Hairpin Stators: The Manufacturing Physics Behind EV Traction Motors

Engineering guide to hairpin stator manufacturing for EV traction motors: 0.75 slot fill arithmetic, skin depth and AC copper loss at 800 Hz, the fifteen-step process chain, copper laser welding windows and porosity control, 800 V PEEK insulation and partial discharge, end-of-line test, and line economics at a 100-second takt.

·

Hairpin Stators: The Manufacturing Physics Behind EV Traction Motors

Almost every new 150 kW traction motor announced for an Indian or global EV uses the same winding architecture: rectangular copper bars pushed axially into stator slots, bent into U shapes at one end, twisted and laser-welded at the other. The hairpin stator has become the default because it converts a manual, variable process — winding round wire into slots — into a deterministic stamping-and-welding line. That determinism is what makes 150,000 units a year at automotive quality possible.

But the process is unforgiving in ways that are not obvious from the marketing literature. The same rectangular conductor that raises slot fill from ~0.40 to ~0.75 also exposes the winding to skin and proximity losses at 800 Hz, turns every joint into a 300 A connection that must hold for 15 years, and makes insulation a partial-discharge problem rather than a dielectric-strength problem. This is a working guide to the numbers behind those statements, and to what a shop actually has to control to make the parts.

1. The fill-factor arithmetic and what it buys

Slot fill factor is the ratio of copper cross-section to usable slot area:

where n_c is the number of conductors in the slot, w and h the conductor width and height, and A_{\text{slot}} the slot area available after the liner and wedge. Round wire in a random-wound slot packs at roughly 0.35–0.45 because circles cannot tile a rectangle; the hexagonal packing ceiling for equal circles is \pi/(2\sqrt{3}) \approx 0.907, and random winding never approaches it. Rectangular conductors with 0.1 mm assembly clearances achieve a packing ceiling near:

Real stators land between 0.60 and 0.80. Six 4.05 × 2.03 mm bars occupy 49.3 mm²; a 4.2 × 15.6 mm usable slot turns that into 49.3/65.5 = 0.75, while the same six bars in a 19.7 mm deep slot fall to 0.60 — slot depth bought for cooling or wedge clearance is fill factor spent. Series production EV motors sit near 0.70–0.75 after insulation build and tolerance stack-up.

The payoff is direct current resistance:

with \rho_{20} = 1.724\times10^{-8}\ \Omega\cdot\text{m} and \alpha = 0.00393\ \text{K}^{-1}. Doubling copper area halves R_{dc}; at 100 °C, copper resistivity rises to 2.27\times10^{-8}\ \Omega\cdot\text{m}, a 31% penalty that any thermal model has to carry.

Take a 150 kW, 18,000 rpm, 48-slot/8-pole interior-PM machine with 300 A rms per phase and 3.4 × 1.4 mm conductors. Published design data for exactly this machine gives 9.2 kW of stator copper loss at 5,000 rpm and 7.8 kW at 18,000 rpm, with current density of 29.5 A/mm² falling to 21.7 A/mm² as the field weakens. The 38 mm² of copper per slot — eight bars of 3.4 × 1.4 mm — is what keeps that current density survivable: the slot walls, not the copper, do the heat transfer, and the liner is only 0.25–0.35 mm of aramid paper and film.

Hairpin windings also shorten the end winding. A conventional 37 mm winding head on a 48-slot stator can be reduced to 24.5 mm by printing the crown directly onto the pins with laser powder-bed fusion of pure copper — 12.5 mm of axial length and 0.564 kg of copper saved per stator. That is not yet a production process, but it shows where the remaining DC resistance lives: outside the iron, in copper that produces torque only by leaking flux.

2. The frequency penalty nobody puts on the datasheet

Direct current resistance is the easy half. The conductor height that maximizes fill also maximizes eddy-current losses, because the relevant dimension is compared against skin depth:

For copper at 20 °C, \mu_0 = 4\pi\times10^{-7} H/m gives:

At 100 °C the skin depth grows to 3.79 mm, 2.68 mm and 1.69 mm respectively because resistivity rises faster than the frequency term changes. An 18,000 rpm 8-pole machine runs at 1,200 Hz electrical; a 6-pole at 900 Hz. Either way, a 2 mm tall bar is no longer small compared with \delta, and the current crowds toward the slot opening.

The measured consequence on the 150 kW reference machine is a ratio of total to DC copper loss rising from 1.05 at 5,000 rpm to 1.66 at 18,000 rpm. That is 66% above the DC value at the speed where the motor is already thermally stressed, in the same copper volume, with the same cooling.

Designers respond in three ways, all of which the manufacturing engineer inherits:

  1. Reduce conductor height. A 2 kHz high-speed machine with 1.48 mm skin depth wants ~1 mm conductors; manufacturing tolerance pushed the chosen height to 1.5 mm. The 48-slot/8-pole 150 kW design settled on 1.4 mm after balancing 270 Nm at 5,000 rpm against 18,000 rpm loss.
  2. Increase layer count. Four layers of 3 mm bars become eight layers of 1.4 mm bars in the same slot. More layers mean more welds: 48 slots × 8 layers ÷ 2 = 192 joints, against 96 for a four-layer layout.
  3. Transpose or use parallel paths deliberately. Incorrect transposition creates circulating currents between parallel branches that no amount of insulation will fix — it is a winding-scheme error, not a build error.

The review literature puts a practical ceiling near 72 slots and 8 layers; beyond that the weld count and tooling complexity grow faster than the electromagnetic benefit. That ceiling is a manufacturing constraint wearing an electromagnetic costume.

3. The process chain: fifteen steps, three of them decide yield

A hairpin stator is not wound. It is assembled from preformed bars. The RWTH Aachen process chain maps fifteen steps, and every one has a measurable output:

flowchart LR
    A[Straighten<br/>wire] --> B[Strip<br/>insulation]
    B --> C[Cut to<br/>length]
    C --> D[Bend<br/>3D hairpin]
    D --> E[Insert slot<br/>liner]
    E --> F[Pre-assemble<br/>basket]
    F --> G[Axial<br/>insertion]
    G --> H[Widen<br/>ends]
    H --> I[Twist to<br/>final position]
    I --> J[Trim and<br/>align]
    J --> K[Laser weld<br/>pairs]
    K --> L[Weld<br/>busbars]
    L --> M[Insulate<br/>weld ends]
    M --> N[Impregnate<br/>and cure]
    N --> O[End-of-line<br/>test]
    classDef hot fill:#1e3a5f,stroke:#38bdf8,color:#ffffff
    class K,L hot

Step · What is controlled · Failure if drifted

Straighten · Residual curvature < 0.5 mm/m, two-axis · Insertion misalignment, liner damage

Strip · Insulation residue RFU < 10, no fringing · Porosity, spatter, open joint

Cut · Squareness, 4-side chamfer, length ±0.2 mm · Slot liner cut, poor fit-up

Bend · Springback, outer-radius enamel thickness · Enamel cracks, corona at bend

Line slots · Liner wrinkle-free, projection flared · Ground fault, phase-to-phase short

Insert basket · Joint gap < 0.1 mm, no shaving · Liner shear, copper swarf in slot

Widen/twist · Tangential parallelism, height tolerance · Weld gap, misaligned pairs

Weld · Penetration, connection area, porosity · Hot joint, open circuit

Impregnate · Resin uptake, void-free slots · Partial discharge, hot spot

Test · Surge, hipot, resistance symmetry · Escape to customer

The 3D bend is the quietest yield killer. Bending 4 × 1.4 mm enamelled copper around a radius that is too tight thins the insulation on the outside of the bend and cracks it on the inside, and the damage is invisible until a surge test or a 500-hour endurance run. Springback on Cu-ETP is 2–5° depending on temper and bend radius; the tool has to over-bend by that amount, and the over-bend is a function of wire batch hardness, which is why incoming wire inspection includes yield strength, not just dimensions.

Stripping is the second. Laser stripping burns the enamel and then blasts the residue off with the absorbed beam; mechanical milling removes insulation plus a few micrometres of copper. Both are acceptable, and both fail the same way: residue at the root of the strip zone. That residue vaporizes in the weld keyhole, and every 0.1 mm³ of trapped gas is a pore in a joint that must carry 300 A.

4. Welding copper: a 5% absorption problem

Copper is the worst common metal to laser weld at 1,070 nm. An untreated surface reflects about 95% of near-infrared radiation, so the process only starts once irradiance is high enough to establish a keyhole — on the order of 10^6–10^7 W/cm². Once the keyhole forms, multiple reflections raise absorption to roughly 70%, and the process becomes stable. Getting there requires 3–8 kW of fibre or disc laser power focused to a 68–255 µm spot.

The usable window is narrow and is usually expressed as energy density:

Published hairpin welding trials place good welds at 11–13 MJ/cm², with lack of fusion below ~8 MJ/cm² and over-melting, large pores and spatter above ~16 MJ/cm². At the 4.2 × 2.5 mm pin face used in those trials, 12 MJ/cm² corresponds to about 1.26 J delivered into a 0.105 cm² area — a millisecond-scale event, repeated 120 to 288 times per stator.

The joint quality numbers from optimised parameters on 4.2 × 2.5 mm Cu-ETP pins are worth memorising:

That last figure is the one that matters commercially. A weld process with Ppk below 1.33 cannot be released to an automotive line without 100% inspection; 1.88 is what allows sampling plus in-line monitoring.

Porosity, oxygen and beam shaping

Porosity is measured as area fraction on a cross-section:

Two mechanisms dominate. Process pores above 0.1 mm form when the keyhole collapses as the beam crosses the gap between pins; metallurgical pores below 0.1 mm come from dissolved oxygen in the copper. The practical levers, in order of effect:

Weld-level acceptance in production is typically a minimum connection area at least equal to the conductor cross-section, pore volume below a few tenths of a cubic millimetre, and no spatter on adjacent enamel. CT is the reference method; laser-ultrasound and in-process photodiode/OCT monitoring are the production-compatible proxies.

Blue lasers and conduction mode

A 3 kW blue diode laser at 720 µm spot can weld hairpins in conduction mode rather than keyhole mode. At 50–150 mm/s with 2–7 scans, it produces sound joints with markedly less spatter — the trade is speed. Best porosity was about 4% at 100 mm/s and six scans; the worst case, at the lowest scan speed, reached 10% because the melt pool sat exposed to atmosphere longer and absorbed more oxygen. Conduction mode also makes external bead dimensions correlate with strength and resistance, which is attractive for inline vision inspection. It is slower than keyhole welding and not yet the volume choice, but it is the obvious answer where spatter damage to neighbouring insulation is the dominant scrap mode.

5. The 800 V insulation problem

At 400 V systems, magnet wire enamel of 60–100 µm is adequate. Above roughly 650 V, partial discharge starts in the air voids between conductors, and PD erodes enamel until breakdown. Every 800 V traction motor is therefore a PD-managed design, not a dielectric-strength design.

The manufacturing answer is a thicker, pinhole-free insulation layer. Extruded PEEK over enamel gives 150 µm walls with PDIV of 1,600–1,700 V and breakdown above 6 kV, with no PD at 1,500 V in typical constructions. PEEK also survives tighter bend radii than enamel without cracking, which is why it dominates 800 V hairpin programs despite costing more per kilogram than the copper it insulates.

Three interfaces must be controlled:

Impregnation is not cosmetic. Trickle epoxy penetrates the slot by capillary action while the stator rolls, then gels and cures in a controlled profile. Its jobs are to fill voids that would otherwise host PD, to bond the bars against vibration fretting, and to conduct heat from copper to iron. Resin uptake on the 19.36 kg reference stator is about 0.08 kg of epoxy plus 0.10 kg of secondary insulation — small mass, large effect. A void-free slot is worth more than a thicker liner.

6. End-of-line test: what actually catches a bad weld

A hairpin stator with 192 joints has 192 chances to be open, high-resistance, or intermittently connected. The test sequence is built around finding those before the customer does:

Test · Detects · Typical level

Four-wire resistance per phase · High-resistance joints, wrong wire · mΩ, matched to ±2% between phases

Surge comparison · Turn-to-turn and phase faults · kV pulse, waveform difference

AC hipot · Ground insulation failure · 1.5–2× rated + 1,000 V, 1 s

Insulation resistance / PI · Moisture, contamination · > 100 MΩ, PI > 2

Partial discharge · Void PD at 800 V · pC threshold, PDIV margin

CT or laser ultrasound (sampling) · Internal porosity, connection area · Pore volume, mm³

Phase-resistance symmetry is the cheapest and most sensitive check. A joint that is 200 µΩ high on a phase of a few milliohms shows up as an imbalance long before it becomes a field failure — and at 300 A that joint dissipates 18 W in a few cubic millimetres, which is enough to carbonize adjacent enamel within a few thousand kilometres.

7. Line economics: why 100 seconds per stator is the number that matters

A commercial hairpin line for 2–10 layer stators runs a 100-second cycle in a 30 m × 12 m footprint, fully MES-integrated. On a 60-second takt, the laser has 50–200 ms of on-time per pin pair, with the rest of the cycle spent indexing, clamping and verifying. Those two numbers bracket the capital decision:

At 100 s and 60% OEE over 300 days, that is 518 stators/day, or about 155,000 per year. At 75% OEE, 194,000. A line that costs €3 million in prototyping configuration and €10 million in series configuration only makes sense above roughly 50,000 units a year; below that, the tooling amortization per stator exceeds the copper cost.

Copper is the other number. The reference stator carries 5.04 kg of wound copper. At ₹1,342/kg (MCX, 12 November 2026), that is about ₹6,760 of copper per stator before scrap. At 8% process scrap the gross buy is 5.48 kg, or ₹7,350. At 150,000 stators a year that is 822 tonnes of copper and roughly ₹110 crore of annual copper spend — which is why a 3% reduction in weld-end scrap is a bigger saving than a 3% improvement in laser cycle time.

And the quality cost is asymmetric. At 192 welds per stator across 150,000 units, the line makes 28.8 million joints a year. A stator-level defect rate of one in a thousand corresponds to 5.2 ppm at the joint. That is the number a customer's supplier quality engineer will ask for, and it is achievable only if every weld has a parameter record tied to its serial number.

8. What this means for the Indian supply chain

India's traction motor base has moved from assembly to real manufacturing. Sona Comstar reports 67 EV programmes across 35 customers and has been producing traction motors since 2021, and is now developing an indigenous motor and inverter stack with Denso under the Bharat Motor programme. The company's own position is strongest in two- and three-wheeler motors, where volumes are high and unit prices low — exactly the segment where hairpin's 100-second takt is hardest to justify and where round-wire stators still win on capital efficiency.

For a tier-2 shop looking at hairpin work, the entry barriers are not the laser. They are:

The bending and insertion tooling is where variant flexibility dies. Each stator geometry needs its own twist tool; that is why prototyping lines cost €3 million and why additive-manufactured winding heads are being researched as a way to buy flexibility back. A shop that can quote quick-turn hairpin prototypes with a CNC bender and a manual weld cell has a real niche — as long as it can hold the same joint resistance the series line holds, and prove it.

9. The five failure modes to design out first

  1. Enamel damage at the bend radius. Outer-radius thinning and inner-radius cracking. Control bend radius ≥ 2× conductor height, inspect with an enamel continuity test, and prefer PEEK over enamel for 800 V.
  2. Insulation residue after stripping. Vaporizes in the keyhole, creates porosity and spatter. Control with RFU < 10 and a vision check of the strip zone.
  3. Lateral gap at the weld. Resistance rises linearly to 0.5 mm offset. Control twist position, not laser power.
  4. Weld spatter onto adjacent enamel. Creates a PD site that fails at 800 V after thermal cycling. Control with beam shaping, or move to conduction-mode blue laser welding.
  5. Resin voids in the slot. The stator passes hipot and fails PD after 500 hours. Control impregnation vacuum, resin temperature and roll profile, and verify by cross-section on a sampling basis.

None of these are electromagnetic problems. They are process-control problems, and they are why hairpin manufacturing is a welding-and-metrology business that happens to make motors.


FabFlow connects manufacturers with customers who need exactly this kind of process competence — from hairpin prototype bending to 800 V stator testing. If you run a winding, welding or impregnation cell and want to quote EV motor work, list your capabilities or post a job.

More FabFlow blog posts