基于改进EMD与CMSE的风机叶片音频信号去噪方法  

Denoising method of audio signal of fan blade based on improved EMD and CMSE

作　　者：彭媛宁 姚恩涛[1] 石玉[1] 周克印[2] Peng Yuanning;Yao Entao;Shi Yu;Zhou Keyin(College of Automation Engineering, Nanjing University of Aeronautics ＆ Astronautics, Nanjing 210016 ,China;College of Materials Science and Technology,Nanjing University of Aeronautics ＆ Astronautics, Nanjing 210016 ,China)

机构地区：[1]南京航空航天大学自动化学院,南京210016 [2]南京航空航天大学材料科学与技术学院,南京210016

出　　处：《电子测量技术》2018年第2期93-100,共8页Electronic Measurement Technology

摘　　要：风机叶片旋转产生气动噪声,其特征参数与叶片的表面裂纹大小分布、表面平整度等损伤存在必然的关系。但是实测的风机叶片气动噪声不可避免地受到周围环境噪声的干扰,如何在强背景噪声下提取气动噪声,是利用音频信号实现监测风机叶片损伤的前提。由于环境噪声相对于气动噪声属于低频噪声且二者频带部分重合,针对待测信号的这一特性提出一种基于改进EMD和CMSE的风机叶片音频信号去噪方法,先将信号经验模态分解（empirical mode decomposition,EMD）为若干个表征不同频率的本征模态函数（intrinsic mode function,IMF）,然后利用改进的连续均方误差（consecutive mean square error,CMSE）准则筛选出有用信号占主导的本征模态函数,重构得到去噪后的信号。在信噪比极小的情况下,CMSE准则失效,需要在EMD分解时对IMF分量辅以循环随机排列-重构-累加-平均算法,提高信号的信噪比。为了验证所提方法的有效性,分别以一种传统的EMD去噪方法（EMD＋相对熵去噪方法）和所提出的改进EMD和CMSE的去噪方法对实测的风机叶片音频信号进行去噪。实验结果表明,该方法可以有效地去除噪声,优于传统的去噪方法,不受主观参数的影响。创新性地将循环随机排列-重构-累加-平均算法应用到所提出的去噪方法中,保证了该方法可以广泛适用于不同信噪比信号的去噪中,具有自适应的优点。There is a certain relationship between the characteristic parameters of the aerodynamic noise generated by the rotation of the fan blades and the crack size,distribution,flatness and other damage on the blade surface. In fact,the aerodynamic noise generated by the fan blades will inevitably receive interference from ambient noise. How to extract aerodynamic noise from strong background noise is a prerequisite for detecting blade damage by using audio signals. The ambient noise is a low frequency noise compared to aerodynamic noise,and the frequency bands of the two are partially coincident. The paper presents a method for denoising audio signal of fan blades based on improved EMD and CMSE. First,the signal is decomposed into a number of different frequency characteristics of the intrinsic mode function by the EMD operation. And then by using the continuous mean square error criterion,the IMF dominated by the useful signal is selected.Finally,these IMF components can be reconstructed. However,when the signal-to-noise ratio is small,the continuous mean square error criterion is disabled. It is necessary to cyclically carry out random permutation-reconstruction-accumulationaverage operation to improve the signal to noise ratio of the signal in the EMD. In order to verify the effectiveness of the proposed method,the noise of the measured fan blade audio signal is denoised by a traditional EMD denoising method（ EMD＋ Relative Entropydenoising method） and the improved EMD and CMSE denoising methods proposed in this paper. The experimental results show that the method proposed in this paper can effectively remove the noise,which is superior to the traditional denoising method and is not influenced by subjective parameters. The cyclic random permutation-reconstructionaccumulating-average operation is applied to the proposed denoising method creatively,which ensures that the method can be widely applied to the denoising of signals with different signal to noise ratios,and has the advantages of adaptive.

关 键 词：风机叶片 经验模态分解 连续均方误差 去噪 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]