Every custom machine on time and right first time
In a plant that builds ultrasonic welding equipment almost everything is made to order: custom tooling, a recipe validated in the lab, its own configuration. A delay, or a machine that does not start first time, puts the customer's line launch at risk. The titanium delay arrives by email, the reworked horn sits on an offline panel, the recipe lives in a spreadsheet, the issue in the supervisor's head. Nobody sees the whole order until it is too late. The previous pieces, traveler, panel, email, earpiece, tablet, lab, receiving, tuning bench, inspection and configuration cameras, feed a single memory per order and serial number.
Nobody sees the whole order until it is already late.
In a plant that builds ultrasonic welding equipment almost everything is made to order: custom tooling, a recipe validated in the lab, its own configuration. A delay, or a machine that does not start first time, puts the customer's line launch at risk. The titanium delay arrives by email, the reworked horn sits on an offline panel, the recipe lives in a spreadsheet, the issue in the supervisor's head. Nobody sees the whole order until it is too late.
- In a plant that builds ultrasonic welding equipment almost everything is made to order: custom horn and nest, a recipe validated in the lab, its own configuration of frequency, booster and generator.
- A delay, or a machine that does not start first time, puts the customer's line launch at risk, and that is what the customer remembers at the next award.
- The signals exist but sit apart: the titanium delay in an email, the reworked horn on an offline EDM panel, the recipe in a spreadsheet, the missing booster in the supervisor's head.
- Each one on its own looks manageable. Together they decide the date, and nobody adds them up until assembly is already waiting and the customer is calling to confirm the FAT.
Agents with JIDOKA AI and SMED AI — the eleven pieces feeding one memory per order.
The previous pieces, traveler, panel, email, earpiece, tablet, lab, receiving, tuning bench, inspection and configuration cameras, feed a single memory per order and serial number. iLEAN Agents spots which order is going to be late and why, proposes the re-plan and prepares the delivery dossier. JIDOKA AI stops what does not match; SMED AI speeds up tooling changes. The operations director decides with the full picture.
No single case moves on-time delivery by itself. What moves it is that traveler, panel, email, earpiece, tablet, lab, receiving, tuning bench and both camera checks feed the same memory per order and serial, so the risk shows up days before the date instead of on it.
Orders seen in pieces versus orders seen whole
| Layer | Cases it draws on | What it gives the operations director |
|---|---|---|
| Titanium and ceramics supply | Spec change (3) + receiving (7) | Which orders depend on a late titanium or ceramics lot |
| Horn shop | EDM panel (2) + horn inspection (9) | Which horns are being re-cut or held |
| Recipe and tuning | Lab spreadsheet (6) + tuning bench (8) | Whether the recipe and frequency are closed |
| Assembly | Traveler (1) + earpiece (4) + tablet (5) | Real progress and open issues per machine |
| Release | Configuration check (10) + dossier (11) | Right-first-time evidence for shipping |
| Plan | Agents across all layers | Which order will be late, why, and a re-plan |
Before: delays found late and failed start-ups. After: risks anticipated days ahead; OTD +10 to +25 points.
Impact estimate — to be validated with your numbers.
The block below is an estimate to be validated against your plant's actual data. We put it forward so the committee has an order of magnitude; we refine it during the assessment.
- Estimated payback 6-12 months, to be validated with your delivery data.
- On-time delivery up 10 to 25 points, because risks are anticipated days ahead instead of found at assembly.
- Less rework and fewer warranty visits: JIDOKA AI stops what does not match the order and SMED AI speeds up tooling changes in the horn shop and the lab.
- And a delivery dossier that is already prepared when the machine ships, so on time also means documented, which is what the customer's quality team checks before accepting the FAT.
Payback 6-12 months · on-time delivery +10 to +25 points (estimate to be validated). On-time delivery up ten to twenty-five points, less rework and fewer warranty visits. Estimated payback of six to twelve months; an estimate to be validated with your delivery data.
And the fair question from the production manager
“Do we need all twelve cases before this does anything?” — no. The layers are built one at a time and each case pays, or enables, on its own from the start. The prediction itself rests on anchored inputs: each signal is extracted by an anchored task where the best models drop below 1.5% error [1], and Agents only proposes the re-plan. The operations director decides with the full picture.
[1] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.
What people ask about on-time delivery of custom ultrasonic machines
Where should an ultrasonic equipment maker start?
Usually with the assembly traveler and the tuning bench, which have few dependencies and show results in weeks. The other layers are added on top, in the order your delivery data points to. Each one pays or enables on its own, so there is no big-bang project.
How does Agents know an order will be late?
It compares what each order still needs, material, horn, recipe, tuning and assembly steps, with what is actually happening on the floor, and flags the order whose path no longer fits its date. It explains why, for example a horn re-cut plus a late booster, so the director can act on the cause.
Does it change our ERP or planning system?
No. The ERP remains the reference for orders and configurations. iLEAN reads it, cross-checks it with the floor and proposes changes for a person to approve.
What do JIDOKA AI and SMED AI do in this case?
JIDOKA AI stops a machine or a step that does not match the order. SMED AI shortens tooling and recipe changeovers in the horn shop and the applications lab, which frees capacity for the orders at risk. Both act on the floor; neither replaces the people who run it.
Can the gain in on-time delivery be estimated before committing?
Yes. With your last year's commit dates, actual ship dates and failed start-ups, we estimate the gain per layer during the AI Discovery. That tells you which layer to build first and what it should return.
Tell us what share of your custom machines shipped on the committed date last year.
We work on your plant's real data, not ours. Assessment with no commitment.
Request estimated ROI within 48h ‹ See all cases of ultrasonic welding equipment See metal fabrication and capital equipment