Independent Component Analysis in Frequency Domain and Its Application in Structural Vibration Signal Separation

In structural vibration signals, many kinds of signals (such as sinusoidal signals and impulsive signals) are mixed together. Convolutive mixtures often take place in vibration signals. In order to separate effectively the useful signals from the signals mixed by noise, the paper firstly transforms time signals into frequency signals through FFT, then applies FastICA to separate the useful signals from mixed signals, transforms frequency signals into time signals through IFFT, finally accomplishes the separating of the useful signals and noise. Two different experiments have been done to separate the steel structural vibration signals generated by hammer and vibration exciter in laboratory. The experiments results show that the mixed signals can be separated successfully when the exciters and sensors are placed on the suitable locations.