Application of blind source separation in industrial noise prediction and control

Blind Source Separation (BSS) has been widely used for speech separation, imaging processing, and other applications, but not in the field of industrial noise separation. In this study, the algorithms of BSS are applied to the separation of industrial noises. Two BSS algorithms, namely: Independent Component Analysis (ICA) and Degenerate Un-Mixing Estimation Technique (DUET) are used. Both modeling and experimental sound signals are tested, and the accuracy of the prediction is evaluated. When the modeled mixtures are separated, results show that both the ICA and DUET algorithm can separate the sound sources successfully. However, as compared to ICA, the DUET method can detect the number of sound sources and predict the relative delay between the sound sources. When the measured sound mixtures are used for the separation, the accuracy of prediction is reduced as the sound contains both the amplitude attenuation and phase delay but also the reverberation.