A mechanical fault characteristic extraction method based on orthogonal non-negative matrix factorization

A new characteristic extracting approach based on orthogonality non-negative matrix factorization with principle component analysis solving the initialization problem (PCA-ONMF) is presented. The research for this algorithm is added the orthogonality constraints to increase the orthogonality, locality and independence of the basic vectors, and make the factorization results easy to explain. On the other hand, the initialization of principle component analysis improves the robustness of the algorithm. Simulation experiments show that the characteristic basis vectors based on PCA-ONMF have smaller reconstruction error. And the gear box fault recognition experiment results with higher recognition accuracy confirm the effectiveness of the method.