Beverage end-of-line palletizer — the shift ends with no stop because the motor current warned days earlier.

The problem

The palletizer's warning arrives late because nobody compares shifts.

The beverage end-of-line palletizer has been working without complaint for years. It builds its pallet, it does the format changeover, and nobody looks at it until one Friday at 19:40, with two trucks waiting, the drive alarm trips and an axis stops moving. What happened could have been read two days earlier — but the mimic only shows the snapshot of the moment:

1. The loaded-axis current had already been rising a stable step shift after shift on the same format — a clear symptom of a bearing, a misaligned axis or a gearbox asking for service.
2. The cadence of the line upstream (filler, labeller) changed subtly — more small micro-stops, a jumpier burst of bottles — and forced the palletizer to work flat out for days.
3. Last week's SKU changeover introduced a taller, heavier format, and nobody retrained the drive alarm for the new profile.

The classic system (drive alarm + monthly maintenance check) works 99% of the time. That 1% is Friday at 19:40 — and it translates into maintenance team overtime, a shipment that does not leave, and a salesperson facing the customer the next day. It is not a lack of maintenance; it is a lack of criteria in time.

How it fits into the IRIS system

iLEAN Edge does not add another dashboard — it seals the crack between the drive, the cadence MES and the production plan.

The piece that solves the pain is iLEAN Edge: a physical terminal in the palletizer's cabinet that reads the loaded-axis motor current the way a maintenance manager who walks the line every day would read it. It carries a neural network trained on the normal current signature per format and SKU, and cross-checks it in real time with the cadence arriving from upstream.

Edge sees the palletizer's current. Connect reads the production plan and the SKU changeover wherever it lives. The agent cross-checks the drift with the available maintenance window and hands it to the shift manager. The person signs — the palletizer does not stop by itself.

The three iLEAN pieces applied to the palletizer's motor current:

• Edge — a terminal in the palletizer's cabinet, reading RMS current, active power and estimated torque of the loaded axis. The CNN learns the normal signature per format and per SKU, not a fixed threshold. When the signature rises a stable step over several shifts, it raises a warning. It works with no network: if the plant loses WiFi, Edge keeps watching the current and recording the ramp, because what is critical cannot depend on WiFi.
• Connect — captures the production plan and the SKU changeover whether it comes from the ERP, the MES, or a spreadsheet the planning manager updates in the morning. It also captures the filler micro-stop upstream, the email from the new format's supplier and last week's message from the maintenance manager with the observation nobody turned into a work order.
• Agent — cross-checks the current ramp with the scheduled maintenance window, the SKU changeover in the plan and the team's availability. It proposes to the shift manager: “there is a 35-minute window at 17:30 between formats to inspect the axis, it avoids a stop at tomorrow's closing”. The manager signs. The system does not stop the palletizer on its own — except where the management team decides to raise autonomy.

See the full IRIS architecture →

Before and after

Palletizer with a drive alarm vs. palletizer with iLEAN Edge on the motor current

Aspect	Drive + monthly check	With iLEAN Edge on the palletizer current
Signal that triggers the reaction	Fixed drive threshold, the same for every format	Learned current signature per format and SKU
Warning horizon	Only when drift is already severe — minutes before stopping	Days earlier — the slow ramp is visible when it starts
Shift-to-shift comparison	Does not exist, except in the veteran manager's head	Edge keeps it always, per format and product
Production plan cross-checked	No — maintenance is planned separately	The agent proposes real windows from the plan
Operation with no network	Local alarm yes, context no	Edge keeps watching the current with the cabinet's power
Post-incident dossier	Rebuilt by hand	Current ramp, decisions and human signatures, automatic

Impact estimate

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

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

• Beverage bottling plant (water, soft drinks, beer, wine) with 1-3 lines and a multi-format end-of-line robot palletizer, with one or two stop incidents from a late alarm per quarter.
• Edge pilot on the palletizer (clamp-on current sensor + integration with the drive, the cadence MES and the production plan). First expected value in a few weeks: detection of the current ramp days before the drive threshold.
• Indicative payback between 4 and 9 months, depending on the frequency of documented incidents and the average cost of a stop at the end of shift (lost shipment + overtime + unfinished format).
• Expected reduction in stops from a late palletizer alarm of ≥ 30% in the first year, depending on the maturity of the maintenance plan. The hard lever is a single shipment saved — it pays for the pilot.

And the plant manager's reasonable doubt

“What if the AI invents a ramp that does not exist and makes me stop the palletizer for no reason?” — the proposal does not execute itself. iLEAN works on anchored tasks: it reads a physical signal that exists (current) and compares it with the product's learned signature. On anchored tasks, the best models brought error below 1.5% ^[1]. And even so, the shift manager sees the ramp, sees the proposal, and signs. The palletizer does not stop by itself — except where the management team configures it.

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