摘要
噪声情况下精确地对齿轮箱进行故障诊断是齿轮箱故障诊断的难题。为了解决该难题,采取自适应小波对自适应噪声完全集合经验模态分解(CEEMDAN)分量进行分解降噪与重组,并提出卷积神经网络(CNN)结合Inception模块的一维卷积神经网络(BICNN)提取重构信号的基本数字特征,同时使用长短期记忆提取BICNN所提取到的特征之间的相关性特征,对齿轮箱进行故障诊断研究。诊断结果表明:所提出的方法具有较高的抗噪声能力,并且齿轮箱在受到-4 dB噪声干扰的情况下,所提出的方法仍然可以获得99.63%的训练精度。
Accurate fault diagnose of gearbox under noise condition is a difficult problem in gearbox fault diagnosis.In order to solve this problem,the noise reduction method of decomposing and reorganizing complete ensemble empirical mode decomposition with adaptive noise analysis(CEEMDAN)by adaptive wavelet was adopted,and the convolution neural network based on inception(BICNN)was put forward to extract the basic digital characteristics of the reconstructed signal and long short-term memory(LSTM)was adopted to extract the correlation features among the features extracted by BICNN.The method was used to study the fault diagnosis of the gearbox.The diagnosis results show that the proposed method has high anti-noise ability,and it can still obtain 99.63%training accuracy when the gearbox is disturbed by-4 dB noise.
作者
蔡超志
白金鑫
池耀磊
张仲杭
CAI Chaozhi;BAI Jinxin;CHI Yaolei;ZHANG Zhonghang(School of Mechanical and Equipment Engineering,Hebei University of Engineering,Handan Hebei 056038,China)
出处
《机床与液压》
北大核心
2022年第24期171-180,共10页
Machine Tool & Hydraulics
基金
河北省自然科学基金项目(E2020402060)。
关键词
自适应小波分解与重构
自适应噪声完全集合经验模态分解
卷积神经网络
长短期记忆
抗噪声能力
齿轮箱故障诊断
Adaptive wavelet decomposition and reconstruction
Complete ensemble empirical mode decomposition with adaptive noise analysis
Convolution neural network
Long short-term memory
Anti-noise capability
Gearbox fault diagnosis
作者简介
蔡超志(1982-),男,博士,副教授,研究方向为基于深度学习理论的机械故障诊断研究。E-mail:caichaozhi1983@163.com。
