摘要
为了准确识别机械设备当前所处的退化状态,研究了一种基于PCA(主成分分析)和CHMM(连续型隐马尔可夫模型)结合的性能退化状态识别方法。首先提取设备振动信号全寿命周期的时域、频域、时频域的特征,经过初步筛选后组成新的特征集,使用PCA方法对其进行降维处理;然后利用降维后的数据,训练一个全寿命周期CHMM用来确定退化状态数目,再针对每个退化状态训练一个CHMM,通过比较观测序列处于各个模型下的似然概率值判断设备当前所处的退化状态;最后通过实验对比了PCA+CHMM和PCA+SVM、PCA+KNN、PCA+CART方法的各退化状态识别准确率,结果表明PCA+CHMM的平均识别准确率最高、识别效果较好,适用于设备退化状态的识别。
In order to accurately identify the degradation state of mechanical equipment,this paper researched a recognition method of performance degradation state based on PCA( principal component analysis) and CHMM( continuous hidden Markov model). Firstly,it extracted the vibration signal’s time domain,frequency domain and time-frequency domain features in full life cycle,then constructed a new feature set by screening the features,then performed PCA dimensionality reduction for this set. Secondly,it trained a full life cycle CHMM to determine the number of degraded states by using of reduced dimension feature data,and then trained a CHMM for each degraded state,judged the degradation state of the device by comparing the likelihood probability of the observation sequence under each model. Finally,it compared the accuracy of PCA + CHMM and PCA + CHM,PCA + KNN and PCA + CART methods that to identify each degraded state. The results show that the average recognition accuracy of PCA + CHMM is the highest,the recognition effect is good,and it is suitable for the identification of device degraded state.
作者
钟飞
宁芊
周新志
赵成萍
Zhong Fei;Ning Qian;Zhou Xinzhi;Zhao Chengping(College of Electronics & Information Engineering,Sichuan University,Chengdu 610065,China;Science & Technology on Electronic Information Control Laboratory,Chengdu 610036,China)
出处
《计算机应用研究》
CSCD
北大核心
2019年第1期136-139,共4页
Application Research of Computers
基金
国家"973"计划资助项目(2013CB328903)
关键词
性能退化
主成分分析
连续隐马尔可夫模型
特征降维
退化状态识别
performance degradation
principal component analysis
continuous hidden Markov model
feature dimensionality reduction
degenerate state recognition
