Autonomous maintenance TPM — if it demands paper, it doesn't work. You know it and so do we.

The problem

TPM pillar 1 collapses on the floor because the tool weighs more than it gives back.

Any maintenance manager who has tried to roll out TPM pillar 1 knows the pattern. The first weeks there is goodwill: the operator fills in the checklist, ticks the cleaning box, writes down the odd vibration. Three weeks in, the checklist is unsigned. Two months in, the forms get signed at the end of the shift, all at once, without anyone having walked out to the machine. TPM pillar 1 does not collapse out of bad faith — it collapses because the tool asks for paperwork and returns nothing.

1. The channel is a barrier — open the CMMS, find the machine, type the issue, close it. Five minutes per report. The operator has a line to run; the report stays in their head and is lost at the end of the shift.
2. The report never comes back with an action — the logged issue drops into the CMMS and the operator never finds out what happened to it. Next time they see something odd, why report it?
3. The veteran's knowledge is never captured — the veteran operator knows the sound of their machine, tells a normal vibration from an odd one, sees the failure coming two shifts ahead. That knowledge walks out with them the day they retire.

The consequence: unplanned stoppages keep arriving, corrective maintenance eats the preventive budget, and TPM pillar 1 becomes the framed picture on the wall with the committee photos.

How it fits the IRIS system

iLEAN doesn't ask the operator to become an administrator — it gives them a voice and an agent that listens.

The IRIS category (Industrial Reality Intelligence Systems) starts from one principle: plant reality first, the report afterwards. In TPM that means changing the contract with the operator: you see what happens, I (the system) do the paperwork. iLEAN acts as the filler between the operator, the CMMS and the maintenance technician — without asking you to change CMMS or to touch the certified procedure.

Connect listens through the channel the operator already uses. Writer documents. The agent proposes the action. The person signs off.

The three iLEAN pieces applied to digital autonomous maintenance:

• Connect — listens through every channel: voice into the earpiece, a photo from the phone, WhatsApp, dictation, industrial tablet. Bidirectional: the agent can answer through the same channel (“received, spare part ordered, technician notified, I'll tell you when it arrives”). The operator reports the way they would tell a colleague, not the way they would fill in a form.
• Writer — turns the operator's natural report into a structured event in the CMMS: machine, failure code, description, attached photo, proposed priority. Everything that used to take 5 minutes of typing is prepared by the system and validated by the person.
• Agent — cross-references the report with that machine's history, with recent reports from sister machines, with the preventive maintenance plan and with the spare parts catalogue. It proposes the corrective action (cleaning, lubrication, spare part replacement, work order to the technician). If the customer has granted autonomy for routine actions, it executes; if it touches a critical work order, it waits for the signature.

See the full IRIS architecture →

Before and after

Paper-based autonomous TPM vs. digital autonomous TPM with iLEAN

Aspect	TPM pillar 1 on paper / classic CMMS	With iLEAN Connect + Writer + Agent
Reporting channel	Checklist signed at the end of the shift	Voice, photo, WhatsApp, tablet — whichever channel is already in use
Operator time per report	5 minutes of typing	15 seconds talking or taking a photo
Veteran's knowledge	Lives only in their head	Every voice report becomes a persistent pattern
Feedback to the operator	None — the issue disappears	The agent answers through the same channel with the action
Load on the maintenance technician	Receives badly filled-in orders	Receives a structured order with photo, history and proposal
Adoption at 3 months	Falls to around 30% — the form gets signed in a rush	Rises because the tool stops weighing and starts helping

Impact estimate

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

• Plant with 40-150 critical machines, TPM pillar 1 already existing on paper or in an unevenly used CMMS, real adoption of autonomous maintenance below 50%.
• Connect + Writer + Agent pilot on one pilot area (a line or a machine family). First value in a few weeks: as soon as the operator sees their voice report come back with the action, reporting frequency rises on its own.
• Increase in the number of voluntary reports of the order of 3x to 5x in the first months (estimate to be validated). Reduction in unplanned stoppages of the order of 15%-30% in the first year.
• Indicative payback between 4 and 9 months. The hard lever: unplanned stoppages avoided + operator time freed from paperwork + capturing the veteran's knowledge before they retire.

And the works council's fair question

“Is this about listening in on the operator?” — Connect picks up what the operator chooses to report about their machine, not their conversation. Voice is captured when the operator presses the earpiece button, dictates or speaks to the app — like a WhatsApp voice message, not an ambient microphone. And the rule of the system is a shop-floor rule: the agent proposes, the person signs off. What changes is not oversight of people (a framing that does not fit how we work): what changes is the contract with the operator, who now contributes what they know and gets something back. The agent does not invent either: it works on anchored tasks — this machine, this history, this report — and hallucination is a problem of free generation, not of anchored tasks; on anchored tasks the best models brought the error below 1.5% ^[1].

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