PE/PP extruder motor current with AI — anticipate the jam without stopping the line.

The problem

The screw does not warn you in words — it warns you in current, and nobody reads it in time.

A screw jam on a PE/PP extruder is one of those stoppages you can see coming, but only if you look in the right place:

1. The pellet arrives with moisture slightly out of spec, or with a supplier batch change that the ERP never passed down to the shop floor.
2. Melt viscosity rises. The motor current starts rising with it, but the operator is watching die temperature and film thickness, not the ammeter.
3. The drive holds — until it doesn't. It goes into overcurrent and trips. The line stops, the die has to be opened, the barrel purged, everything cleaned.
4. The next start-up means kilos of scrap, and the blowing or winding line downstream runs out of material for hours.

The shift supervisor knows it. The operator knows it. What is missing is a layer that learns the normal current pattern per recipe and per screw speed, cross-checks it with pellet moisture and temperatures, and tells the operator "drop to 80 rpm or change the filter" before the stoppage. The veteran's intuition — the one who could hear the jam coming in the sound of the extruder — retires next year.

How it fits into the IRIS system

iLEAN Edge does not add one more sensor — it learns the normal current signature and watches the drift.

The extruder's problem is not a lack of sensors: the motor drive already has current, the barrel has thermocouples, the die has pressure. What is missing is a layer that learns the normal current pattern per recipe and per screw speed, detects the drift before the drive trips and cross-checks it with everything else. iLEAN is the filler that closes the gap left open by the extruder SCADA, the plant MES and the shift supervisor's memory.

Edge sees the current signature as nobody had seen it before. Connect captures pellet moisture and recipe. The agent cross-checks and proposes. The operator acts before the stoppage.

The three iLEAN pieces applied to PE/PP extruder motor current:

• Edge — a terminal in the drive cabinet reading current (Modbus from a modern drive, or a clamp ammeter if the drive is old and you would rather not touch OT). A CNN trained on the normal pattern per recipe (PE 100, blown PE-HD, PP copolymer) and per screw speed. It detects drift against the pattern, not a fixed threshold. It works without a network: it keeps analysing on cabinet power alone and raises an alert through a dry contact to the local panel.
• Connect — captures from the inside the batch recipe from the MES, the barrel zone temperatures from the SCADA, the pellet moisture if there is a hygrometer (or the manual value the operator types into the extruder screen), the jam history from the CMMS. And from the outside, the pellet supplier's delivery note with the batch supplied, the warehouse messages about lot changes.
• Agent — cross-checks the current drift against recipe, moisture and temperatures, and proposes a proportionate corrective action: reduce screw speed, change the inlet filter, stop and clean before the forced stoppage. The shift supervisor signs off. The line never stops on its own during a critical reel change, and the veteran's knowledge about what to do in each situation is captured as an agent rule.

See the full IRIS architecture →

Before and after

A jam heard by ear vs. a jam anticipated with Edge

Aspect	Classic operation	With iLEAN Edge motor current
Reading the normal pattern	In the veteran's head	Learned per recipe and speed, checked in real time
Drift detection	When the drive trips	Much earlier — in the subtle change of signature
Corrective action	Reaction to overcurrent, forced stoppage	Reduce speed / change filter / stop in time
Start-up scrap	Unavoidable, several kilos	Reduced — many jams are avoided altogether
Intervention on OT	n/a	Zero — external clamp ammeter if needed
The veteran's knowledge	Retires with him	Captured as an agent rule

Impact estimate

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

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

• PE/PP extrusion plant with 3-8 lines (blown film, pipe, profile), a mix of modern and previous-generation drives, regular screw jams on some recipes.
• Edge pilot on one critical line + integration with MES/SCADA. External clamp ammeter if the drive is old, without touching OT. First value expected within a few weeks: normal pattern learned and first jam anticipated.
• Indicative payback between 4 and 9 months, depending on the historical jam frequency and the kilos of scrap per start-up.
• Expected reduction in screw jams of ≥ 30% against baseline in the first half-year.
• Hard lever: every jam avoided is OEE hours recovered + scrap kilos avoided + no delay to the downstream line. Three or four jams avoided in the half-year usually pay for the pilot.

And the CAIO's reasonable doubt

«What if the AI fakes the current signature and stops the line for nothing?» — hallucination is a problem of free generation, not of anchored tasks. On tasks where the AI does nothing more than compare a current time series against a pattern learned per recipe, the best models brought the error below 1.5% ^[1]. And even so, the critical call is never made alone: iLEAN proposes the corrective action, the person signs off. On non-critical batches it can operate with self-approval to reduce speed; on critical reel changes, human validation is mandatory. The three rings exist precisely for this.

^[1] OpenAI paper «Why Language Models Hallucinate», 2025 — on the reliability of AI in anchored tasks.