One-shot neural architecture search for fault diagnosis using vibration signals

作者：

Highlights：

• A one-shot neural architecture search method is applied for fault diagnosis.

• The one-shot method contains supernet training and evolutionary search two stages.

• Cosine performance metrics and weights reuse strategy are proposed.

• The proposed method can find optimal networks and achieve state-of-the-art results.

摘要

•A one-shot neural architecture search method is applied for fault diagnosis.•The one-shot method contains supernet training and evolutionary search two stages.•Cosine performance metrics and weights reuse strategy are proposed.•The proposed method can find optimal networks and achieve state-of-the-art results.

论文关键词：Neural architecture search,Fault diagnosis,Vibration signals,One-shot model

论文评审过程：Received 22 November 2020, Revised 13 August 2021, Accepted 1 October 2021, Available online 6 November 2021, Version of Record 10 November 2021.

论文官网地址：https://doi.org/10.1016/j.eswa.2021.116027