基于信号稀疏性的静电层析成像降噪方法研究

Noise Reduction of Electrostatic Tomography Based on Signal Sparsity

关键词:  滑油检测  荷电磨粒  静电层析成像  稀疏分解  降噪

Key words:  Lubricating oil monitoring  Charged debris  Electrostatic tomography  Sparse decomposition  Noise reduction

摘要:

静电层析成像（Electrostatic tomography，EST）技术因其无辐射、非入侵、可视化、实时性高、成本低等优势在滑油磨粒在线监测中获得广泛研究，但实际测量中静电信号幅值微弱且含有大量噪声，严重影响图像重建质量。针对上述问题，本文提出一种基于稀疏分解的静电信号降噪方法。首先，对EST传感器测量得到的静电信号数据构建相应的字典，然后用正交匹配追踪（OMP）算法在字典中寻找信号的稀疏表示矩阵，并用其与字典的乘积来表示信号，最后将稀疏表示后的信号代入基于原始对偶内点法（PDIPA）的EST图像重建，并与两种经典降噪方法进行对比。实验结果表明：经过数据降噪处理，重建图像误差相对于降噪前下降5.5%，相对于小波分析或经验模态分解（Empirical Mode Decomposition，EMD）方法具有较高的准确性；采用本文提出的降噪方法，明显提高了管道内单个荷电颗粒和两个荷电颗粒在不同径向位置时的成像质量。

Abstract:

Electrostatic tomography （EST） has been widely studied in the area of online oil debris monitoring due to its advantages of non-radiation， non-invasion， visualization， high real-time performance and low cost. While in practical measurements， the electrostatic signals are both weak and noisy， which has seriously affected the quality of image reconstruction. In order to solve the above problems， a sparse decomposition-based denoising method for electrostatic signal is proposed. First， a dictionary is constructed for the electrostatic signal measured with EST sensor. Then the sparse representation matrix of the signal is picked up from the dictionary by adopting the orthogonal matching pursuit（OMP） algorithm， and the signal can be represented by the product of the matrix and the dictionary. Finally， the sparse signal is substituted into EST image reconstruction based on the primal dual interior point method （PDIPA）， and the results are compared by two classical denoising methods. Experimental results demonstrated that the error of the reconstructed image is reduced by about 5.5% compared to that before noise reduction， and the proposed method obtained higher accuracy than the wavelet analysis or EMD method. By utilizing the proposed denoising method， the imaging quality for single charged particle and two charged particles located in different radial positions can be obviously improved.