Lightscline
03 / EDGE INTELLIGENCE
← Smart sampling
RUNNING LIGHTWEIGHT MODELS

Intelligence at the machine.

Fault inference from 5–10% of the signal—validated across 10 CWRU bearing conditions.

SMART-SAMPLED INPUTFailure-mechanism physics retained
LIGHTSCLINESIUNCompact inference
CONDITION IDENTIFIEDNormalHealthy baseline
ON-SITE INFERENCE

Running on a
Jetson Nano.

Machine-side bearing monitoringAT THE MACHINE
EDGE DEVICE ONLINE

Bearing-to-decision intelligence runs beside the machine—without full raw-data backhaul.

THE MANUFACTURING PAYOFF

High-accuracy decisions without the 100%-data infrastructure.

CWRU validated results
5–10%

of raw data used

instead of moving every sample

96.0%

10-class accuracy

across CWRU bearing conditions

435×

fewer FLOPs

for radically lower compute

1000×

faster

than traditional ML models

Retain domain physicsRecognize multiple fault typesMake faster decisions at the machine