摘要
针对齿轮故障诊断中单一传感器采集信息不完全、容错性不佳及一种神经网络模型具有局限性,传统信号处理技术提取特征困难等问题,提出了多深度学习模型决策融合的齿轮箱故障诊断分类方法,构建了基于卷积神经网络(convolutional neural networks,CNN)和改进堆叠降噪自动编码器(stacked denoising autoencoders,SDAE)的混合网络模型,根据改进的Dempster-Shafer(D-S)证据理论实现决策级融合诊断。以时频信号作为CNN的输入,以频域信号作为SDAE的输入,采用Adam优化算法和dropout、批量归一化技术训练该混合模型。实验结果表明:利用该融合方法对齿轮进行故障诊断相比单个的网络模型CNN和SDAE诊断正确率有所提高,为齿轮故障智能诊断分类提供了新路径。
Aiming at the problems of incomplete information collected by a single sensor,poor fault tolerance,limitations of a neural network model and difficulty in extracting features by traditional signal processing technology,a gearbox fault diagnosis and classification method based on multi-depth learning model decision fusion was proposed,and a hybrid network model based on conventional neural networks(CNN)and improved stacked denoising autoencoders(SDAE)was constructed.According to the improved Dempster-Shafer(D-S)evidence theory,the decision-level fusion diagnosis was realized.The time-frequency signal was used as the input of CNN and the frequency-domain signal as the input of SDAE.Adam optimization algorithm,dropout and batch normalization technique were used to train the hybrid model.Experimental results show that the accuracy of gear fault diagnosis based on this fusion method is higher than that of single network model CNN and SDAE,which provides a new path for intelligent diagnosis and classification of gear faults.
作者
陈科
段伟建
吴胜利
邢文婷
CHEN Ke;DUAN Wei-jian;WU Shen-li;XING Wen-ting(College of Traffic and Transportation, Chongqing Jiaotong University, Chongqing 400074, China;Chongqing Traffic Management Bureau, Chongqing 400054, China;Schoool of Management Science and Engineering, Chongqing Technology and Business University, Chongqing 400067, China)
出处
《科学技术与工程》
北大核心
2022年第12期4804-4811,共8页
Science Technology and Engineering
基金
国家自然科学基金(51705052)
重庆市自然科学基金(cstc2019icyj-msxmX0779)
国家社会科学基金(17CGL003)。
