Blind Source Deconvolution Research Articles

Segmentation MED method based on kurtosis-frequency curve and its application in bearing diagnosis

Signature extraction of fault impacts is a significant task for rolling bearing diagnosis. A series of blind source and target deconvolution methods respectively represented by minimum entropy deconvolution (MED) and maximum correlated kurtosis deconvolution (MCKD) are proposed to enhance the fault impulses from the complex interference components. However, blind source deconvolution methods are usually fragile for strong shock interference and target deconvolution methods require a rigorous high-accuracy prior knowledge. In addition, many methods based on correlation kurtosis indicator are no longer applicable under the variable speed condition. These problems form a huge obstacle to their application in fault diagnosis of rolling bearing. To overcome these limitations, a segmentation MED guided by the kurtosis-frequency curve is proposed in this paper. The presented method calculates the kurtosis-frequency curve by selecting appropriate filter parameters. Signals are adaptively segmented to several parts corresponding to the actual component according to the kurtosis-frequency curve. MED for each segmentation can effectively extract fault signatures and reduce the influence of other interference components. Compared with MED, the new method is more robust for strong interference impacts. Compared with the MCKD, this method does not need a high-accuracy input parameter and can woks at variable speed. Its effectiveness has been verified by several simulation signals and actual railway bearing datasets.

Blind Source Separation Based on ICA Algorithm Applied to Multispectral Fluorescence Imaging

Immunofluorescence is one of the most used techniques in optical fluorescence microscopy and has substantial applications in biology and pathology. It aims to detect and localize one or more proteins thanks to the use of specific dyes. One of the major issues with immunofluorescence is the intrinsic fluorescence, called auto-fluorescence present in some biological specimen. Several approaches, based on blind source deconvolution, are developed to deal with this problem and to isolate the extrinsic fluorescence from the intrinsic one. In this paper, we present two Independent Component Algorithms based on second order and higher order statistics. Experimental results have revealed that the blind identification algorithm based on second-order statistics is more adaptedly fit to fluorescence sources un-mixing problem than algorithms based on higher-order statistics.

Undercomplete Blind Subspace Deconvolution

We introduce the blind subspace deconvolution (BSSD) problem, which is the extension of both the blind source deconvolution (BSD) and the independent subspace analysis (ISA) tasks. We examine the c...

Natural Gradient Works Efficiently in Learning

When a parameter space has a certain underlying structure, the ordinary gradient of a function does not represent its steepest direction, but the natural gradient does. Information geometry is used for calculating the natural gradients in the parameter space of perceptrons, the space of matrices (for blind source separation), and the space of linear dynamical systems (for blind source deconvolution). The dynamical behavior of natural gradient online learning is analyzed and is proved to be Fisher efficient, implying that it has asymptotically the same performance as the optimal batch estimation of parameters. This suggests that the plateau phenomenon, which appears in the backpropagation learning algorithm of multilayer perceptrons, might disappear or might not be so serious when the natural gradient is used. An adaptive method of updating the learning rate is proposed and analyzed.