Fault Diagnosis of time-frequency images based on non-negative factorization and neural network ensemble
Qinghua Wang, Youyun Zhang, Zhu Yongshen, Junyan Yang · 2008
Considering unstable characteristics of vibration signals with mechanical failure, the Wigner-Ville distributions (WVD) of vibration acceleration signals, which were acquired from the cylinder head in eight different states of valve train, were calculated and displayed in grey images. Non-negative matrix factorization (NMF) as a useful decomposition for multivariate data and neural network ensembles (NNE) with better generalization capability for classification than a single NN were introduced to perform intelligent diagnosis without further fault feature (such as eigenvalues or symptom parameters) extraction from time-frequency distributions. The experimental results show that the time-frequency images can be classified accurately by the proposed methods.