摘要
提出了一种新颖的、基于独立分量分析 ( ICA)的复合神经网络 ,用于不同机械状态模式的特征提取。利用支持向量机 ( SVM)进行最终分类。与通常的基于经验风险最小化 ( ERM)原理的神经网络方法相比 ,基于结构风险最小化 ( SRM)原理的支持向量机分类方法具有更好的推广能力。而借助多个独立分量分析网络 ,隐藏于多通道振动观测信号中的不变特征得到有效提取 ,从而实现了支持向量机分类器在分类能力和推广性两者间的合理平衡。
A novel multi- neural network based on Independent Component Analysis (ICA) was proposed for feature extraction of different mechanical condition modes,followed by a Support Vector Machine (SVM) which implements the final classification.The SVM based on the principle of SRM has a greater ability to be generalized, compared with the traditional neural network based on Empirical Risk Minimization (ERM).On the other hand, invariable features embedded in multi-channel vibration measurements can be captured by the use of multi-ICA networks. Thus, stable SVM classifier was constructed, which can find a good balance between the classification performance and the generalization power. The classification results imply great potential of the new compound ICA-SVM classifier in health condition monitoring of machines, by comparison with a Multi-Layer Perceptron (MLP) classifier.
出处
《中国机械工程》
EI
CAS
CSCD
北大核心
2004年第1期62-65,共4页
China Mechanical Engineering
基金
国家自然科学基金资助项目(50205025)
浙江省自然科学基金资助项目(5001004)
关键词
独立分量分析
残余总体相关
经验风险最小化
结构风险最小化
independent component analysis (ICA)
residual total correlation (RTC)
empirical risk minimization (ERM)
structural risk minimization (SRM)
