Adaptive weighted data fusion-driven multi-layer discriminative dictionary learning method for intelligent fault diagnosis of rotating machinery

Abstract

Dictionary learning is an effective intelligent diagnosis model to achieve fault classification of rotating machinery. However, existing dictionary learning methods mostly employ a single-layer architecture for dictionary learning, which hinders the learning of deeper discriminative features. Besides, due to limited feature information and noise interference, dictionary learning models learned from single-domain data cannot guarantee that the vibration signals achieve a satisfactory sparse representation. In this paper, an adaptive weighted data fusion-driven multi-layer discriminative dictionary learning method (AWDF-MLDDL) is proposed for intelligent fault diagnosis of rotating machinery. First, a multi-layer discriminative dictionary learning framework is proposed to learn discriminative dictionaries with a deep architecture, in which a structured incoherence term is employed to improve the independence of sub-dictionaries associated with different categories, while still enabling feature sharing between categories. Second, an adaptive weighted fusion method applied to dictionary learning is proposed to improve the representation capability of the category-associated discriminative sub-dictionaries. Finally, a sparse recognition method with an adjustable decision fusion strategy is designed to realize jointly intelligent fault diagnosis. Two datasets are used to quantitatively verify the superiority and effectiveness of the developed AWDF-MLDDL under small-sample and noise scenarios, indicating that the AWDF-MLDDL has superiority for intelligent fault diagnosis of rotating machinery.