BearingNAS: Obtaining In-Sensor Intelligent Fault Diagnosis Systems for Bearings Using a Laptop

Short summary

BearingNAS is a hardware-aware neural architecture search framework that shifts fault-diagnosis intelligence onto the sensor die, targeting extreme micro-budgets of 4-8 kiB RAM and 16-32 kiB Flash. Using a derivative-free search strategy running entirely on a laptop CPU, it converges in under an hour and achieves 99.50% diagnostic accuracy on the LSM6DSO16IS intelligent sensor processing unit. Results demonstrate viable low-cost, production-scale bearing fault diagnosis inside the sensor package.

• •HW-NAS framework targeting 4-8 kiB RAM and 16-32 kiB Flash microcontrollers for in-sensor fault diagnosis
• •Derivative-free search runs on a laptop CPU and converges in under an hour
• •Achieves 99.50% diagnostic accuracy on STMicroelectronics LSM6DSO16IS ISPU

Generated with AI, which can make mistakes.