基于MSSA+IESN+MFFN组合算法的齿轮箱早期故障智能诊断  

Early fault intelligent diagnosis of gearbox based on MSSA+IESN+MFFN combined algorithm

作　　者：冯贺平[1] 杨敬娜[1] 吴梅梅[2] 薛林雁[3] 王德永 FENG Heping;YANG Jingna;WU Meimei;XUE Linyan;WANG Deyong(Department of Intelligent Engineering,Hebei Institute of Software Technology,Baoding 071030,Hebei,China;Department of Network Engineering,Hebei Institute of Software Technology,Baoding 071030,Hebei,China;College of Quality and Technical Supervision,Hebei University,Baoding 071000,Hebei,China;Department of Design,Xinhuachang(Beijing)Electronic Technology Co.,Ltd.,Beijing 030002,China)

机构地区：[1]河北软件职业技术学院智能工程系,河北保定071030 [2]河北软件职业技术学院网络工程系,河北保定071030 [3]河北大学质量技术监督学院,河北保定071000 [4]芯华创(北京)电子科技有限公司设计部,北京030002

出　　处：《中国工程机械学报》2023年第2期172-177,共6页Chinese Journal of Construction Machinery

基　　金：河北省高等学校科学技术研究基金项目(ZD2022068);河北省教育厅青年基金项目(QN20131137);河北省教育厅项目(Z2017122)。

摘　　要：齿轮箱故障诊断存在变速工况、样本数量偏少以及会形成强噪声情况,提出了一种通过多尺度特征融合网络(MFFN)实现故障诊断技术。对初始时域信号拓展形成多特征域,建立造多维堆栈稀疏自编码器(MSSA)对不同特征域进行故障采集,通过粒子群算法优化回声状态网络(IESN)进行信号处理。研究结果表明:样本充足条件下,MFFN模型诊断时,定速工况为99.15%,变速工况为98.46%,达到了更高准确率并降低了标准差。在样本不足条件下,深度特征融合网络(DEFN)和MFFN对于样本数量减少表现出了优异鲁棒性,MFFN达到了更优的性能。在噪声干扰场景下,采用MFFN依然能够达到85%的准确率。该算法具备更优抗干扰性能,采用多维特征提取能够更好地适应处于强噪声干扰环境。该研究为实现传动系统的稳定运行提供了理论参考。Gearbox fault diagnosis is characterized by variable speed conditions,small number of samples and strong noise.A multi-scale feature fusion network(MFFN)is proposed to realize fault diagnosis.The initial time domain signal was expanded to form a multi-feature domain,and the multidimensional stack sparse autoencoder(MSSA)was built to collect faults in different feature domains,and the signal was processed through the echo state network(IESN)optimized by particle swarm optimization.The results show that:under the condition of sufficient samples,the MFFN model diagnosis,constant speed condition is 99.15%,variable speed condition is 98.46%,achieving higher accuracy and lower standard deviation.Under the condition of insufficient samples,the deep feature fusion network(DEFN)and MFFN show excellent robustness to the reduction of the number of samples,and MFFN achieves better performance.In noise interference scenes,MFFN can still achieve 85%accuracy.The algorithm has better anti-interference performance,and the multi-dimensional feature extraction method can better adapt to the environment with strong noise interference.The research can provide a theoretical reference for realizing the stable operation of the transmission system.

关 键 词：齿轮箱 故障诊断 深度学习 多堆栈稀疏自编码器(MSSA) 多尺度特征融合网络(MFFN) 

分 类 号：TH137[机械工程—机械制造及自动化]