Multi-sensor gearbox fault diagnosis by using feature-fusion covariance matrix and multi-Riemannian kernel ridge regression

Abstract Intelligent fault diagnosis of gearbox holds important implications for the safety assessment and risk analysis of rotating machinery. Due to many monitoring variables in engineering practice, it is often necessary to install multiple sensors to monitor the operating conditions. To achieve multi-sensor information fusion and improve gearbox fault diagnosis performance, a fault diagnosis approach is proposed with feature-fusion covariance matrix (FFCM) and multi-Riemannian kernel ridge regression (MRKRR) in this paper. Firstly, FFCM is constructed to fuse multi-domain statistical features from multi-sensor data. FFCM not only has the characteristics of simple calculation and strong adaptability, but also can preserve the interaction relationship between different sensors. Secondly, by incorporating FFCM into the framework of the Riemannian manifold, a MRKRR model is proposed with the concept of multiple kernel learning (MKL), avoiding the selection of kernel function and its kernel parameter, and fully leveraging the Riemannian structure information of FFCM. Finally, the experiment results on two multi-sensor datasets verify that the proposed approach has excellent diagnosis performance consistently.