The Noise Reduction of Structural Multichannel Signals Based on Independent Component Analysis

In practice, there are various noises mixed into structural vibration signals and the useful signals maybe covered up by noise. Independent component analysis (ICA) is one kind of effectual signal processing technology. Using of this method, the multichannel signals can be separated into some independent components. Because noise and vibration signal are statistical independent to each other, the ICA method is introduced to denoise the structural vibration signals. A noise channel is added to amplify the sensor signals to satisfy the calculation condition of ICA. Because the noise is unknown, the usual method simulating noises by numerical dummy channel needs to adjust the noise type unceasingly to obtain the perfect denoising effect. In this paper, an actual testing noise channel substitutes for the simulating channel. In the laboratory, a sensor, static relatively to the ground, is placed outside the vibration structure and regarded as noise channel. The signals of vibration experiment are processed by ICA. The effect of noise reduction compared with wavelet is satisfied and the denoised signals do not change its dynamic characteristic.