bearing

The paper presents BearingNAS, a hardware-aware neural architecture search framework that designs intelligent fault diagnosis systems for bearings, targeting microcontrollers and sensor processing units with extremely limited memory (4-8 KiB RAM, 16-32 KiB Flash) while running entirely on a laptop CPU and achieving 99.50% diagnostic accuracy.