Source separation based on nonnegative matrix factorization and independent component correlation algorithm

Because the initialization of Nonnegative Matrix Factorization (NMF) has a great impact to the final result, a new method that combines Independent Component Analysis (ICA) with NMF is put forward. Firstly, ICA is used to process the raw data and the initial value of NMF can be obtatined. Secondly, according to the characters of machine faults, the Hierarchical Alternating Least Squares (HALS-CR) is adopted to extract fault feature from complex system. Finally, simulations and application data show that the proposed approach is effective in improving the separation of traditional NMF.