Natural Mode Prediction of a Cantilever Beam Using a Physics-Informed Neural Network

In this study, a physics-informed neural network model is developed to predict the natural modes of the entire structure with only a few frequency response functions, and its effectiveness and practical applicability is subsequently examined. The network model is used to propose a method to obtain the associated natural mode after determining the natural frequencies from frequency response functions. The frequency response functions are acquired from two randomly-selected measurement points on the cantilever, and 12 collocation points are uniformly distributed to predict the 1st, 2nd, and 3rd natural modes. The developed artificial neural network model consists of three hidden layers with 20 nodes used in each. The proposed method successfully predicts the natural mode. The accuracy of the predicted natural mode depending on the number and distribution of measurement and collocation points was also investigated. Based on the results, a discussion is presented regarding how this method can be utilized in a practical experimental modal test.