基于多目标交叉熵优化的轮对轴承故障特征提取方法

Fault Feature Extraction of Wheel-bearing Based on Multi-objective Cross Entropy Optimization

从复杂干扰中提取轮对轴承的故障特征,需要同时从冲击性和循环平稳性两个方面考虑。而在提取过程中,单一的指标很难统一并平衡二者的权重。为此,提出一种基于多目标优化的非对称实Laplace小波解调方法。在第一个目标中,以最大化平方包络的峭度为适应度函数,利用窄带信号包络的时域稀疏性表征故障的冲击性特征。在第二个目标中,以最大化平方包络谱的峭度为适应度函数,利用窄带信号包络的频域稀疏性表征故障的循环平稳性特征。并通过将非支配排序和拥挤距离排序引入交叉熵算法实现了多目标最优非对称实Laplace小波解调以降低冲击性噪声或循环平稳性噪声的干扰。试验结果表明,该方法可实现复杂干扰下轮对轴承的故障特征提取,并通过与单目标优化方法的对比分析,验证了该方法的优越性。

It is crucial to take impulsiveness and cyclostationarity into account simultaneously in the fault features extraction of wheel-bearing, especially with the occurrence of complex interferences from wheel-rail contact relation. However, these two aspects can hardly be synthesized and balanced by one index. Therefore, a novel multi-objective optimized anti-symmetric real Laplace wavelet filtering method is proposed to deal with this problem. The first fitness function is designed by maximizing the kurtosis value of squared envelope, which is representing the impulsiveness. And, the second fitness function is designed by maximizing the kurtosis value of squared envelope spectrum, which is representing the cyclostationarity. Through combining the non-dominated sorting and crowded comparison, the parameters of the wavelet filter are optimized by the improved cross entropy algorithm, which is immune to impulsive or cyclostationary noises. One vibration signal from a faulty wheel-bearing is investigated to illustrate the effectiveness of the proposed method, and some comparisons with the single-objective method are also conducted to show its superiority in extracting the repetitive transients.