Feature extraction for fault diagnosis utilizing supervised nonnegative matrix factorization combined statistical model
Yuanming Chen, Maolin Li, Lin Liang, Guanghua Xu, Huizhong Gao · 2016
A new method for separating the feature automatically with supervised nonnegative matrix factorization (NMF) is proposed for fault diagnosis. Because of the shortage of lacking prior knowledge in existed NMF, a supervised NMF combined statistical model for fault diagnosis is proposed. The basis matrix achieved in training stage is treated as the sources' features. Besides, Gaussian Mixture Model is introduced to estimate the distributions of the base vectors and then keep them as the prior knowledge. The rotation machinery with faults was used to evaluate the performance of the proposed method. The results show that the proposed method has a good source separation capability. Its performance is better than NMF.