Adaptive cancellation of background machine noise based on combination of ICA-R and RBFNN

Extraction of machine fault signals from background machine noises is of great help in improving the accuracy of machine fault diagnosis. In this paper, a prediction model of time series based on RBF neural network (RBFNN) is proposed to learn the priori knowledge of background machine noise that obscure in a template noise which is tailored from the historical samples of background machine noises. By defining the mean square error of prediction to candidate independent component with the proposed RBFNN model as the contrast function, a new ICA-R algorithm is proposed to extract the `pure´ background machine noise which is then taken as reference input of a Volterra Adaptive Noise Cancellation (VANC) system. The simulation shows that the combination of ICA-R and VANC system prevails over a standard VANC system. The new VANC system is easier to be implemented in engineering applications due to its sensor-position independent characteristics.