Enhanced Dictionary Design-based Sparse Classification Scheme Towards Machinery Intelligent Diagnostics

Abstract

Machinery intelligent diagnostics is taking on a key role in enabling smart operation and maintenance of modern industrial equipment, especially in the prospective era of industry 4.0. Sparse representation-assisted intelligent diagnostics (SR-ID) framework shows great prospects to obtain promising diagnostic performance without designing complex deep network architectures compared with deep learning models. However, the existing SR-ID approach still suffers to obtain superior and robust diagnostic accuracy in noisy circumstances. To tackle this challenge, a novel enhanced dictionary design-based sparse classification (EDD-SC) scheme is developed in this study, which comprises of enhanced dictionary design and intelligent health diagnostics. Firstly, the periodic similarity of vibration data is leveraged to fuse the physical priori information with dictionary design, thus enhancing reconstruction capability of EDD-SC. Secondly, a minimal sparse approximation error strategy is developed to accomplish superior health diagnosis. The presented EDD-SC scheme has been detailedly verified on the challenging task of planetary drivetrain fault diagnostics, showing that EDD-SC can yield robust and superior diagnostic results even in comparison to several state-of-the-art benchmarks. This work has provided a promising framework and paved a new direction towards robust data-driven machinery intelligent diagnostics.