Battery-Powered Excursion Detection That Survives 14 Days Off-Grid
Cold-Chain Integrity Recorder
16 days
battery life on a single primary cell
87%
reduction in false excursion alerts
Smart Manufacturing / Quality Assurance
Inline Weld Seam Inspection Cell
Every Weld Graded in 38 ms, On the Line, Without a Network
Smart Manufacturing
Industry
9 months
Duration
4 → 9
TRL
4
Disciplines
Background
A tier-two automotive fabricator inspected robotic MIG welds by sampling - an operator pulled roughly one part in forty to a light booth. Defects that escaped sampling surfaced at the customer, and each containment event cost more than a week of production margin.
The problem
What made it hard.
Inspection had to finish inside the robot cell's 45 ms index window or it would slow the line, which ruled out sending frames to a server. Weld appearance shifted constantly with spatter, tip wear and shielding gas flow, so a fixed-threshold vision recipe drifted within a shift. The plant also refused any inspection system that required opening the OT network.
Approach
What we built.
We built a self-contained cell: two GigE cameras on a rigid frame with controlled ring lighting, feeding a Jetson Orin NX running a segmentation model quantised to INT8 through TensorRT. The model outputs a seam mask and a per-millimetre defect class - porosity, undercut, burn-through, insufficient fill. Inference lands at 38 ms end to end, inside the index window, with the verdict pushed to the cell PLC over OPC UA as a simple pass/divert bit. Nothing leaves the cell. A local ring buffer keeps the last 5,000 part images so quality engineers can retrain against real escapes; retraining runs offline and ships back as a signed model bundle on removable media.
Outcome
What it measured.
38 ms
inference latency, inside the 45 ms index window
100%
of welds inspected, up from 2.5% sampling
21 ppm
escaped defects, down from 340 ppm
The cell moved the fabricator from 2.5% sampling to 100% inspection with no cycle-time penalty. Escaped-defect rate to the customer fell from 340 ppm to 21 ppm over the first two quarters, and the plant retired two of three light-booth stations. Because the verdict is a PLC bit rather than a network call, the system passed the customer's OT security audit without an exception.
Our role
Optical design; cell mechanics; model development and quantisation; PLC integration; commissioning and operator training.
Technologies
NVIDIA Jetson Orin NXTensorRTGigE VisionPyTorchOPC UAC++Profinet
Gallery
Inside the build.
Illustrative of the inspected feature: bead geometry, undercut and spatter are the visual characteristics graded on every weld rather than on a 2.5% sample.
Illustrative of the edge compute class used: the TensorRT model runs on a Jetson Orin NX module inside the cell cabinet, keeping the 38 ms inference entirely off the plant network.
Illustrative of commissioning work at the cell: reference images were recollected as consumable wear and shift lighting changed weld appearance, and results were handed to the PLC over OPC UA and Profinet.
Next case study
Tell us what you’re building.
Send the constraint that worries you most - a latency budget, a power budget, a certification date. We’ll tell you straight whether we’re the right team.