Distillation-enhanced fast neural architecture search method for edge-side fault diagnosis of wind turbine gearboxes

作者：

Highlights：

• Automatic model design is sped up using a multibranch neural architecture search.

• Lightweight and accurate searched models are balanced by auto-distillation.

• Lightweight requirements for edge devices are met by the designed models.

• The effectiveness was verified using simulated and measured cases.

摘要

•Automatic model design is sped up using a multibranch neural architecture search.•Lightweight and accurate searched models are balanced by auto-distillation.•Lightweight requirements for edge devices are met by the designed models.•The effectiveness was verified using simulated and measured cases.

论文关键词：Fault diagnosis,Lightweight model,Knowledge distillation,Neural architecture search,Wind turbine gearboxes,Edge devices

论文评审过程：Received 30 January 2022, Revised 15 June 2022, Accepted 1 July 2022, Available online 12 July 2022, Version of Record 16 July 2022.

论文官网地址：https://doi.org/10.1016/j.eswa.2022.118049