Machine fault diagnosis based on Frequency-Domain Blind Deconvolution

On the basis of introducing the model of Frequency-Domain Blind Deconvolution (FDBD), key techniques in mechanical signal extraction were comprehensively related and analyzed in this paper, which include the methods of suppressing the difference between circular and partial convolution by coordinating the relationship between FFT size and length of each frequency bin or Modified Discrete Fourier Transform, the methods of removing the permutation indeterminacy (methods based on consistency of filter coefficients, DOA methods, Split Spectrum methods, methods based on Nonlinear Function, etc.), the applications of Complex-Domain Blind Separation Algorithms which based on explicit tensor eigenvalue decomposition and nonlinear functions. Aimed at vibration and acoustic signal feature extraction in complex environment and equipment with complex mechanical structures, the application values of FDBD and its research status in machinery condition monitoring and fault diagnosis were reviewed and summarized. Finally, the main problems which need to be studied further in this area were pointed out.