基于信号时频特性与GA-SVM的液压泵故障诊断方法

Fault diagnosis method of hydraulic pump based on time-frequency characteristic of signal and GA-SVM

为提高液压泵故障诊断的准确率和速度,提出一种小波包频带能量结合信号时域统计量方差和均方根值的信号特性表示方法,以及一种用于寻找支持向量机最优惩罚因子和径向基核函数模型参数的实值编码遗传算法。实验结果表明这种信号特性表示方法能够很好地展示液压泵不同工作状态下的特征,使不同状态下的信号具有明显的区分度。通过与幂级数分格的网格搜索法对比,验证了实值编码的遗传算法能够有效且快速地找到支持向量机的最优参数。

To improve the accuracy and speed of hydraulic pump fault diagnosis, a signal characteristic representation method based on wavelet packet band energy, signal variance and mean square value, and a real-coded genetic algorithm used for seeking for the optimum penalty factor of support vector machine and the model parameter of the radical-base kernel function were proposed. The experimental results show that such signal characteristic representation method can reveal the characteristics of the hydraulic pump well under different working states, apparently distinguish the signals under different states. By comparison with the grid search method with power series of grid division, the optimal parameters of support vector machine can be found effectively and quickly by adopting the real-coded genetic algorithm.