Multichannel Structural Inverse Filtering

Abstract

A technique has been implemented for predicting the multiple input force time histories acting on a machinery component through the use of multiple vibration measurements and a stored matrix of mechanical frequency response functions. A least squares pseudo-inverse of this transfer matrix is computed in the frequency domain using the singular value decomposition. This decomposition is used to condition the pseudo-inverse against potential singularities caused by uncertainties in the data. The constructed inverse filter is capable of separating multiple input force signals even when the vibration signals contain reverberated information from the inputs in overlapping frequency ranges. The technique has been successfully applied to a flat plate system with two inputs and up to four vibration measurements. Guidelines are suggested for reducing the effects of measurement inaccuracies on the force predictions.