现代电子技术

针对恶劣环境下振动信号复杂、噪声难以去除的问题，文中将改进的自适应噪声完备集合经验模态分解、模糊熵（FE）特征提取与改进的小波阈值相结合，提出一种ICEEMDAN⁃FE联合改进小波阈值的振动信号去噪算法。首先，将测得的振动信号经过ICEEMDAN分解为多个固有模态函数（IMF）与具有相对平滑的变化趋势的残余项（Res）；其次，通过模糊熵（FE）特征提取算法计算各IMF模糊熵特征值，通过设定的IMF阈值条件对信息主导部分的IMF进行保留；然后，采用改进小波阈值对仅保留的信息主导分量的各IMF进行相应去噪处理；最后，将残余项与改进小波阈值去噪处理后的IMF进行信号重构，得到最终信号。通过建立仿真信号对滤波效果进行评估，实验结果表明，与ICEEMDAN 去噪、小波阈值去噪以及ICEEMDAN⁃小波阈值去噪相比，所提算法信噪比（SNR）分别提高了3.233 5 dB、1.181 1 dB、1.066 3 dB，归一化互相关（NCC）分别提高了0.033 42、0.009 39、0.008 4，均方根误差（RMSE）分别降低了52.5%、23.81%、21.77%。导入实测振动信号后的去噪结果也表明，所提算法在进行去噪后有效信号更加完整，信号更为平滑，去噪效果较为理想。

In view of the complex vibration signals and difficult noise removal in harsh environments, the improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) and fuzzy entropy (FE) feature extraction are combined with the improved wavelet threshold, and a vibration signal denoising algorithm based on ICEEMDAN ⁃ FE and improved wavelet threshold is proposed. Firstly, the detected vibration signal is decomposed into multiple intrinsic mode functions (IMFs) and residuals with a relatively smooth trend by ICEEMDAN. Secondly, the FE feature extraction algorithm is used to calculate the FE eigenvalues of each IMF, and the IMF of the dominant part of the information is retained by the set IMF threshold conditions. Thirdly, the improved wavelet threshold is used to denoise the IMF of the retained dominant components of information. Finally, the residuals and the IMF after improved wavelet threshold denoising are subjected to signal reconstruction, so as to obtain the final signal. The filtering effect was evaluated by establishing simulation signals. The experimental results show that in comparison with ICEEMDAN denoising, wavelet threshold denoising and ICEEMDAN⁃wavelet threshold denoising, the signal⁃to⁃noise ratio (SNR) of the proposed algorithm is increased by 3.233 5 dB, 1.181 1 dB and 1.066 3 dB, respectively, its normalized cross⁃correlation (NCC) increased by 0.033 42, 0.009 39 and 0.008 4, respectively, and its root mean squared error (RMSE) decreased by 52.5%, 23.81% and 21.77%, respectively. After importing the measured vibration signal, the denoising results also show that the effective signal is more complete and smoother, and the denoising effect is more ideal.