基于CNN和WSVDD的深孔镗削加工过程状态评估与监测  

Condition Assessment and Monitoring for Deep-hole Boring Process Based on CNN and WSVDD

作　　者：李欣欣 周学良 鲍武 Li Xinxin;Zhou Xueliang;Bao Wu(School of Mechanical Engineering,Hubei University of Automotive Technology,Shiyan 442002,China)

机构地区：[1]湖北汽车工业学院机械工程学院

出　　处：《湖北汽车工业学院学报》2019年第2期40-46,共7页Journal of Hubei University Of Automotive Technology

基　　金：湖北省自然科学基金(2016CFB402);湖北汽车工业学院博士科研启动基金(BK201601);湖北汽车工业学院研究生创新基金(Y2017307)

摘　　要：针对深孔镗削加工过程中易出现颤振以及传统监测方法监测效率低的问题,提出了一种将深度卷积神经网络和加权支持向量数据描述法相结合的深孔镗削加工过程颤振评估与监测方法。首先基于深度卷积神经网络提取加工过程状态内振动信号的特征矢量,然后以正常加工状态下的特征矢量训练加权支持向量数据描述模型,得到相应的描述加工正常状态下的超球体模型,再计算当前加工过程状态特征矢量与超球体之间的相对距离,作为加工过程状态的评估指标,并对稳定度阈值进行设定。结果表明:与文献中其他方法相比,文中方法的监测能力更强,对加工过程状态的稳定度描述更加准确。Aiming at the problem of chatter which is easy to occur in deep hole boring process and the low monitoring efficiency of traditional monitoring methods, a method of chatter evaluation and monitoring in deep-hole boring process was proposed, which combines deep convolution neural network with weighted support vector data description method. Firstly, the feature vectors of vibration signals in the process state were extracted based on deep convolution neural network. Then, the weighted support vector data description model was trained by the feature vectors in the normal process state, and the corre? sponding hypersphere model describing the normal process state was obtained. The relative distance be? tween the current process state feature vector and the hypersphere was calculated which was used as an evaluation index of the process state, and the stability threshold was set. The results show that the meth? od has stronger monitoring ability and more accurate description of the process state stability comparing with other methods in the references.

关 键 词：深孔镗削 颤振监测 深度卷积神经网络 加权支持向量数据描述 

分 类 号：TH165.4[机械工程—机械制造及自动化]