Float glass for automotive with AI — the inclusion shows on the table, not in the furnace.
The float glass tempered for automotive windscreens and windows carries tin inclusions, bubbles, knots and scratches that become critical in the tempering furnace or at the OEM. iLEAN Vision reads the pane on the loading table, sets the part aside before the quench and leaves the rejection in the quality manager's hands. The person signs off.
The defect that weighs the most is the one discovered after tempering or, worse, on the OEM's line.
The pane entering the automotive tempering furnace has already paid for cutting, edging, washing and screen-printing. Once in the furnace, it pays for the full cycle and the air bed. If the pane comes out with an inclusion visible in the field-of-vision zone, a knot that breaks the temper anomalously, or a scratch the OEM detects in its inspection, the part is expensive scrap or an expensive return. The end buyer does not accept "99% correct": they ask for the order of 25 PPM [1].
- The defect is subtle in flat glass, obvious after tempering or on the OEM's line. A half-millimetre inclusion in a critical zone passes the operator's eye and becomes a return.
- Human inspection on the table does not scale. The inspector sees two square metres per second under general lighting; subtle defects need directed lighting and filters, which the person is not wearing.
- The cost accumulates with every process. Tempering, bagging and shipping a pane the OEM is going to return is a cost you do not see until the notice arrives.
The classic system works almost always. The OEM's return is the one the customer remembers.
iLEAN Vision seals the crack between the loading table and the tempering line — it adds no system, it fills the gaps.
Surface control on automotive glass is not missing because nobody knows how to do it: it is missing because until now it was expensive to install over every loading table, and because the OEM's tolerances and the cutting recipe lived in different systems. iLEAN acts as the filler that closes that gap without asking you to change the tempering line or the MES.
Edge sees the pane before the furnace. Connect reads the cutting recipe and the OEM's tolerances. The agent cross-checks defect, the pane's position in the vehicle and the standard — if the zone is critical, it sets the part aside. The person signs off.
The three iLEAN pieces applied to float glass inspection for automotive:
- Edge — a terminal with machine vision (CNN) over the tempering line's loading table. Directed lighting (bright/dark field, polarised for stresses) and a neural network trained for tin inclusions, bubbles, knots and scratches. It triggers a light signal or diverts the pane to the rejection table before the quench. It works with no network: as long as the cabinet has power, the cycle continues.
- Connect — captures the cutting recipe (which pane it is, which position of the vehicle it goes to), the OEM's tolerances (ECE R43 and internal criteria) and the furnace's history, whether they come from the MES, the ERP or the OEM's email with the latest specification. The information arrives at second zero.
- Agent — cross-checks the detected defect, the zone within the pane (field of vision vs. periphery), the position in the vehicle and the OEM's tolerances. If the combination is critical, it sets the pane aside and alerts the quality manager. The person validates and signs; the line never restarts on its own for critical parts.
Visual table inspection vs. cross-checked inspection with iLEAN Vision
| Aspect | Inspector + general lighting | With iLEAN Vision + Connect + Agent |
|---|---|---|
| Detection point | Before or after tempering, depending on the plant | Loading table, before the quench |
| Coverage | Sampling + the operator's eye | 100% of the panes entering the tempering line |
| Tolerance by zone and SKU | The operator's memory | The OEM's rules loaded into the agent |
| Cost of the OEM's rejection | Return + reverse logistics + audit | Pane set aside before the furnace (an order of magnitude less) |
| Operation with no network | n/a | Edge keeps operating on cabinet power |
| File for the OEM | Manual reconstruction of the lot | Dossier per pane with the defect's photo |
Impact estimate for your plant — to be validated with your numbers.
The block below is an estimate to be validated with the specific figures of your plant. We set it out so the committee has an order of magnitude; we refine it during the diagnostic.
- Automotive glass plant with cutting, edging, washing, screen-printing and a horizontal air-bed tempering line.
- Edge pilot on the tempering line's loading table (directed lighting + cameras + integration with the cutting recipe). First expected value within a few weeks.
- Reduction in panes rejected after tempering or returned by the OEM: ≥ 30% as a defensible floor.
- Indicative payback between 4 and 9 months, depending on the documented frequency of OEM returns and the average cost of the finished pane in your mix.
- The hard lever is the avoided cost of the OEM's return: the audit it triggers weighs as much as the pane's material cost.
And the quality manager's reasonable doubt
"What if the AI flags as critical an inclusion the OEM would have accepted?" — hallucination is a problem of free generation, not of anchored tasks. In image classification against a learned pattern + a cross-check with the OEM's tolerance table, the best models brought the error below 1.5% [2]. And even so, what is critical is never decided alone: iLEAN sets aside the pane and the person signs off. The three safety rings are there for exactly this.
[1] Automotive quality standard in the order of 25 PPM (Symestic).
[2] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.
What people ask about float glass inspection for automotive
What typical defects appear in float glass for automotive?
In the float cut for windscreens and windows the typical ones are: tin inclusions from the float bath, bubbles from the furnace, knots and streaks, scratches and scuffs from transport on rollers, dust contamination before washing and residual stresses that burst in tempering. Most are subtle in flat glass and become a problem in the tempering furnace or on the manufacturer's assembly line.
Why inspect before tempering rather than after?
Tempering is one of glass's most expensive processes: an air-bed furnace cycle, energy-intensive, non-recoverable. Once the pane enters the furnace, the part has already paid for cutting, edging, washing, screen-printing and the energy of the full tempering cycle. An inclusion set aside before the furnace is one pane; the same inclusion discovered after tempering is a pane plus a complete furnace cycle. And in automotive the part is nearly finished — setting it aside weighs heavily. iLEAN Vision reads the pane on the tempering line's loading table and sets compromised parts aside before the quench.
How does iLEAN Vision read a pane on the loading table?
iLEAN Edge combines directed lighting (bright/dark field, polarised for stresses) with line-scan or matrix cameras over the tempering line's loading table and a neural network trained for the typical automotive float defects: inclusions, bubbles, knots, scratches. When a defect appears that is critical for the position (driver's field-of-vision zone vs. periphery), it triggers a light signal or diverts the pane to the rejection table. The operator has the defect's image on screen in milliseconds to validate the rejection.
How does it tell a critical inclusion from a tolerable one?
That is precisely the agent's reason for being: automotive standards (ECE R43 for windscreens, the OEM's criteria for side windows) define different tolerances depending on the part's position in the vehicle and the zone within the pane. iLEAN cross-checks the defect detected by Edge with the cutting recipe (which SKU the pane is, which position it goes to), the OEM's tolerances loaded in Connect and the furnace's history. The agent grades the rejection and proposes — the person validates and signs. The line never restarts on its own for critical parts.
What payback is reasonable to expect in an automotive glass plant?
In a plant tempering automotive glass, the cost of a rejection after the furnace is very high: the part has already paid for cutting, edging, washing, screen-printing and the full tempering cycle. The order of magnitude of an Edge pilot on the loading table covers directed lighting + cameras + integration with the tempering line. The reasonable payback to present to the committee sits between several months and a year, and the hard lever is the cost of a single pane set aside before the furnace versus one rejected afterwards or, worse, returned by the OEM. Send us your plant's data and we will send back the estimated ROI within 48h.
Tell us your case and within 48h we will send you the estimated ROI of this AI project for your automotive glass plant.
We work on your plant's real data, not on ours. Diagnostic with no commitment.
Request estimated ROI in 48h See iLEAN Vision