基于隐马尔科夫模型和支持向量机的曲面加工颤振识别与预报  

Chatter Recognition and Prediction for Curve Surface Processing Based on HMM and SVM

作　　者：李欣[1,2] 邓小雷 张玉良[1] 余建平 LI Xin;DENG Xiaolei;ZHANG Yuliang;YU Jianping(Key Laboratory of Air-Driven Equipment Technology of Zhejiang Province, Quzhou University, Quzhou 324000, China;College of Mechanical Engineering, Zhejiang University, Hangzhou 310027, China)

机构地区：[1]衢州学院浙江省空气动力装备技术重点实验室,衢州324000 [2]浙江大学机械工程学院,杭州310027

出　　处：《航空制造技术》2019年第6期14-20,共7页Aeronautical Manufacturing Technology

基　　金：国家自然科学基金(51505253;51605253;51876103);浙江省公益技术研究计划项目(LGG19E050013);浙江省博士后择优资助项目(ZJ20180077)

摘　　要：针对曲面加工过程容易产生加工颤振而导致表面加工质量降低的问题,提出一种基于隐马尔科夫模型和支持向量机(HMM-SVM)的颤振早期识别与预报方法。首先,根据曲面加工颤振发展较快,孕育阶段时间短,难以与正常加工及颤振爆发阶段区分的现象,结合HMM模型有较强的相似性归类能力和SVM有较强的二类分类能力的特点,设计了基于HMM-SVM混合模型的颤振识别与预报系统;其次,应用加速度传感器采集曲面加工过程中刀具振动信号,完成反映加工状态的特征信号的获取;最后,利用HMM、HMM-SVM分别进行曲面加工状态识别试验,并进行结果分析与比较。试验结果表明:与单独使用HMM模型相比,基于HMM-SVM混合模型可以大大提高识别准确率,3种加工状态识别率均达95%以上,并具有较好的识别实时性,识别时间1.5s以内,可实现颤振快速识别与预报,为后续颤振抑制环节提供依据和保证。Chatter occurs frequently during the curve surface machining process, and it results in poor quality of finished surface. In order to identify chatter quickly and accurately, a method based on hidden Markov model (HMM) and support vector machine (SVM) for chatter recognition and prediction was proposed in this paper. Firstly, according to the phenomenon that the transition period of formation process of the curve surface machining chatter is short and difficult to distinguish from normal processing and chatter burst stages, a chatter identification and prediction system based on HMM– SVM hybrid model was designed, which combined the strong similarity classification ability of HMM and the strong classification ability of SVM. Then, the acceleration sensor was used to measure the tool vibration signal during the curve surface machining process, and the characteristic signals of machining states was obtained. Finally, HMM and HMM– SVM were used to carry out recognition experiments of curve surface machining state, and the results were analyzed and compared. The experimental results show that the proposed HMM–SVM method drastically improve the recognition accuracy rate, compared with HMM model alone. The recognition accuracies of the three processing states are over 95%, and the recognition time is less than 1.5s. Rapid identification and prediction of chatter are realized, which provide basis and guarantee for the subsequent chatter suppression.

关 键 词：颤振预报 状态识别 曲面加工 隐马尔科夫模型 支持向量机 

分 类 号：TH161.6[机械工程—机械制造及自动化] V261[航空宇航科学与技术—航空宇航制造工程]