Manifold learning-based subspace distance for machinery damage assessment

• A damage assessment scheme is proposed by using manifold subspace distance.
• Manifold subspace analysis is extended into the domain of machinery damage assessment.
• An effective subspace distance is defined for damage assessment.
• The damage index reveals intrinsic pattern variation in the mechanical state.
• The proposed method can outperform some other damage assessment methods.

Damage assessment is very meaningful to keep safety and reliability of machinery components, and vibration analysis is an effective way to carry out the damage assessment. In this paper, a damage index is designed by performing manifold distance analysis on vibration signal. To calculate the index, vibration signal is collected firstly, and feature extraction is carried out to obtain statistical features that can capture signal characteristics comprehensively. Then, manifold learning algorithm is utilized to decompose feature matrix to be a subspace, that is, manifold subspace. The manifold learning algorithm seeks to keep local relationship of the feature matrix, which is more meaningful for damage assessment. Finally, Grassmann distance between manifold subspaces is defined as a damage index. The Grassmann distance reflecting manifold structure is a suitable metric to measure distance between subspaces in the manifold. The defined damage index is applied to damage assessment of a rotor and the bearing, and the result validates its effectiveness for damage assessment of machinery component.