Multi-cavity injection mold — per-cavity hydraulic pressure keeps short shots out of scrap.

The problem

The molding machine says «OK». Cavity 9 has been limping for three cycles.

In an 8, 16 or 32 cavity mold, the global hydraulic pressure of the molding machine is an average that hides what happens inside each cavity. When one cavity fills with less pressure —partially blocked channel, unbalanced hot runner, locally compromised cooling, worn insert—, out comes a short shot, flash or a dimensionally out-of-spec part. The molding machine raises no alarm; the global cycle is within band. The problem lives on three islands:

1. The per-cavity pressure signature (where piezo transducers exist) or the screw curve + position signature that gives away which cavity is limping — nobody watches it cycle by cycle.
2. Hot runner balance (zone temperatures) shifts from shift to shift when the material picks up moisture or when the supplier's batch arrives with a different MFI.
3. Vision on the part as it comes out almost always depends on the end-of-line operator or on sampling inspection — and by then the short shot is already travelling in a box.

The classic system (molding machine + thermal balance controller + operator visual inspection) works most of the time. When it doesn't, it turns into accumulated scrap, a customer claim (automotive, packaging) and hours of mold setup against the clock. It is not a lack of skill; it is information living on islands that never meet at second zero.

How it fits into the IRIS system

iLEAN Edge does not add yet another supplier panel — it seals the crack between the molding machine, the cavities and the part coming out.

The piece that solves the pain is iLEAN Edge: a physical terminal on the plant floor, next to the molding machine, that sees the mold the way a veteran molding supervisor would if they could read transducer by cavity, hot runner balance and a photo of the part off the robot all at the same time.

Edge closes hydraulic pressure per cavity (with a transducer, or by inference from screw curve + vision). Connect reads the material batch, the hot runner balance and the incident logged on the thermal controller. The agent cross-checks and proposes the correction to the molding supervisor. The person signs off — short shots never leave the shop.

The three iLEAN pieces applied to the multi-cavity mold:

• Edge — terminal next to the molding machine, reading (where they exist) the per-cavity pressure transducers, and/or the screw curve + hydraulic pressure + vision on the take-out robot. The CNN learns the normal signature per SKU and mold and detects the unbalanced cavity within a few cycles. Where appropriate, it triggers ejection of the defective part with a millisecond actuator — the same pattern as the automotive powder coating case. It works with no network: the shop keeps scrap under control even without the molding machine supplier's cloud.
• Connect — captures the material batch (virgin/regrind, supplier ticket, MFI), the mold recipe, the hot runner balance from the thermal controller, and SKU changeovers. It also captures what comes from outside: the supplier email warning that the next batch arrives with higher moisture, the shift handover note detailing the balance adjusted by hand.
• Agent — cross-checks the unbalanced cavity with mold history, material batch, hot runner balance and mold servicing. It proposes to the molding supervisor: retune zone N of the hot runner, schedule an inspection of the feed channel of cavity 9, hold the current SKU until material drying is complete. The supervisor signs off. It does not retune zones on its own — except where the management team configures autonomy per family.

See the full IRIS architecture →

Before and after

Classic multi-cavity molding vs. molding with iLEAN Edge on per-cavity pressure + vision

Aspect	Molding machine + operator visual inspection	With iLEAN Edge on per-cavity pressure + vision
Granularity of control	Global machine pressure	Per-cavity signature, cycle by cycle
Short shot detection	Sampling or the customer	Within the cycle, with automatable rejection
Cross-check with hot runner balance	Separate screen, separate person	Cross-checked by the agent on the plant floor
Cross-check with material batch and moisture	Not done	Connect brings it in at second zero
Operation with no network / no manufacturer cloud	Reduced	Edge keeps working as long as the panel has power
Dossier per box for the customer	Rebuilt by hand if there is a claim	Per cycle and cavity, automatic, with a human signature

Impact estimate

Impact estimate for your plant — to be validated with your numbers.

The block below is an estimate to be validated with the concrete data of your plant. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• Automotive or packaging molding shop with multi-cavity molds (8/16/32+), hot runner, take-out robot and sampling-based quality control. Documented short-shot / dimensional scrap, with occasional customer claims.
• Edge pilot on one mold (camera over the take-out robot + reading of global pressure and ideally cavity transducers + MES integration). First value expected within a few weeks: identifying the limping cavity before final inspection.
• Indicative payback between 4 and 9 months, depending on the frequency of documented incidents and the average cost of scrap + rework + impact of a customer claim.
• Expected reduction of short-shot / cavity-imbalance scrap of ≥ 30% in the first year. A reminder of the automotive quality standard: in the order of 25 PPM ^[1] — incompatible with unbalanced cavities not detected within the cycle.

And the molding supervisor's fair objection

«What if the AI invents an unbalanced cavity that is actually running fine and makes me throw away good parts?» — the proposal does not execute itself and, in what does execute (rejecting a part with a clear short-shot mark), the decision is anchored: vision on that specific part + cycle pressure, not free-running inference. On anchored tasks, the best models brought error below 1.5% ^[2]. And adjusting hot runner balance or the recipe is not done on its own; it stays with the molding supervisor's signature. The three safety rings are there for exactly this.

^[1] Automotive quality standard ~25 PPM — source Symestic.

^[2] OpenAI paper «Why Language Models Hallucinate», 2025 — on the reliability of AI in anchored tasks.