An end-to-end framework combining time–frequency expert knowledge and modified transformer networks for vibration signal classification
作者：
Highlights：
• The proposed end-to-end framework combined expert knowledge and deep neural networks.
• The Transformer network was modified to fit time–frequency features extraction.
• Multi-sensor fusion improved the performance without much extra computation cost.
• Joint optimization was realized throughout the framework to avoid human intervene.
• The proposed framework has potential broad applicability on different vibration signals.
摘要
•The proposed end-to-end framework combined expert knowledge and deep neural networks.•The Transformer network was modified to fit time–frequency features extraction.•Multi-sensor fusion improved the performance without much extra computation cost.•Joint optimization was realized throughout the framework to avoid human intervene.•The proposed framework has potential broad applicability on different vibration signals.
论文关键词：Vibration signal classification,Time–frequency feature extraction,modified Transformer network,Multi-sensor fusion
论文评审过程：Received 20 October 2019, Revised 10 November 2020, Accepted 2 January 2021, Available online 6 January 2021, Version of Record 24 January 2021.
论文官网地址：https://doi.org/10.1016/j.eswa.2021.114570
