Improvement of Experimental SEA model accuracy using Independent Component Analysis

Statistical Energy Analysis (SEA) is a suitable tool to predict vibration stationary responses. It is roughly categorized into analytical, FEM based, and experimental SEA due to the process of model construction. SEA is expected to be used for designing machines with optimal vibration transfer paths. However experimental SEA model accuracy especially in low frequency is highly dependent on experimental condition and target structure. It is required to develop model construction process for more convenient use of SEA. In this paper, preprocessing algorithm using independent component analysis (ICA) is proposed for improvement of model construction of SEA. Here, ICA is a signal processing method which is originally developed for bio-signal analysis and it is used for separation of complex mixture of several vibration sources. Feasibility of the proposed method is examined through an experiment with a test structure, composed of three flat steel plates.