Humanoid extraction of abnormal engine sounds by using ICA-R and VANC
Li Zhang, Yaowu Shi, Luquan Ren · 2012
The accuracy of engine noise diagnosis depends greatly on SNR (Signal-to-Noise Ratio) of fault-featured engine sounds. By simulating the way that human technicians use to distinguish abnormal engine sound from observed engine acoustics, a humanoid ANC (Adaptive Noise Cancellation) system is proposed. With a RBFNN (RBF Neural Network) based measurement that is defined to evaluate the closeness between asynchronously sampled time series, a new ICA-R (Independent Component Analysis with Reference) algorithm and a Volterra ANC system are designed. The proposed method simulates the way that human technicians use to discern and then counteract the abnormal engine sounds according to the healthy engine acoustics that stored in their memories. The simulations prove that the proposed humanoid system is functional. Compared with standard VANC system, the humanoid VANC system is more effective in noise cancellation performance, and is little affected by sensor locations. The proposed method that used for extraction of interested signals from engine acoustics is fit for being extended to other applications that the priori knowledge of background noise is fully contained in its historical samples.