Wire Harness Engineering: Crimp Compaction, Bundle Derating and the 18 AWG That Carries 2.9 A

Engineering guide to wire harness design: ampacity derating, voltage drop, crimp compaction, contact resistance, bend radius, shielding, hipot testing and cost.

·

Wire Harness Engineering: Crimp Compaction, Bundle Derating and the 18 AWG That Carries 2.9 A

The 18 AWG wire in the drawing is rated 16 A in free air. Route twelve of them into one bundle, put that bundle inside an enclosure that sits at 50 °C, and the same conductor carries 2.87 A. Nothing about the wire changed. The copper is still 0.823 mm², the insulation is still rated for the same temperature, and every catalogue in the drawer still prints 16 A on the front page.

That 5.6× collapse is the first thing a harness engineer learns and the last thing a schematic captures. A wire harness is not a circuit diagram with lines thickened. It is three designs stacked on top of each other, and they routinely disagree:

flowchart LR
    E[Electrical: ampacity and voltage drop] --> H[Finished harness]
    T[Thermal: bundle, ambient, connector grouping] --> H
    M[Mechanical: bend radius, supports, strain relief] --> H
    S[Standards: IPC/WHMA-A-620 class and test] --> H

A harness that passes on the bench and fails in the field almost always failed one of the three that was never written down. This guide works the arithmetic for all of them: conductor geometry, the derating stack, crimp compaction, contact constriction resistance, routing strain, shield transfer impedance, test levels, and the cost model that decides whether the design is manufacturable at all.

1. The conductor is a thermal device, not a pipe

AWG is a geometric series, not an arbitrary table. The diameter in millimetres follows directly from the gauge number:

For 18 AWG that gives d = 1.024 mm and A = \pi d^2/4 = 0.823 mm². Resistance follows from the resistivity of annealed copper at 20 °C, \rho = 1.724 \times 10^{-8}\ \Omega\cdot\text{m}, or more conveniently 0.01724\ \Omega\cdot\text{mm}^2/\text{m}:

AWG · Diameter (mm) · Area (mm²) · R at 20 °C (Ω/km) · Free air (A) · Bundled (A)

24 · 0.511 · 0.205 · 84.2 · 3.5 · 2.1

22 · 0.644 · 0.326 · 53.0 · 7 · 3

20 · 0.812 · 0.518 · 33.3 · 11 · 5

18 · 1.024 · 0.823 · 20.9 · 16 · 7

16 · 1.291 · 1.31 · 13.2 · 22 · 10

14 · 1.628 · 2.08 · 8.28 · 32 · 20

12 · 2.053 · 3.31 · 5.21 · 41 · 25

10 · 2.588 · 5.26 · 3.28 · 55 · 35

8 · 3.264 · 8.37 · 2.06 · 73 · 50

The last two columns are not properties of copper. They are the output of a heat balance. A conductor carrying current I dissipates I^2R per unit length and sheds heat from its surface by convection and radiation:

where h is the convection coefficient (W/m²K), d the conductor diameter, \Delta T the permitted temperature rise, and R' the resistance per metre. For 20 AWG in still air with h = 12 W/m²K and a 60 K rise, this predicts 7.4 A. The catalogue's 11 A "chassis" rating implies either a higher convection coefficient or a much hotter conductor — typically a 90–105 °C conductor temperature, which is exactly what the insulation rating allows and exactly what shortens the life of anything nearby.

This is why ampacity tables must be read as insulation-temperature tables, not wire tables. A PVC-insulated wire rated 105 °C and a PTFE wire rated 200 °C have the same copper and very different current ratings, because the limiting component is the polymer, not the metal.

Stranding changes the mechanical design, not the ampacity

IEC 60228 classifies conductors by flexibility. Class 1 is solid, class 2 is stranded for fixed installation, class 5 is flexible, and class 6 is extra-flexible. A 0.5 mm² class 5 conductor is typically 16 strands of 0.21 mm; the same 0.5 mm² in class 6 is 28 strands of 0.16 mm. The copper area is identical. What changes is the bending fatigue life: finer strands bend further before the outer fibres reach yield, which is why drag-chain and robot-arm cables specify class 6 and why a class 2 conductor fails after a few thousand cycles at the same radius.

Two rules follow, and both are enforced by IPC/WHMA-A-620:

2. The derating stack: where 16 A becomes 2.9 A

Ampacity tables assume a single conductor in free air at 30 °C. A real harness violates that assumption on both counts. The correction is multiplicative:

The bundle factor comes from NEC Table 310.15(C)(1), which is written for raceways but is the de facto convention for harness design:

Current-carrying conductors · Adjustment factor

4–6 · 0.80

7–9 · 0.70

10–20 · 0.50

21–30 · 0.45

31–40 · 0.40

41 and over · 0.35

Ambient correction for 90 °C-rated insulation is 0.91 at 40 °C, 0.82 at 50 °C, and 0.71 at 60 °C. For a 12-conductor bundle at 50 °C, an 18 AWG conductor rated 7 A bundled becomes:

That is the number the harness carries, not 16 A. If the enclosure is at 60 °C instead of 50 °C, it drops to 2.49 A. If the bundle grows to 25 conductors, the factor falls to 0.45 and the result is 2.24 A.

Altitude enters through the convection coefficient. Natural convection scales roughly with the square root of air density (h \propto \rho^{0.5}), and density at 3,000 m is about 74% of sea level, so a still-air harness at altitude sheds heat about 14% less effectively. That is not catastrophic for a control harness, but it matters for a 100% duty-cycle power feed in an unpressurised avionics bay.

Fusing current is not ampacity

There is a second, much higher current at which the conductor melts. Preece's equation, derived from experiments on telegraph wires in the 1880s, gives the fusing current for copper:

with d in inches and I in amperes. For 20 AWG (d = 0.0320 in), that is 58.5 A. Onderdonk's time-dependent form predicts about 47 A to reach the melting point in 10 seconds. Either way, the wire fuses at roughly 6× its bundled rating. The conductor is not a fuse. A 5 A-rated 20 AWG wire will carry 30 A long enough to ignite the insulation around it. Circuit protection must be sized to the wire's ampacity, not to its fusing current.

3. Voltage drop is the constraint that actually sizes wires

Ampacity sets the minimum gauge for safety. Voltage drop usually sets the minimum gauge for function, and it is the constraint most often skipped because it does not appear in a wire table.

The factor of two accounts for the return conductor. n is the number of contact interfaces in the loop — two per connector (pin and socket, source and load sides), and R_c is the contact resistance per interface.

Take a 12 V system, 5 A load, 6 m one-way run in 20 AWG, through two connectors:

The load sees 9.8 V. A 12 V solenoid may pull in; a 12 V router may reboot. The wire is thermally fine — 5 A in a 5 A-rated conductor — and electrically useless.

To hold a 3% budget (0.36 V at 12 V), the copper area must satisfy:

which gives 2.87 mm² — between 13 AWG (2.62 mm²) and 12 AWG (3.31 mm²). The design needs 12 AWG, two gauges larger than the ampacity table demanded. Note also that the four contact interfaces consume 0.20 V, more than half of the 0.36 V budget. Contacts are not a rounding error; at low voltage they are a primary design constraint.

ABYC E-11 codifies the two-tier convention that most low-voltage systems copy: 3% maximum for panel feeders, navigation lights, electronics, and other voltage-sensitive loads; 10% maximum for general lighting and non-critical circuits. The rule is applied to the round-trip length, which is why the formula carries the 2.

System voltage · 3% budget · 10% budget

12 V · 0.36 V · 1.20 V

24 V · 0.72 V · 2.40 V

48 V · 1.44 V · 4.80 V

At 48 V the same conductor goes four times further than at 12 V for the same percentage drop, which is the entire reason 48 V DC distribution exists in telecom and data-centre racks.

4. Crimp physics: compaction, not crushing

A crimp is a cold-weld process. The tool deforms the terminal barrel and the stranded conductor together until the strands are compacted into a solid mass, the oxide films on both surfaces fracture, and virgin metal extrudes into contact. On release, the barrel and the wire spring back by different amounts; because the terminal's springback is smaller, a residual contact force remains and holds the joint together for decades.

Solder does the opposite. Capillary action wicks solder up the strands beyond the joint, creating a stiff section that ends abruptly. Every flex cycle concentrates strain at the wicking boundary, and the strand fractures there. This is why IPC/WHMA-A-620 prohibits pre-tinning before crimping and why a soldered joint in a vibration environment needs strain relief that a crimp does not.

The compaction ratio

Crimp quality is quantified by how much the barrel compresses the conductor. SAE/USCAR-21 Revision 4 specifies a compaction ratio of 15–20%:

where S_1 is the cross-sectional area inside the crimped barrel, voids included, and S_0 is the nominal conductor area. A 0.5 mm² conductor crimped into a barrel whose internal area measures 0.59 mm² gives C = 18\%, inside the window. At 12% the joint is under-compacted: the strands are not touching each other or the barrel wall over enough area, resistance is high, and the crimp is loose. At 25% the strands are over-deformed, the barrel may crack, and the conductor cross-section is reduced.

A published response-surface study on a 0.5 mm² terminal found the optimum at a crimp height of 1.25 mm, crimp depth 2.36 mm, and crimp width 1.48 mm, producing 18.63% compaction and a 13.60 kgf (133 N) pull force. Holding crimp height to ±0.05 mm keeps the compaction inside the USCAR window — which is why crimp height, not pull force, is the production control variable. Pull force is destructive and sampled; crimp height is non-destructive and measured every setup, often every hour.

Geometry that matters

Pull force floors

AWG · UL 486A terminals (lbf / N) · SAE AS7928 (lbf) · NASA-STD 8739.4 (lbf)

26 · 3 / 13 · 7 · 8

24 · 5 / 22 · 10 · 8

22 · 8 / 36 · 15 · 13

20 · 13 / 58 · 19 · 21

18 · 20 / 89 · 38 · 32

16 · 30 / 133 · 50 · 41

14 · 50 / 222 · 70 · 65

12 · 70 / 311 · 110 · 103

Dividing the UL floor by the conductor area gives a useful rule of thumb: about 100 N per mm² of copper up to 12 AWG. The ratio falls for 10 AWG and 8 AWG because the terminal, not the wire, becomes the limiting component. Aerospace floors are roughly double the UL values for small gauges, reflecting the vibration and thermal cycling of the application rather than a different crimp process.

Failure modes and what causes them

Failure · Mechanism · Symptom

Under-crimp · Barrel not closed to spec; low compaction · High resistance, heating, loose wire; passes continuity

Over-crimp · Strands flattened, barrel cracked · Reduced cross-section, strand fracture at barrel edge

Missing bellmouth · Strands sheared at wire entry · Failures under vibration; pull force may still pass

Insulation in conductor crimp · Strip length short or wire under-inserted · No metal-to-metal contact; open circuit or high resistance

Nicked strands at strip · Blade depth too deep · Reduced pull strength, fatigue crack initiation

Double crimp · Re-termination without cutting back · Cracked barrel, work-hardened copper

Industry defect data consistently puts incorrect crimp height first, strand damage during stripping second, and insulation damage third — together roughly 58% of A-620 non-conformances. All three are process-control failures, not operator-care failures. A calibrated crimp tool, a strip-depth gauge, and a crimp-height chart catch most of them before the harness reaches test.

5. The contact interface: 10 mΩ is a physical dimension

A mated connector is not a short circuit. It is a constriction. Current passes from one metal to another through microscopic asperities, and the resistance of that constriction is given by Holm's model for a single circular contact spot:

For copper (\rho = 1.724 \times 10^{-8}\ \Omega\cdot\text{m}) and a measured 10 mΩ, the equivalent spot radius is 0.86 µm. One spot, less than a micrometre across, carrying the whole circuit. Real contacts use many spots and several newtons of normal force to keep them, but the order of magnitude is the point: contact resistance is set by real area of contact, which is set by force, plating, and surface film — not by the bulk conductivity of the pin.

This is why a 10 mΩ contact is a respectable datasheet number (Molex Micro-Fit specifies 10 mΩ maximum; 1.25 mm-pitch board connectors are commonly 20–30 mΩ) and why it can degrade without anything visibly changing. Tin plating oxidises; vibration of 10–50 µm amplitude causes fretting, which breaks the oxide and then re-forms it as insulating debris; resistance climbs from 10 mΩ to 100 mΩ or more over a few thousand cycles. Gold plating solves the oxide problem but only if the porosity is low enough; a 0.1 µm gold flash over nickel is a wear layer, not a corrosion barrier.

The power balance at the contact

At 10 mΩ and 10 A, each contact dissipates 1 W. A mated pair dissipates 2 W in a volume smaller than a sugar cube. UL 1977 caps connector temperature rise at 30 °C at rated current, so the contact system must present no more than 15 K/W to the environment for that pair. At 20 A the same 10 mΩ contact dissipates 4 W, and the pair 8 W — which is why connector current ratings are per contact, with all contacts energised, and why a 10.5 A-rated connector in a 12-way housing is not a 126 A connector.

Three practical consequences:

  1. Derate connector current in hot enclosures. The 30 °C rise is on top of ambient. A connector rated 10 A at 25 °C is not rated 10 A at 70 °C.
  2. Do not mix signal and power in one housing without checking grouping. Adjacent loaded contacts heat each other; the datasheet's single-contact rating assumes a worst case that may not match your loading pattern.
  3. Measure resistance, not continuity. A 100 mΩ crimp passes a continuity beeper at 1 mA. At 10 A it dissipates 10 W and destroys the housing.

6. Routing: the mechanical design is a fatigue calculation

Wire does not fail because it is bent. It fails because it is bent repeatedly at a strain it cannot survive. The surface strain in a conductor bent to radius R is:

For a 0.4 mm conductor (0.2 mm radius) bent to a 5 mm radius, \varepsilon = 4\%. Annealed copper yields at roughly 0.05–0.1% elastic strain, so the outer strands are far into plastic deformation on every bend. A single static bend is harmless — the strands rearrange and the copper work-hardens. A dynamic bend at that radius, repeated, is a fatigue test with a finite life measured in thousands of cycles, not millions.

That is the basis for the bend-radius rules:

Cable type · Static minimum · Dynamic minimum

Single conductor · 4× OD · 10× OD

Multi-conductor cable · 6× OD · 10× OD

Braided shield · 6× OD · 12× OD

Coaxial · 6–8× OD · 15× OD

Flat ribbon · 3× width · 6× width

Harness bundle · 4× bundle OD · 10× bundle OD

SAE AS50881, the aerospace wiring standard, recommends 10× the diameter of the largest wire or cable in the bundle as the preferred bend radius. For a 12-conductor bundle of 2.1 mm wires, the finished OD is about 8.4 mm, so the preferred radius is 84 mm — a number that frequently does not exist inside the enclosure the harness has to fit.

Support spacing is the other half of the mechanical design. The convention is a clamp every 300 mm for benign, low-vibration runs, 200 mm for moderate vibration, and 150 mm or less for high-vibration environments. IPC/WHMA-A-620 adds a specific rule for breakouts: the first restraint after a breakout must be within three bundle diameters or 100 mm, whichever is less. Without it, the breakout point becomes a hinge, and every flex cycle works the same few strands.

Other rules worth writing into the drawing:

7. Shielding: the pigtail is the failure

A braided shield with 85–95% coverage looks like a Faraday cage. It is not, if it is terminated with a 20 mm wire pigtail to chassis. The pigtail is an inductor:

A 20 mm pigtail is roughly 20 nH. At 1 MHz that is 0.13 Ω — negligible. At 10 MHz it is 1.3 Ω; at 100 MHz it is 12.6 Ω. The shield current, which is supposed to flow freely to chassis, now sees a significant impedance in series with the shield, and the cable radiates or picks up exactly as if the shield were not there. The fix is a 360° clamp — a shielded backshell, a compression gland, or a conductive gasket — so the shield terminates in a low-inductance ring rather than a single wire.

The second decision is where to ground the shield. Both ends is correct for RF: it forms a transmission line around the conductors and keeps the shield at chassis potential along its length. One end is correct for low-frequency analogue signals where a ground loop would inject more noise than the shield removes. The compromise is a hybrid: a 360° connection at both ends for RF, with a capacitor or a ferrite breaking the low-frequency loop.

Twisted pairs handle the other half of the problem. Twist pitch controls how well the two conductors see the same interfering field; a pair with a 25 mm pitch stops being balanced above roughly 100 MHz, where the pitch is a significant fraction of the wavelength. For a 1 m cable, that is a real limit, not a theoretical one.

8. Sealing, temperature, and the environment nobody specifies

IP67 is 1 m of water for 30 minutes. IP69K is 80 °C water at 80–100 bar, 14–16 L/min, from four angles. Both are tests of the interface, not the connector body: a sealed housing with an unsealed crimp is an unsealed harness.

The weak points are predictable:

Temperature rating is a chain: the lowest-rated component in the assembly governs. A 200 °C PTFE wire crimped into a 105 °C nylon housing with a 125 °C heat-shrink boot is a 105 °C harness.

9. Test: what to specify so the harness is actually verified

A continuity test proves that copper exists between two points. It does not prove that the connection will carry current. The test plan should be written into the drawing, not chosen by the supplier.

Test · Method · Typical acceptance

Continuity · 4-wire resistance, all points · Point-to-point, resistance below drawing limit (often 0.1–1 Ω)

Dielectric withstand (hipot) · AC or DC, conductor-to-conductor and conductor-to-shield · 2× operating + 1000 V (UL/IEC); A-620 Class 3 uses 1500 VDC or 1000 VAC

Insulation resistance · 500 VDC megohmmeter · ≥ 100 MΩ (connector datasheets commonly specify 1000 MΩ)

Crimp pull · Destructive tensile, 25–250 mm/min · Above the UL 486A / AS7928 floor; sampling per class

Crimp cross-section · Mount, grind, etch, measure · Compaction 15–20%, no cracks, no voids above limit

Crimp height · Micrometer or crimp-height gauge · Per terminal application spec, typically ±0.05 mm

Dimensional · First article and fixture check · Per drawing, including branch lengths and breakouts

Seal · IP67/IP69K per IEC 60529 / ISO 20653 · No ingress after the specified exposure

Two details matter more than the table:

Hipot voltage is not arbitrary. For a 24 V system, 2 \times 24 + 1000 = 1048 V. The same harness on a 400 V DC bus needs 1800 V. Testing a 400 V harness at the 24 V level proves nothing; testing a 24 V harness at 1500 V may be acceptable (A-620 Class 3) but should be agreed with the connector supplier, because the housing, not the wire, is usually the weakest dielectric.

Sampling is class-dependent. A-620 Class 1 is a quality-plan test; Class 2 requires first article plus periodic sampling; Class 3 requires first article plus every lot, with documentation. The class is set by the customer, not the manufacturer. Specifying Class 3 "to be safe" doubles inspection cost and does not improve a consumer product that will never see vibration.

10. DFM and cost: where harness quotes actually go wrong

Harness cost is dominated by operations, not materials, at low and medium volume. A typical breakdown for a 12–24 wire assembly:

Cost layer · Share · Driver

Wire and cable · 15–25% · Gauge, insulation, copper price

Connectors and housings · 15–30% · Pin count, sealing, brand

Terminals and contacts · 5–10% · Plating, size, crimp type

Sleeving, tape, labels · 3–8% · Protection and marking

Direct labour · 25–45% · Termination count, routing, test

Tooling and NRE · 5–15% · Crimp applicators, test fixture, nail board

Test and QC · 3–8% · Continuity, hipot, pull, FAI

The labour line scales with termination count, not wire length. Cutting and stripping 12 wires takes seconds; crimping 24 terminals at 6–10 s each is 2.5–4 minutes before routing, taping, and testing. A design that reduces terminations from 48 to 36 removes about 2 minutes of labour and 12 terminals — roughly 10% of the piece price on a small harness.

A worked model for a 12-wire, 0.8 m harness with two connectors and 24 crimps, at Indian job-shop rates:

Line · Assumption · Cost

Wire · 12 × 0.8 m × ₹10/m · ₹96

Terminals · 24 × ₹2 · ₹48

Connectors · 2 × ₹120 · ₹240

Sleeving, tape · — · ₹40

Material subtotal · · ₹424

Labour · 10 min at ₹300/h · ₹50

Test · continuity + hipot · ₹40

Overhead · 18% of ₹514 · ₹93

Cost · · ₹607

Price at 20% margin · · ₹758

The same harness at 5,000 units does not cost 5,000× less, but the NRE (nail board, test fixture, first article) spreads across the volume, material purchasing improves, and labour per unit falls with fixture-based assembly. A prototype at ₹1,500–2,500 and a production unit at ₹450–600 is a normal curve; a supplier who quotes the same number for 1 and 10,000 has not costed the NRE.

The ten DFM rules that move the number

  1. Reduce termination count. Every crimp is labour, a terminal, a pull-test sample, and a potential defect.
  2. Standardise on two or three wire gauges. Mixed gauges multiply setup changes and minimum order quantities.
  3. Do not over-specify temperature. Silicone at 200 °C costs 3–5× PVC at 105 °C; specify the ambient plus rise, not the worst case in the catalogue.
  4. Specify the A-620 class explicitly. Class 1, 2, or 3 changes inspection and documentation cost by 2–3×.
  5. Give a length tolerance, not a length. ±10 mm on a 1 m branch is normal; ±2 mm is a measurement operation.
  6. Design breakouts with room for the first restraint. Three diameters or 100 mm, whichever is less.
  7. Keep power and signal in separate bundles. It is cheaper than shielding.
  8. Use keyed, polarised connectors. A 1-in-16 chance of mis-mating is a warranty claim waiting to happen.
  9. Put the label on the drawing. Wire marking is a process step; if the marking scheme is unclear, every unit is a first article.
  10. Send the test specification with the RFQ. "Tested" means nothing; "hipot 1500 VDC, IR ≥ 100 MΩ at 500 V, pull test per UL 486A" means something.

11. What the harness is

A wire harness is the only part of an electrical system that is designed three times — once for current, once for heat, once for motion — and manufactured under a fourth discipline, the crimp, that none of the first three models. The 18 AWG conductor that carries 16 A on the datasheet, 7 A in a bundle, and 2.87 A in a hot 12-wire loom is the same copper in all three cases. The difference is the design around it.

The numbers in this guide are the ones worth carrying into a design review: 15–20% crimp compaction, 10 mΩ per contact, 30 °C connector rise, 3% voltage drop, 10× bend radius, 300 mm support spacing, and a hipot level of twice the operating voltage plus 1000 V. Each is a check that can be done on paper, before a single wire is cut — and each one, skipped, becomes a field failure that no amount of continuity testing will find.

If you are sourcing a harness rather than building one, put those numbers in the RFQ. The suppliers who can answer them are the ones who will build the same harness you designed.

More FabFlow blog posts