Physical AI for solar module manufacturing

Every module,
peak power.

Panevex is the autonomous operations layer for solar PV module lines. Seven AI agents perceive every cell, predict every defect, and close the control loop across stringing, lamination, binning, and a full module digital twin.

+3.2 Wavg power recovered / module
−41%microcrack escape rate
24/7closed-loop line autonomy
The problem

Gigawatts of solar are shipped with defects no human can see

A modern module line runs faster than any inspection team can follow. Microcracks, cold solder joints, cure faults, and cell mismatch slip through — and each one steals watts for 25 years in the field.

Invisible damage

Microcracks form during stringing and lamination but hide beneath the glass until field degradation reveals them.

Open-loop machines

Stringers and laminators run fixed recipes. When materials drift, nothing corrects the set-points in real time.

Mismatch losses

Cells binned on flash data alone leave watts on the table when real defect data is ignored.

Module digital twin Ribbix strings Elumnex sensing Lamira cure Binova match Twinvex simulate

One perception-to-control loop, not seven point tools

Panevex unifies the whole line under a single autonomy layer. Perception feeds control; control feeds the twin; the twin feeds every agent. The line gets measurably better with every module it builds.

  • Perceive — EL, PL, and RGB vision on every cell and string.
  • Predict — crack-risk and power impact scored in-line.
  • Control — set-points corrected on the machine, in real time.
  • Learn — the module twin compounds every run into better policy.
$26B
Module-line automation TAM
Serviceable across global capacity
3.2 W
Avg power recovered / module
From matching + defect avoidance
41%
Fewer field-critical microcracks
Versus open-loop baselines
9 mo
Typical payback per line
At Line-tier deployment
How it works

From raw wafer to graded module — under one policy

01

Interconnect

Ribbix controls the stringer bond in real time — thermal profile, pressure, dwell, and alignment — for sound, crack-free joints.

02

Inspect

Elumnex reads EL/PL and vision on every cell, segmenting defects and scoring crack-risk before damage propagates.

03

Laminate

Lamira models cure kinetics and closes the loop on temperature, vacuum, and dwell per material lot.

04

Match & simulate

Binova sorts by true power class while Twinvex simulates the module IV to lock in the target grade.

445W 450W 455W 460W 465W +3.2W avg / module recovered
Power-class bins, 445–465 W — watts recovered per module.

Recover watts that flash-only binning throws away

Binova fuses live IV and EL defect data to compose strings that minimize mismatch loss — re-optimizing continuously against your actual cell population and power targets. Every recovered watt is margin that compounds across gigawatts.

  • Bin on real defect state, not just flash class
  • Compose strings that cancel mismatch
  • Re-optimize live as the cell population drifts
Proof

What autonomy looks like on the line

“Panevex caught a lamination cure drift our SPC charts wouldn't have flagged for two shifts. That's a full lot of modules saved.”
PLLine 3 Process LeadTier-1 module maker (pilot)
“The recovered power alone paid for the deployment. The defect avoidance is what keeps our warranty team calm.”
VPVP of Manufacturing5 GW integrated producer
Built for the factory floor

Grounded, auditable, and real-time by design

Edge-native

Perception and control run on Jetson-class edge GPUs at the machine — no round-trip to the cloud in the control loop.

Auditable autonomy

Every set-point change is logged with the evidence that justified it. Nothing acts without a traceable reason.

MES-friendly

Integrates with your existing MES, SCADA, and equipment PLCs — Panevex augments the line, it doesn't replace it.

See Panevex run your line

Book a technical walkthrough with our manufacturing engineers. We'll map your stations to the agent stack and model the watts you're leaving on the table.

The autonomy stack

Every module, peak power.