Bearing fault diagnosis of hard and soft threshold denoising

Commonly used bearing fault diagnosis method for vibration analysis, by analyzing the vibration acceleration sensors to collect data, using traditional signal processing methods analyzing the characteristic of fault fault type. Environmental noise is bigger, often bearing the actual working condition of the signal in the analysis, the influence of the final signal analysis and judgment, therefore the signal denoising has become an important aspect in the process of bearing fault diagnosis, the common method to signal denoising are mainly based on transform and independent component analysis, empirical mode decomposition and principal component analysis and sparse decomposition methods, this paper mainly introduces decomposition noise reduction method. In mathematics, the problem of denoising is the nature of a function approximation problem, namely how to by the generating function scale and translation version of the generative function space, according to the proposed measure criterion, finding the best approximation of the original signal, to complete the original signal and noise signal. Is also from the actual signal space to the optimal mapping function space, in order to get the best to restore the original signal. From the perspective of the signal of denoising is a signal filtering problem, to a large degree and although noise can be regarded as a low pass filter, but because after denoising can successfully retain the signal characteristics, so at this point is better than traditional low-pass filter. After decomposition, it is generally believed coefficient larger represent the main characteristics of the signal, and the coefficient of small part for more noise, so the threshold for implementing about reduction coefficient of small signal denoising process. Threshold denoising mainly adopts hard threshold and soft threshold method. Hard threshold popular speak for less than the threshold decomposition coefficient is set to 0, is greater than the threshold coefficient is constant, and the soft threshold value is less than the coefficient of threshold set to 0, part is greater than the threshold value minus the threshold. Look from the principle, the key lies in the choice of the threshold denoising, reasonable threshold value can get better denoising result. At present commonly used threshold for: lambda = = sigma * SQRT ( 2log( N) ) , the sigma is noise intensity, N for the length of the signal. Hard and soft threshold value exist some shortcomings, thus put forward a lot of improved algorithm, such as soft hard threshold denoising of a kind of improved eclectic method, the improved threshold function is between soft and hard thresholding function of a flexible choice. Formula is really is a good method, at present there are open source code, is a good choice.