Maximum margin Riemannian manifold-based hyperdisk for fault diagnosis of roller bearing with multi-channel fusion covariance matrix

For rotating machinery, the sudden failure of roller bearing would lead to the downtime of the whole system and even catastrophic accidents. Therefore, multiple accelerometers are usually arranged to comprehensively evaluate the health of roller bearing, enhancing the stability and reliability of monitoring results. This paper proposes a novel fault diagnosis framework by utilizing a multi-channel fusion covariance matrix (MFCM) and Riemannian manifold-based hyperdisk. First, 22 statistical features are acquired from each channel data. Then, MFCM is calculated as the fault feature representation of roller bearing to achieve multi-channel feature fusion, where the element of MFCM represents the correlation information between different channels. Finally, since MFCM is a symmetric positive definite (SPD) matrix, lying on a Riemannian manifold, we design a maximum margin Riemannian manifold-based hyperdisk (MMRMHD) classifier to conduct fault classification, where Log-Euclidean metric (LEM) is introduced to calibrate the distribution of MFCMs. Moreover, to further improve the classification ability of nonlinear SPD data, we map MFCMs into a high-dimensional Hilbert space with the LEM-based kernel function and construct a novel kernelized MMRMHD model. The experimental results on two bearing datasets with multi-channel vibration signals demonstrate the effectiveness and superiority of the proposed fault diagnosis framework.