A tier-2 automotive supplier needed an 8-cavity tool for a glass-filled PA66 connector housing held to a ±0.01 mm class tolerance on three critical features. Balanced eight-drop gating, a hardness-matched steel choice and two measured rounds of steel correction brought cavity-to-cavity variation from 0.048 mm to 0.007 mm, with a 98.6% first-pass yield at T2.
Key takeaways
- On an 8-cavity tool, cavity-to-cavity variation — not the mean dimension — is what fails a tolerance check.
- A ±0.01 mm class tolerance on a connector is a tool decision first. Gating, cooling and steel grade set the ceiling; process setting only works inside it.
- Glass-filled PA66 shrinks anisotropically, roughly 0.35–0.55% with the flow and 0.65–0.85% across it. Cut cavity dimensions per direction, not to a single shrink factor.
- Measure 30 parts per cavity, free and constrained, before correcting any steel. One cavity tells you nothing about the tool.
- Hardened core and cavity inserts hold the geometry across a million-plus cycles. On an automotive tool, steel choice is a cost-of-ownership decision, not a purchase-price one.
The project at a glance
Part: sealed connector housing for an under-hood wiring harness, 46 × 28 × 19 mm, with an 8-way terminal block and a perimeter seal groove
Material: PA66-GF30, black, heat-stabilised
Nominal wall: 1.6 mm body, 1.1 mm ribbed lid, 0.9 mm seal groove floor
Critical requirements: terminal pitch within ±0.01 mm; seal groove width within ±0.02 mm; 8 cavities within 0.03 mm of each other; warpage under 0.15 mm across the mating face
Volume: 1.4 million parts a year across 8 cavities
Tooling: one 8-cavity production mold, hot runner, hardened 1.2344 (H13) inserts
Delivered from: RCH Plastic, Dongguan, China
The challenge: eight cavities that all had to agree
A single-cavity tool is judged on whether it hits nominal. A multi-cavity tool is judged on whether all eight cavities agree with each other.
A ±0.01 mm class tolerance on a 1.4 million part program
Terminal pitch and the two locating features decide whether the connector mates on the assembly line. At ±0.01 mm on critical dimensions the tolerance band is 0.02 mm wide, so every source of cavity-to-cavity difference has to go: fill imbalance, cooling imbalance and insert machining variation.
Volume makes it worse. At 1.4 million parts a year the tool runs continuously, so variation that looks acceptable on a trial sample shows up as drift over a production week.
Glass-filled PA66 does not shrink evenly
PA66-GF30 is a semicrystalline resin with 30% glass fibre. The fibres align with the flow and restrain shrinkage in that direction, typically 0.35–0.55%. Across the flow they restrain less and shrinkage rises to roughly 0.65–0.85% — a ratio near 1:1.6, and not uniform across a part with ribs, a terminal block and a seal groove. A cavity cut to one average shrink factor lands outside tolerance in both directions at once.
Warpage fighting a sealing face
The perimeter seal groove has to compress a gasket evenly. If the mating face warps after ejection, groove depth varies around the perimeter even when the steel is correct. The specification was 0.15 mm maximum warpage, measured in the constrained state, because that is how the part is used.
What we did
Five decisions, in order. Balance the flow, then the cooling, verify the process is stable, and only then cut steel.
1. A naturally balanced eight-drop manifold
The original concept fed eight cavities from a single central drop with a branching runner. We replaced it with a naturally balanced hot manifold: eight drops arranged so that the flow length from the inlet to every gate is identical, within 4 mm, so each cavity fills at the same rate and packs at the same pressure.
Balanced fill is the precondition for everything downstream. If one cavity fills 8 ms later than its neighbour it packs less, shrinks more, and lands outside the cavity-to-cavity spread no matter how good the steel is.
2. Per-cavity cooling rather than a single circuit
A ring layout with one cooling circuit gives the cavities nearest the inlet colder mold steel and the far cavities warmer steel — a recipe for exactly the drift we were eliminating. We built eight independent circuits, one per cavity, each with its own baffled core insert and flow control valve.
Measured mold steel temperature across all eight cavities after stabilisation: 82 °C ±1.5 °C. Balanced cooling is what makes balanced packing repeatable over a production week, not just on a trial sample.
3. Hardness-matched steel inserts
For a 1.4 million part program, the core and cavity inserts were cut from 1.2344 (H13) hardened to 48–52 HRC. The reason is dimensional, not just wear-related: a soft insert yields slightly under repeated high packing pressure, so cavity dimensions drift over the life of the tool. On a ±0.01 mm class tolerance, that drift is the failure.
The inserts were machined on our 13 CNC machining centres, with critical features finished after heat treatment on EDM and WEDM rather than before it, so hardening distortion could not move the terminal pitch.
4. Per-direction shrinkage compensation
Instead of a single shrink factor, we ran a Moldflow study to map shrinkage direction by direction across the part, then cut each cavity to compensate in the flow and cross-flow directions separately. The terminal block, the seal groove and the two locating features each got their own compensation factor.
This is why the DFM stage matters on a filled resin. A tool cut to an average factor would have been outside tolerance before a single part came off the press.
5. Two measured rounds of steel correction
Once the tool was cut and the process found stable, we shot a 30-piece sample from every one of the eight cavities — 240 parts — and measured on a CMM in the free state and in a fixture that constrained the mating face the way the connector sits in service. The residual bias proved repeatable.
Only then did we cut steel. Two rounds took the cavity-to-cavity spread from 0.048 mm to 0.007 mm. A third was not needed, which is the sign the process was stable before the tool was touched.
The result
Every trial round shipped as a documented package: samples, trial photos, a trial video, the process parameter sheet and a CMM dimensional report, issued against the mold number.
T1 verified fill balance across all eight cavities with a short-shot series. T2 ran the production resin, set gate seal with a hold-time study and pulled the 240-piece dimensional sample. T3 ran 5,000 consecutive parts to confirm the process window at both ends of the recommended settings.
| Parameter | Before (supplier tool) | After (RCH tool) |
|---|---|---|
| Cavity-to-cavity dimensional spread | 0.048 mm | 0.007 mm |
| Terminal pitch, worst cavity vs nominal | 0.023 mm | 0.006 mm |
| Seal groove width variation | ±0.035 mm | ±0.012 mm |
| Warpage across mating face (constrained) | 0.28 mm | 0.09 mm |
| Mold steel temperature spread, 8 cavities | 9 °C | 1.5 °C |
| Gates / balance | 1 drop, branching runner | 8 naturally balanced drops |
| Cycle time | 34 s | 27 s |
| First-pass yield at T2 | 91.2% | 98.6% |
| Process capability on terminal pitch (Cpk) | 0.79 | 1.61 |
| Scrap per 10,000 parts | 880 | 140 |
The dimensional result is the headline, but the yield result is what the customer banked. Moving from 91.2% to 98.6% first-pass yield on a 1.4 million part program is roughly 103,000 fewer rejects a year, before sorting. Cycle time fell from 34 to 27 seconds because balanced fill let us inject at lower pressure and shorten the hold.
The Cpk figure tells the same story in the language a quality engineer uses: 1.61 on terminal pitch means the process runs well inside the ±0.01 mm band with margin for ordinary process noise, not at its edge.
Our answer: how we build a tolerance-critical connector tool
RCH Plastic has been building injection molds in Dongguan since 2015, with more than 1,500 molds delivered to customers in North America and Europe, at 20–30 molds a month. Our answer to a tolerance-critical connector starts before steel: we model the part in UG and run it through Moldflow, then return a written DFM report within 3 working days at no cost, so gating, cooling, shrinkage compensation and the tolerance call-outs are settled on screen. Five mold designers averaging 8 years work alongside three project engineers and twelve fitters averaging 6 years — 49 staff, 37 in production and 12 in the office.
The machines that hold the geometry are 13 CNC machining centres with travels up to 1,600 mm, 6 EDM machines including Sodick mirror-finish and twin-head units, 3 wire EDMs, and CMM inspection on trial samples. Our machining tolerance is ±0.005 mm on critical mold dimensions; the molded part tolerance that tool can then hold is ±0.01 mm on critical dimensions, subject to the resin and the part geometry. The two numbers are different, and worth keeping separate when you compare suppliers.
Tools go up to 1,600 × 1,000 × 500 mm and 15 t, with up to 48 cavities, in steels including 1.2344 (H13), 1.2343 (H11), 1.2738HH and S136, on LKM, DME or HASCO bases, with hot runners from HRS, Yudo, Husky, Incoe or Mold-Masters. Lead time to T1 is 35 days on a simple tool, 42 medium, 50 complex, and 30 days when the drawing is finalised and components are in stock. Tool life is up to 1,000,000 shots depending on steel grade and maintenance, with a 12-month warranty. We make molds in-house and run trials and production through vetted partners, which keeps our engineering on the tool, where cost and lead time are decided.
What carries to your project
Five things transfer to any multi-cavity tool running a filled resin to a tight tolerance.
- Judge a multi-cavity tool on spread, not on the mean. A tool can hit nominal on average and still fail because cavity 3 disagrees with cavity 6. Specify the cavity-to-cavity tolerance explicitly.
- Balance fill first, cooling second, steel last. Steel correction multiplies whatever imbalance is already in the tool. Fix the tool before you cut the tool.
- Cut per-direction shrinkage compensation on any filled resin. A single average shrink factor cannot hold a class tolerance on glass-filled PA66.
- Sample at least 30 parts per cavity, free and constrained. One cavity and one state is not a measurement, it is an anecdote.
- Buy hardened inserts on a high-volume program. Steel grade is a tool-life and drift decision; the wear cost shows up long after the purchase order.
None of this is specific to under-hood connectors. Any multi-cavity tool feeding an assembly line — a sensor body, a switch housing, a terminal block — is judged on the same two things and fails for the same reasons. The order never changes: model the flow, balance the cooling, prove the process is stable, then cut steel.
Related guides in this series
- Injection molding defects and quality guide — the hub for this topic: defects, inspection and process capability
- Mold design and DFM guide — gating, cooling, shrinkage compensation and the DFM checklist
- Multi-cavity mold design — balancing, layout and cavity-count economics
- Injection molding tolerances guide — what a tool can actually hold, by feature
- How to choose mold steel — P20 vs NAK80 vs S136, and when hardening pays
- Injection molding cooling system design — circuits, baffles and temperature balance
- Hot runner vs cold runner — when a balanced hot manifold earns its cost
- Case study: warpage on a glass-filled PA66 bracket — the same material, a different defect
- Case study: thin-wall medical device enclosure — flatness and gasket sealing on one wall
Working on an automotive connector?
Send the 3D model, the resin grade, the GD&T drawing and the annual volume. We will come back with the likely failure mode on a multi-cavity layout, the first three things to check, and whether the fix is a process change, a tool modification or a new tool. Ask our engineers for a free DFM and flow review.