机械故障模式识别的ICA基神经网络方法

首先利用 ICA及基于残余互信息的二次特征抽取策略 ,进行不同机械状态模式 (包括正常和齿轮点蚀状态 )的特征提取 ,随后以此训练某一典型神经网络 (如多层感知器、径向基或自组织映射网络 ) ,以实现模式的最终分类。借助 ICA,隐藏于多通道振动观测中的高阶特征得以有效提取 ,从而实现机械状态模式的准确识别。对照分类实验结果表明 ,基于 ICA SOM分类方法不仅具有较好的故障模式分类能力 ,且实现简单 ,在机器运行状况监测中有较大的应用潜力。

ICA Based Neural Networks for Pattern Recognition of Mechanical Faults

Artificial neural network (ANN), especially the self-organizing map (SOM) based on unsupervised learning is a kind of excellent method for patterns clustering and recognition. Moreover, the independent component analysis (ICA) is a powerful tool for analyzing nongaussian data. In this paper, we used ICA and the further feature extraction strategies based on the residual mutual information (RMI) for feature extractions of different mechanical patterns (including normal and gear pitting), which were then used to train a typical ANN (for example MLP, RBF or SOM) for realization of the final classification. By means of ICA, the features higher than the second order embedded in multi-channel vibration measurements can be effectively captured to ensure that the mechanical fault patterns are correctly recognized. The results from the contrastive experiments show that the compound ICA-SOM classifier can be constructed in a simpler way, and classify various fault patterns at a considerable accuracy, both of which imply its great potential in health condition monitoring of machines.