基于量子高斯混合模型的振动信号降噪方法

De-noising algorithm of vibration signals based on quantum Gaussian mixture model

由于机械设备振动信号受到背景噪声的干扰,造成机械设备故障状态特征不明显,因此提出了一种基于量子高斯混合模型的振动信号降噪方法。首先,对振动信号进行双树复小波包变换,对双树复小波包系数建立高斯混合模型,根据贝叶斯最大后验估计准则,得到双树复小波包系数收缩函数;然后,利用双树复小波包系数父代和子代的空间相关性,结合量子叠加态理论计算噪声信号和有用信号小波系数出现的概率值;最后,根据量子叠加态得到的概率参数值调节高斯混合模型中的小波包系数收缩函数,使小波包系数自适应非线性收缩,提高高斯混合模型的局部自适应性,实现机械振动信号的降噪处理。仿真信号和实测行星齿轮箱振动信号实验结果表明,该方法能够有效地去除振动信号中的噪声,凸显机械设备的故障状态特征。

Vibration signals of machinery equipment are often disturbed by background noise to cause machinery equipment’s fault features being not obvious. Here, a de-noising algorithm of vibration signals based on quantum Gaussian mixture model was proposed. Firstly, the dual-tree complex wavelet packet transform was performed on vibration signals, and Gaussian mixture model was established for dual-tree complex wavelet packet coefficients. According to Bayesian maximum posteriori estimation criterion, shrinkage function of dual-tree complex wavelet packet coefficients was acquired. Then the spatial correlation between dual-tree complex wavelet packet coefficients’ father generation and child one was used to combine the quantum superposition state theory, and calculate appearing probabilities of noise signal and useful one’ wavelet coefficients, respectively. Lastly, the shrinkage function of dual-tree complex wavelet packet coefficients was adjusted with probability parameters achieved with the quantum superposition state theory to make wavelet packet coefficients shrink adaptively and nonlinearly, and the local adaptability of Gaussian mixture model was improved to realize machinery vibration signals’ de-noising processing. The test results of simulated signals and measured planetary gearbox vibration signals indicated that this proposed method can be used to effectively get rid of noise in vibration signals and highlight fault state features of machinery equipment.