基于新型Contention网络的滚动轴承早期故障诊断方法研究  

Research on rolling bearing early fault diagnosis based on a novel Contention neural network

作　　者：赵俊豪 郑煜 王英 王凯[4] Zhao Junhao;Zheng Yu;Wang Ying;Wang Kai(School of Mechanical Engineering,Shaanxi University of Technology,Shaanxi Hanzhong,723001,China;School of Mechanical Engineering,Shaanxi Polytechnic Institute,Shaanxi Xianyang,712000,China;Universities Engineering Research Center of Composite Movable Robot of Shaanxi Province,Shaanxi Xianyang,712000,China;School of Mechanical and Precision Instrument Engineering,Xi'an University of Technology,Shaanxi Xi'an,710048,China)

机构地区：[1]陕西理工大学机械工程学院,陕西汉中723001 [2]陕西工业职业技术学院机械工程学院,陕西咸阳712000 [3]复合型移动机器人陕西省高校工程研究中心,陕西咸阳712000 [4]西安理工大学机械与精密仪器工程学院,陕西西安710048

出　　处：《机械设计与制造工程》2024年第3期87-91,共5页Machine Design and Manufacturing Engineering

基　　金：国家自然科学基金(32060574);陕西工业职业技术学院科研基金(2023YKZX-013)。

摘　　要：针对滚动轴承早期故障诊断问题,为了同时建模振动信号中的高频和低频特征,实现高精度诊断,提出了一种新的模型Contention。它以一种互补的方式集成了空洞卷积和自注意力机制的优点,具有同时捕捉高频和低频信息的能力。为了验证其诊断能力,首先,在完整信息原则下将振动信号连续构造成数据集;其次,搭建Contention网络并训练,其最终测试集准确度可达100%,t-SNE显示随网络层次的深入特征被明显聚类;最后,设置对照实验,将该模型与传统RNN、CNN、Transformer模型对比,证明该模型具备突出的诊断能力。For the early fault diagnosis problem of rolling bearings,a new model named Contention is proposed to model both the high-frequency and low-frequency features of vibration signals for high accuracy diagnosis.It integrates the advantages of dilated convolution and attention in a complementary way,and has the ability to capture both high-frequency and low-frequency information simultaneously.To verify its diagnostic ability,the vibration signals are firstly divided into datasets using the complete information principle continuously.Then,the Contention network is properly built and trained,resulting in a final test accuracy of 100%.The t-SNE shows that the data is clearly clustered as the network level deepened.Finally,a controlled experiment is set to compare the model with the traditional RNN,CNN and Transformer models,and the experiment verifies that the model has outstanding diagnostic ability.

关 键 词：Contention网络 空洞卷积 自注意力机制 滚动轴承 早期故障诊断 

分 类 号：TP206.3[自动化与计算机技术—检测技术与自动化装置]