双树复小波域隐Markov树模型降噪及在机械故障诊断中的应用

Denoising method based on hidden Markov tree model in dual tree complex wavelet domain and its application in mechanical fault diagnosis

提出一种基于双树复小波变换的隐Markov树模型的信号降噪方法,并将其成功应用于机械故障诊断中。机械设备的振动信号中不可避免地存在着噪声,使得微弱故障信息的提取一直是故障诊断的难点和热点。双树复小波变换具有近似平移不变性,而隐Markov树模型能有效刻画小波系数间的相关性和非高斯性,两种优势的结合可以获得比常规软、硬阈值小波降噪法和小波域隐Markov树模型降噪法更好的降噪效果。它不仅能有效抑制高斯白噪声,还能够去除异常冲击干扰,仿真信号验证了这一点。对于实际滚动轴承信号,使用该方法同样可以获得满意的结果。

Noise is inevitably present in mechanical vibration signal,which makes the extraction of weak fault information become the difficult point and hotspot of fault diagnosis.Since dual tree complex wavelet transform is of the property of approximate translation invariance while hidden Markov tree model can effectively describe the dependency between wavelet coefficients as well as the non-Gaussian nature of these coefficients,a method combining these advantages can achieve better denoising results than conventional soft or hard threshold denoising methods and hidden Markov tree model used alone in wavelet domain.Applications of simulation signals verify that Gaussian white noise can be effectively inhibited,and abnormal impact can be removed by using this method.For actual rolling bearing signal,satisfied results also can be acquired.