滚动轴承多工况故障的特征自动选择核极限学习机智能识别方法

Intelligent identification method using kernel extreme learning machine for rolling bearing multi-working condition multi-feature automatic selection

滚动轴承的智能诊断存在许多不足,特别是对复杂工况下的识别存在特征提取不足及诊断精度低等问题。针对故障类型不同、故障程度不同和负荷不同的多工况条件下滚动轴承故障诊断,提出了基于多特征自动选择的核极限学习机智能识别方法。分别从时域、频域、时频域提取有效故障特征,采用拉普拉斯分数(Laplace Score,LS)根据每个特征的重要性自动选择敏感特征,旨在消除一些冗余信息,提高计算效率。采用模拟退火粒子群优化的核极限学习机(Simulated Annealing Particle Swarm Optimization,Kernel Extreme Learning Machine,SAPSO-KELM),实现滚动轴承多故障状态识别。将该方法应用于滚动轴承变负荷故障识别,与其他识别方法的比较结果表明,该方法具有较高的识别精度和较快的分类速度。

There are many deficiencies in intelligent fault diagnosis of rolling bearing,especially,for fault diagnosis under complex working conditions,there are problems of insufficient feature extraction and low diagnosis accuracy.Here,aiming at fault diagnosis of rolling bearing under multi-working condition with different fault types,different levels of fault and different loads,an intelligent identification method using kernel extreme learning machine based on multi-feature automatic selection was proposed.Firstly,effective fault features were extracted in time domain,frequency one and time-frequency one,respectively.Secondly,Laplace score(LS)was adopted to automatically select sensitive features according to the importance of each feature,the goal was to eliminate some redundant information and improve computational efficiency.Finally,the kernel extreme learning machine with simulated annealing particle swarm optimization was adopted to realize multi-fault state recognition of rolling bearing.The method was applied in fault identification of rolling bearing under variable loads.It was shown that compared with other identification methods,the proposed method has higher recognition accuracy and faster classification speed.