ased on the assumption that the ori

ased on the assumption that the original high-dimensional data lies on a low-dimensional manifold, SSKMFA aims toproject the high-dimensional data in ambient space into a low-dimensional feature space and simultaneously preserve theintrinsic manifold structure. Due to the fact that the different operating conditions of a mechanical system have variousmanifold structures, the fault classification of rolling bearings is multiple manifolds learning problem in principle [27,28].The faulty data in the same class resides on a sub-manifold and the defective samples in different classes are distributed ondifferent sub-manifolds. Thus, a new approach based on SSKMFA is proposed for fault diagnosis of rolling bearings, whichconsiders simultaneously the compactness of the same sub-manifold and the separability of different sub-manifolds.In order to achieve the optimal recognition result in category space, the proposed method finds multiple sub-manifoldsembedded in the raw high-dimensional pattern space and subsequently separates different sub-manifolds in feature space.The strategy of fault diagnosis based on SSKMFA is stated as follows: