Transients Extraction Based on Averaged Random Orthogonal Matching Pursuit Algorithm for Machinery Fault Diagnosis

Abstract

Sparse representation based on matching pursuit (MP) algorithm is one of the effective methods for the extraction of weak feature contaminated by heavy noise. However, the optimal iterative threshold and iterations of MP algorithm are difficult to be determined and the sparsest representation is difficult to obtain for pursuit algorithms. In order to reduce the influence of the threshold and iterations on the performance of MP and obtain appropriate results without seeking the sparsest representation, a new transient feature extraction technique named averaged random orthogonal MP (AROMP) algorithm is proposed. In the proposed method, random orthogonal MP algorithm, which is a greedy algorithm to match the atoms with probability, is utilized repeatedly to generate a group of competitive representations for the mechanical vibration signal. Then by averaging these solutions, the estimated representation vector can be obtained to represent the vibration signal and then transients can be extracted from the noisy signal. The simulation study and experimental analysis show that transients can be extracted effectively from the noisy vibration signal. And comparison results between the proposed method and orthogonal MP show that the proposed algorithm is less dependent on iterations and can obtain a better performance for transients extraction. Comparisons between the proposed algorithm and spectral kurtosis also show the superiority of the proposed method.