ANN-Based Prediction of Resonant Frequencies in Thin Isotropic Plates with CCCS Boundary Conditions

Abstract

 this paper presents the prediction of the non-dimensional resonant frequencies of three edges rotationally restrained and one edge hinged (CCCS) thin rectangular isotropic plates subjected to dynamic loading using an artificial neural network (ANN). The ANN model was developed, trained, and tested using 165 numerical datasets generated via the Ritz energy method and polynomial shape functions. The polynomial shape function was employed to define the deflection functions for fifteen plate configurations, while the Ritz energy method determined the total potential energy under dynamic loading. Aspect ratios ranging from 1 to 2, with an interval of 0.1, were considered. The ANN model utilized a three-layer feed forward architecture with a back propagation algorithm, featuring two input parameters (aspect ratio and support conditions), ten hidden nodes, and three output parameters (support conditions, aspect ratio, and frequency). The predicted non-dimensional frequencies were compared with those obtained from the Ritz energy method. A high correlation coefficient (R = 0.99973) between the ANN-predicted and Ritz-derived frequencies indicates near-perfect agreement. Additionally, the error histogram shows that most residual errors fall between -0.5 and 0.6, with the majority below 0.001, suggesting minimal and insignificant outliers. These results confirm that the ANN model effectively predicts the non-dimensional resonant frequencies for varying aspect ratios and boundary conditions.