应用Bootstrap方法构造机械故障特征库

An Application of Bootstrap Resampling Method to Knowledge Base Design in Machinery Fault Diagnosis

由于生产实际中往往无法采集到大量故障样本以建立故障特征库 ,本文应用 Bootstrap技术在小样本情况下建立故障特征库 ,讨论了如何根据各种指标在变工况下的稳定程度选择故障特征的问题。作为实例 ,本文以火车车轮滚动轴承在四种状态下振动信号的功率谱上 ,按谱峰的稳定程度选择作为特征指标 ,并应用所提出的方法对确定的指标进行统计模拟 ,构建了相应的故障特征库 ,说明了构造机械故障特征库的算法步骤 ,在实际中证明 :状态辨识良好。最后 ,本文将所提方法应用到旋转机械故障诊断中。实践证明 ,这种方法解决了在机械故障诊断特征指标的提取和置信区估计的问题 。

In machinery fault diagnosis, knowledge base design should be based on adequate data collection, which can not be implemented easily in practice. Therefore, bootstrap resampling method is introduced to establish the knowledge base using small samples. To evaluate the stability of selected fault features, a quantitative criterion characterizing the feature steadiness is presented. According to this criterion, appropriate features corresponding to their states are optimized and selected. Based on these extracted fault features, the bootstrap resampling method is used to design their state knowledge base. As an example, the vibration signals of a type of rolling bearing are processed and feature extracted to elucidate the algorithm. Then an example of fault diagnosis of rotating machinery is described. It is verified that the method is simple, feasible and effective.