usl <= lsl or there are fewer than two measurements.
Further reading and resources
- NIST/SEMATECH e-Handbook of Statistical Methods · National Institute of Standards and TechnologyFree, thorough chapters on process capability, control charts and measurement-system analysis.
- Sandvik CoromantA Swedish cutting-tool maker whose website explains cutting data, tool wear and machining formulas.
- TWIThe UK's Welding Institute, where friction stir welding was invented; technical knowledge articles on welding and joining.
- FreeCAD documentation wikiThe CAM workbench generates and simulates tool paths and G-code from a FreeCAD model.
- ISOPublishes ISO 9001 (quality management), ISO 22514 (process capability) and ISO 2553 (welding symbols), among others.
- OSHA, machine guarding · U.S. Occupational Safety and Health AdministrationWhy and how machines are guarded.
- Health and Safety Executive (UK)Guidance on work equipment, welding fume, foundries and presses, written for workplaces.
- Manufacturing Engineering and Technology · Serope Kalpakjian and Steven R. SchmidA standard, broad textbook on manufacturing processes.
- Materials Selection in Mechanical Design · Michael F. AshbyIncludes the process-selection charts and cost model behind the approach in this lesson.