Studying And Appling Of Sparse De-noising Algorithm For Vibration Signal

Signal de-noising is a significant step in the signal processing. For example, it is used in health check of large structures. Changing environment or the imperfect of the sampling device will causes some noise in the sampling signal. Those noise may submerged useful vibration signal and damage the features of vibration signal. If researchers process those signal, directly. It will affect the accuracy of health check analysis result. The error of analysis may lead to serious economic loss and even harm the social public security. So signal de-noising is a significant pretreatment step of the signal processing.Sparse de-noise is a current emerging de-noising method. The main theory of sparse de-noise is signal sparse representation. Signal can sparse expansion on an appropriate dictionary. But noise in the signal can’t sparse mapping on the dictionary and the mapping coefficients are small and not sparse. Then, it use algorithms for sparse decomposition which can remove those small coefficients of noise. And get effect of de-noising. Redundant dictionary can or not accurately retains characteristics of signal will be de-noised is the key of sparse de-noise. Given the vibration signal is conform to the AR model. In this paper, One improved method of redundant dictionary in the sparse de-noise is proposed through the combination of AR model and sparse de-noise. This method increases the effect of sparse de-noise. First, according to the signal features and mathematical expression of AR, an over-completed sparse base and some redundant dictionaries can be build. The second, we use one sparse decomposition algorithm to solve those sparse coefficients. The last, according to the over-completed sparse base and sparse mapping coefficients, the signal which is de-noised and reconstruct is obtained.The K-SVD algorithm of dictionary training commonly used in sparse de-noise of image. It can increase the compatibility of the dictionary be updated and the signal, thus increase the effect of sparse de-noising. The paper uses a signal matrix which is constituted by multiple acquisition signal sequence to training one redundant dictionary by K-SVD algorithm. Then we use the trained dictionary to sparse de-noise for the vibration signal.Above all, the main research of this topic is automatically improving of sparse domain in the sparse de-noise. It increases the adaptive of dictionary and makes sparse de-noise method to be suitable for many signalizes with different characteristics. The experimental results show that the result of sparse de-noising with adaptive redundant dictionary is excellent. And the adaptive redundant dictionary can accurately keep the signal’s feature, so the refactoring signal can quite close to the original signal.