A multiple faults diagnosis approach for roller bearings based on space reconstuction and nonlinear manifold was proposed according to the fact that vibration signal for roller bearings is non-stationary and time-variation.
After embedding the vibration signal into a high dimensional phase space, the original feature set was acquired by calculating the eigenvalues of the covariance matrix. Using the local tangent space alignment algorithm on the original feature set for feature compression, it will get the new features which were input into K-means classifier, and the output of the K-means classifier was the clustering results. The experimental results show that the proposed method has better clustering performance than the traditional linear principal component analysis method.