Periodic sparsity oriented super-wavelet analysis with application to motor bearing fault detection of wind turbine

Abstract

The demand of high similarity between the matching basis function and expected feature is required by the inner product transform principle. However, without precise information of the potential fault features, the deterministic or statistical characteristics of the investigated features are beneficial to the selection and construction of proper matching bases. According to the intrinsic periodic sparsity phenomena of repetitive impulsive fault features, a periodic sparsity based oriented super-wavelet transform is proposed. The super-wavelet transform is constructed based on the tunable Q-factor wavelet transform (TQWT) and presented as an improvement to the conventional idea of unique and fixed basis. Within the procedure, the super-wavelet dictionary functions are applied to decompose signals; an indicator estimating the periodic sparsity feature energy ratio (PSFER) is adopted to guide the selection of TQWT's parameters; the selected optimal super-wavelet basis is utilized to reveal the hidden fault features in the signal. The proposed technique is applied to acquire the incipient fault features of a motor bearing on a piece of wind power generation equipment, and the extracted features proved to be associated with an actual bearing fault.