Algorithm 2: Signal Separation Method Based on Multi-constraint NMF

Step 1. The algorithm of the short-time Fourier transform (STFT) is performed to obtain a feature matrix with local information.

Step 2. Take the square value of the feature matrix, and the multi-constraint NMF algorithm is used to reduce the dimension, and obtain the base matrix W and the coefficient matrix H.

Step 3. The matrix W and H are recombined in subspace, and the recombined signals with feature components in the time domain are obtained by the inverse short-time Fourier transform (ISTFT).

Step 4. Calculate the WK values of the recombined signals

Step 5. The separation signals with high WK values are selected for envelope spectrum analysis to extract the fault features of bearings.