TPM on a high-takt engine line with AI — the 8 pillars working at once, not in silos.

The problem

Eight TPM pillars, eight systems that never talk to each other.

The TPM manager on a high-takt engine line knows the board by heart: autonomous maintenance runs on the operator's paper checklist, planned maintenance lives in the CMMS, predictive is a vibration sensor with its own software, focused improvement gets discussed in a weekly meeting, and training is noted down on yet another sheet. On paper, the 8 pillars are all there. In daily operation, each one lives on its own island.

1. The operator spots an anomaly — a knock in the spindle, a new vibration on the lathe — and has no way to link it with the breakdown history in the CMMS without stopping to go look for it.
2. Predictive raises an alarm from the vibration sensor — and it is never cross-referenced with the shift load or with the oil batch changed last week.
3. Planned maintenance triggers the intervention — and nobody knows whether the quality of the last batch was already starting to degrade before the stoppage.

The result on a high-takt engine line: unplanned stoppages that predictive "saw coming" in one system but nobody cross-referenced with the data in another; MTBF the committee only learns about two weeks late; MTTR broken down by eye in the maintenance meeting. Classic TPM works — and even so it leaves 20% of the performance on the table because information was not cross-referenced in time.

How it fits the IRIS system

iLEAN does not invent a ninth pillar — it connects the 8 you already have.

TPM is not the problem. The problem is that the 8 pillars do not share one picture of the line at the moment it is useful. iLEAN acts as the putty that fills the cracks between the operator in autonomous maintenance, the CMMS in planned maintenance, the vibration sensor in predictive and the focused-improvement lead, without asking you to change any of them.

Edge sees the line the way a veteran operator does. The Agent cross-references the pillars in real time. Brain computes live MTBF and MTTR. The person decides the intervention — the line never stops on a decision made by the AI.

The three iLEAN pieces applied to TPM on an engine line:

• iLEAN Edge — a terminal on the line with vision (CNN) + vibration/acoustic reading. It sees what the veteran operator sees (a change in spindle noise, drift in the lathe's pitch, an incipient leak) and logs every event. It works with no network: if the plant loses WiFi, Edge keeps capturing and holding. What is critical does not depend on WiFi.
• iLEAN Agent — cross-references the Edge signal with the CMMS (last planned maintenance, spare parts on hand), with predictive (vibration sensor history), with the quality batch (rising defects that point to a worn tool) and with the shift load. It detects degradation before the line stops and proposes the intervention to the maintenance manager. It proposes; the person signs.
• iLEAN Brain — computes MTBF, MTTR, OEE and availability continuously per machine, cell and line. It breaks MTTR down (diagnosis, part, repair) — without that breakdown you cannot tell where to attack. The TPM committee sees live KPIs instead of monthly retrospectives.

See the full IRIS architecture →

Before and after

Classic TPM vs. TPM with iLEAN on an engine line

Aspect	Classic TPM (8 pillars in silos)	With iLEAN Edge + Agent + Brain
Autonomous maintenance	Operator's paper checklist	Screen + earpiece, with the anomaly cross-referenced
Predictive	Isolated vibration sensor	Cross-referenced with quality, load and CMMS
MTBF/MTTR	Monthly report after the fact	Live, broken down into components
Degradation detection	When the alarm goes off	The trend before the alarm goes off
Old equipment (brownfield)	"It cannot be integrated"	Connect reads it however it can (local PC, photo)
Weekly TPM meeting	Reviewing the week's history	Deciding on the current state of the line

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 put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• High-takt engine line (machining + assembly + test), mixed brownfield (equipment 5-25 years old), classic TPM already in place but with unplanned stoppages on the rise.
• Edge + Agent + Brain pilot on the critical cell of the line (typically the crankshaft lathe, the block machining center or leak test). First value expected within a few weeks.
• Expected reduction in unplanned stoppages: ≥30% at 6-12 months, above all those that predictive "saw coming" but never cross-referenced with other systems.
• Indicative payback between 4 and 9 months, dominated by the cost per hour of downtime on the engine line (high at high takt) and by the parts that cannot be recovered when the machine breaks instead of stopping in time.

And the maintenance manager's reasonable doubt

"What if the AI proposes a stoppage we do not need and cuts my takt?" — hallucination is a problem of free generation, not of anchored tasks. When the AI simply cross-references real signals from the Edge, from the CMMS, from the vibration sensor and from the quality history (pure recontextualization), the best models brought error below 1.5% ^[1]. And even so, the Agent proposes; the person signs. The line does not stop on a decision made by the AI — it stops when the manager authorizes it, with the cross-reference already done.

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