Operational transfer path analysis based on deep neural network: Numerical validation

Abstract

In this study, a novel transfer path analysis formulation using an emerging deep neural network model is presented and numerically validated for a multi-structural system. In the proposed formulation, only the operational responses of structures are utilized to identify the contributions of all paths to vibration responses at a point of interest in the frequency domain. To this end, the model parameters of a dense neural network model are initially determined using a training dataset to predict the responses precisely. Next, the contribution of each path is identified from the responses predicted by using the trained network model with the input associated with the selected path eliminated. To establish the correct model, the original operational datasets are augmented using the phase shift and the cross-spectrum. The path contributions estimated using the proposed formulation with another numerically generated operational dataset are compared with the known path contributions. The comparison results show that the proposed method based on a deep neural network model can successfully predict the path contributions using only operational responses.