RUNNING LIGHTWEIGHT MODELS
Intelligence at the machine.
Fault inference from 5–10% of the signal—validated across 10 CWRU bearing conditions.
→
LIGHTSCLINESIUNCompact inference
→
CONDITION IDENTIFIEDNormalHealthy baseline
ON-SITE INFERENCE
Running on a
Jetson Nano.
AT THE MACHINE 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 results5–10%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 physics→Recognize multiple fault types→Make faster decisions at the machine