Factory edge
Jetson Orin / Thor units run perception and closed-loop control at the machine with deterministic latency.
Technology
Physical AI, engineered for the factory floor
Panevex runs perception and control where the modules are made. A hybrid edge-and-cloud architecture keeps the control loop real-time while the cloud trains the models and runs the twin.
Jetsonedge perception + control
DGX/HGXtraining + twin
OVXsimulation
Architecture
Edge and cloud, each doing what it's best at
Training cloud
DGX / HGX clusters train perception and policy models on fleet-scale data.
Simulation
OVX / RTX runs the module digital twin — stringing stress, cure, and IV — before any physical run.
Sensor fusion built for silicon
Panevex fuses electroluminescence, photoluminescence, and RGB vision into a single per-cell model. Fusion is what separates a real defect from a reflection, and a cosmetic mark from a crack that will grow.
- EL + PL + RGB per cell and string
- Sub-cell segmentation and classification
- Robust to line-speed motion and glare
Physics-aware models, not black boxes
Every prediction is grounded in device physics — crack mechanics, cure kinetics, and IV behavior. That's what makes the autonomy auditable: each action carries the physical reason that justified it.
- Crack-propagation mechanics
- Encapsulant cure kinetics
- IV / power simulation
The twin
Simulate before you run
Twinvex — the module digital twin
Before a recipe touches real silicon, Twinvex simulates the stringing stress, lamination cure, cell shift, and resulting module IV. Set-points are optimized in simulation and validated against physics, then deployed to the edge.
- Recipe optimization in simulation
- Control-policy validation before deployment
- Closes the sim-to-real loop with field outcomes
| Component | Hardware class | Role | Latency |
|---|---|---|---|
| Perception edge | Jetson Orin | EL/PL/RGB inference | in-line |
| Control edge | Jetson Thor | Closed-loop actuation | < 10 ms |
| Model training | DGX / HGX | Perception + policy | batch |
| Digital twin | OVX / RTX | Simulation | pre-run |
| Fleet plane | Cloud | Orchestration + audit | async |
Engineered for trust
Autonomy you can defend to an auditor
Traceable actions
Every set-point change stores the perception and physics evidence that produced it.
Guardrailed control
Human-set bounds on every actuator; low-confidence decisions hold and escalate.
Deterministic edge
Control decisions are bounded-latency and run locally, resilient to network loss.
EL / PL imagingRGB visionSensor fusionSemantic segmentationCrack mechanicsCure kineticsIV simulationClosed-loop controlDigital twinFleet learningEdge inferenceMES / SCADA sync
Panevex is hardware-agnostic on the machine side — we augment your existing stringers, laminators, and handling rather than forcing a rip-and-replace.
Go deep with our engineers
Bring your hardest perception and control problem. We'll show you how the stack handles it — edge to twin.