An intelligent index-driven multiwavelet feature extraction method for mechanical fault diagnosis

Abstract

To sufficiently perform the inner product match principle of mechanical fault diagnosis, the construction and selection of the basic functions are the critical issues. Hereinto, the construction of basic functions increases the ability of accurate inner product match, but the selection determines the final matching accuracy for fault diagnosis. Thus, an intelligent index-driven multiwavelet feature extraction method is proposed for mechanical fault diagnosis, which is essentially an accurate inner product matching established by the ‘appropriate’ multiwavelet basic functions guiding by an intelligent data-driven index. First, an intelligent index by the improved weighted square envelope spectrum is designed for the data-adapted selection guidance. Hereinto, the optimal weights of purified envelope spectrum by multiwavelet neighboring coefficient denoising are searched by support vector machine. Second, a basic function library is established by two excellent families of SA4-type and Hermite lifting-based multiwavelets. Third, guiding by the maximization of the intelligent index, the optimal basic functions are selected from the candidate multiwavelet library and employed to extract the fault features. Finally, two experimental case studies of bearing degradation data verify the effectiveness and feasibility of this method with comparisons. The results show that it could extract the fault features at different stages, especially at the very initial stages, offering a useful tool for mechanical fault diagnosis.