基于极点对称模态分解和概率神经网络的轴承故障诊断

摘要

针对复杂非线性的滚动轴承系统,提出了极点对称模态分解(ESMD)和概率神经网络(PNN)相结合的滚动轴承故障诊断方法.ESMD将固有模态函数的定义进行扩充,采用内部极点对称直接插值的方法替代外部包络线插值,引入最优的自适应全局曲线(AGM)的概念优化分解的趋势线,并由此确定最佳的模态分解次数.PNN是一种基于核函数逼近的神经网络分类器,将指数函数引入神经网络用来替代S型激活函数并进行重新构造,突出体现了梯度最速下降法的概念,减少实际和预测的输出函数之间的误差.通过对经验模态分解(EMD)、屏蔽经验模态分解(MEMD)和ESMD方法进行信号仿真分解对比,以及采用ESMD和PNN对故障数据进行处理,结果表明,该方法能够更加有效地对故障信号进行识别.%Aimed at the complex non-linear rolling bearing systems,a new method combining ES-MD and PNN was introduced for bearing fault diagnosis.ESMD expanded the definition of the intrin-sic mode function,and changed the external envelope interpolation to internal pole symmetric interpo-lation.An idea of adaptive global mean(AGM)was used to optimize the last remaining modal,thus to determine the optimal number of decomposition.PNN was a neural network classifier based on kernel function approximation.The exponential functions were introduced to the neural network to replace the S type activation functions and to reconstruct the functions,representing the notion of gradient steepest descent method prominently,and reducing the errors between the actual and predicted output functions.Through the decomposing comparison to the simulation signals among empirical mode de-composition(EMD),making empirical mode decomposition(MEMD)and ESMD,and the diagnosis to the bearing vibration data by ESMD and PNN shows that the new method introduced may diagnose the bearing faults more effectively.