基于信息融合的支撑座早期松动故障诊断  

Early Loosening Fault Diagnosis of Clamping Support Based on Information Fusion

作　　者：孙卫祥[1] 陈进[1] 伍星[1] 董广明[1] 宁佐贵[2] 王东升[2] 王雄祥[2] 

机构地区：[1]上海交通大学振动冲击噪声国家重点实验室,上海200240 [2]中国工程物理研究院结构力学研究所,绵阳621900

出　　处：《上海交通大学学报》2006年第2期239-242,247,共5页Journal of Shanghai Jiaotong University

基　　金：国家自然科学基金资助项目(50335030;10176014);"十五"科技攻关资助项目(2001BA204B06)

摘　　要：采用基于信号分析的无模型检测方案和信息融合技术,对支撑座早期松动故障进行检测诊断.针对支撑座松动的小波包变换特征和功率谱特征进行特征融合与决策融合,同时采用基于熵度量的无监督特征约简方法对功率谱特征进行约简,有效地减少了特征数目,加快了融合和诊断速度.特征融合与决策融合采用分层神经网络实现,该网络综合了局部融合和全局融合的优点,具有很高的故障确诊率和很好的抗噪性能,无噪声样本综合确诊率达94.3%,有噪声样本综合确诊率达88.6%.A novel global Non-Destructive Evaluation（NDE） technique based on information fusion was proposed to diagnose early loosening fault of clamping support. It is a kind of non-modeled method. Two feature extraction methods are used to extract feature, which are wavelet packet （WP） transform and power spectrum density （PSD） analysis based on FFT. During the loosening fault diagnosis, two local decisions are made by using WP feature and PSD feature respectively. Then the two features are fused to make another local decision. After that the three local decisions are fused to make global decision. A hierarchical neural network structure is put forward to implement feature fusion and decision fusion. The network has advantanges of both local fusion and global fusion, also it has high correct diagnosis ratio and good antinoise performance. Using the information fusion network, the correct diagnosis ratios of exemplars with no noise and with random noise reach 94.3 % and 88.6 % respectively.

关 键 词：故障诊断 信息融合 支撑座 松动 特征约简 分层神经网络 

分 类 号：TH113.1[机械工程—机械设计及理论] TP206.3[自动化与计算机技术—检测技术与自动化装置]