Determination of the Mode Dominance for Model Reduction

Abstract

When time domain state-space identification algorithms are employed, identified models may need to be overparameterized due to system nonlinearity or measurement errors. After an initial oversized model is obtained, model reduction must be conducted. An important step in model reduction is to determine those significant modes Several commonly used model reduction methods such as the balanced realization (BR) technique select significant modes by minimizing the impulsive response error between the original model and reduced model. This study proposes an alternative method. The method is based on modal approximation. Two indices are devised to determine significant and insignificant modes. The parameter matrices expressed in the modal realization are reduced by eliminating the terms corresponding to the insignificant modes. Experimental modeling of a flexible manipulator system is used to demonstrate the effectiveness of the proposed method. The results using the BR technique are also given as a comparison.