• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2024, Vol. 60 ›› Issue (12): 65-76.doi: 10.3901/JME.2024.12.065

• 特邀专栏:可解释可信AI驱动的智能监测与诊断 • 上一篇    下一篇

扫码分享

一种面向机械设备故障诊断的可解释卷积神经网络

陈钱1,2, 陈康康1,2, 董兴建1,2, 皇甫一樊1,2, 彭志科1,2,3, 孟光1,2   

  1. 1. 上海交通大学机械系统与振动国家重点实验室 上海 200240;
    2. 上海交通大学振动、冲击、噪声研究所 上海 200240;
    3. 宁夏大学机械工程学院 银川 750021
  • 收稿日期:2023-07-01 修回日期:2024-01-29 出版日期:2024-06-20 发布日期:2024-08-23
  • 作者简介:陈钱,男,1999年出生,博士研究生。主要研究方向为旋转机械智能诊断的可解释性。E-mail:chenqian2020@sjtu.edu.cn;董兴建(通信作者),男,1977年出生,博士,副教授,博士研究生导师。主要研究方向为振动分析与控制、振动超材料和结构疲劳分析。E-mail:donxij@sjtu.edu.cn
  • 基金资助:
    国家自然科学基金资助项目(12272219,12121002)。

Interpretable Convolutional Neural Network for Mechanical Equipment Fault Diagnosis

CHEN Qian1,2, CHEN Kangkang1,2, DONG Xingjian1,2, HUANGFU Yifan1,2, PENG Zhike1,2,3, MENG Guang1,2   

  1. 1. State Key Laboratory of Mechanical System and Vibration, Shanghai Jiao Tong University, Shanghai 200240;
    2. Institute of Vibration, Shock and Noise, Shanghai Jiao Tong University, Shanghai 200240;
    3. School of Mechanical Engineering, Ningxia University, Yinchuan 750021
  • Received:2023-07-01 Revised:2024-01-29 Online:2024-06-20 Published:2024-08-23

摘要: 卷积神经网络(Convolutional neural network, CNN)以其强大的特征提取和分类能力,被广泛应用于机械系统故障诊断任务中。但CNN是一个典型的“黑箱模型”,其决策机理和分类依据并不明确。这不仅降低了智能诊断结果的可信性,还限制了在高可靠性要求故障诊断中的应用。针对这一问题,具有物理意义的Chirplet变换被引入到传统卷积层中,形成具有优异可解释性的Chirplet卷积层和Chirplet-CNN,进而提出将Chirplet-CNN用于故障诊断的完整流程。一系列试验表明,Chirplet-CNN以其提取时频特征的特点,不仅拥有和当前先进方法相近的优异故障诊断能力,而且在可解释性方面具有突出表现,即能够通过频谱分析对CNN提取类别特征和做出判断的频带依据进行解释。此外,进一步的分析结果表明,所提出的Chirplet卷积层具有良好的通用性,与不同深度的CNN模型进行组合,均能有效提高其诊断精度并获得不错的解释结果。

关键词: 卷积神经网络, 可解释性, Chirplet变换, 时频变换, 深度学习, 故障诊断

Abstract: Convolutional neural network(CNN) has been widely used in mechanical system fault diagnosis because of its powerful feature extraction and classification capabilities. However, CNN is a typical“black box model”, and the mechanism of CNN’s decision-making is not clear, which not only reduces the credibility of intelligent diagnosis results but also limits the application in fault diagnosis with high-reliability requirements. Facing this problem, the physically meaningful chirplet transform (CT) is introduced into the traditional convolutional layer to formulate the chirplet convolutional layer and Chirplet-CNN with the complete process of using Chirplet-CNN for fault diagnosis. A series of experiments show that Chirplet-CNN has excellent fault diagnosis ability similar to the current state-of-the-art methods, and has outstanding performance in interpretability. It can interpret the frequency band basis for CNN to extract category features and make judgments through spectrum analysis. In addition, the proposed chirplet convolutional layer has good generality and when combined with CNN models of different depths, it can effectively improve the diagnostic accuracy and obtain good interpretation results.

Key words: CNN, interpretability, Chirplet transform, time-frequency transform, deep learning, fault diagnosis

中图分类号: