基于独立分量分析特征提取的复合神经网络故障诊断法

Multi-Neural Networks for Faults Diagnosis Based on ICA Feature Extraction

首先利用基于固定点迭代的快速算法(FASTICA)提取不同机械状态模式(包括正常、齿轮故障及机座松动)特征,随后以此训练某一典型神经网络(如径向基网络或自组织映射网络),以实现模式的最终分类。借助独立分量分析(ICA)及基于残余互信息(RMI)的二次特征抽取策略,隐藏于多通道振动观测中的高阶特征得以有效提取,进而实现机械状态模式的准确识别。对照分类实验结果表明,基于无导师学习的自组织映射(ICA-SOM)分类方法不仅具有较好的故障模式分类能力,且实现简单直观,在机器健康状况监测中有较大的应用潜力。

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. Independent component analysis (ICA) is a powerful tool for analyzing nongaussian data. In ICA, the FASTICA based on fixed-point iteration is a kind of ANN algorithm with high efficiency, which is specially appropriate to feature extraction of multivariate data at real time. In this paper, the FASTICA is firstly proposed for feature extraction of different mechanical patterns (including normal, gear fault and loose foundation), followed by certain typical ANN (for example RBFN or SOM), which implements the final classification. By means of ICA and the further feature extraction strategy based on residual mutual information (RMI), higher than second order features embedded in multi-channel vibration measurements can be captured effectively. Thus, mechanical fault patterns can be recognized correctly. The results from contrast experiments show that the compound ICA-SOM classifier can be constructed in simpler way, and classify various fault patterns at considerable accuracy, both of which imply its great potential in health condition monitoring of machines.