Generalized Gini indices–enhanced sparsity decomposition for rotating machinery health assessment using a rotary encoder

Abstract

ABSTRACT Rotary encoder signal has been attracting considerable attention for health estimation of the rotating elements, since its advantages in lower testing costs and sensitive to torsional stiffness loss caused by incipient faults. However, the practical encoder signal often keeps low signal-to-noise ratio due to its instantaneous angular acceleration conventional method, which brings new challenges for the commonly used feature extraction methods. Sparse decomposition is an efficient method for fault impact features extracting of rotating machinery. However, limited by the quality of initial dictionary and the learning efficiency of K-SVD algorithm, commonly used sparse decomposition is susceptible to the signal noise, especially when unsuitable initial atoms are applied. To overcome this problem, a simple but effective framework term Generalized Gini indices-enhanced sparsity decomposition is established in this paper for practical encoder signal analysis. Firstly, the de-trended and frequency domain weighting methods are used for instantaneous angular acceleration estimation. After that, a Generalized Gini indices guided tunable Q-factor wavelet filter is introduced for the atom selection. Then, the shift-invariant K-SVD method is introduced for dictionary updating and sparse coefficients calculation. Through the analysis of simulation and experiment, the transient characteristics of the gear can be extracted well which provides an alternative methodology for incipient fault detection of rotating machinery.