Fast analysis and prediction approach for geometrically nonlinear bending analysis of plates and shells using artificial neural networks

Abstract

For the first time, this paper introduces a Fast Analysis and Prediction (FAP) approach for geometrically nonlinear bending analysis of plates and shells. The proposed approach consists of two phases: fast analysis and prediction using artificial neural networks (ANNs) combined with the Bayesian regularization algorithm. In the first phase (fast analysis), geometrically nonlinear bending analysis of a structure is performed using only several load steps. This enables the efficient computation within a short period of time. In the second phase (prediction using ANNs), the deflections obtained in the first phase are utilized as the data for training the ANNs. The trained networks are then used to quickly predict deflections at any load step. The calculation process in the second phase is also very fast. For the analysis package in the first phase, the formulation of this paper is originated from the literature for geometrically nonlinear analysis of composite shells. High accuracy and efficiency of the FAP approach were demonstrated via solving nonlinear bending problems of plates and shells including an isotropic square plate, a composite cylinder subjected to internal pressure, a FG-CNTRC panel and a pinched cylinder with end diaphragms. The proposed approach offers significantly computational savings (up to 70%) compared to the conventional one. ANNs combined with the Bayesian regularization algorithm, which are proposed for the FAP approach in this work, possess the outstanding advantages compared to many existing neural networks. The FAP approach can be applied to simulations, observations, or experiments where nonlinear responses are present or data is limited. Applying the FAP approach to such areas can significantly save labor, time, and money.