Accuracy Prediction of Higher-Order Structural Theories for 1D Aerospace Structures Using Neural Networks

This study introduces a Machine Learning (ML) framework to assess the accuracy of higher-order structural theories for aerospace structures. Stiffened panels are modeled using one-dimensional elements and analyzed using refined theories based on the Carrera Unified Formulation (CUF). To evaluate accuracy, the framework compares the average error of the first ten natural frequencies with that of a reference model. An ML tool automatically predicts this error and identifies the most relevant generalized variables for each structural configuration. The dataset used to train the ML tool is generated using CUF and includes material properties, geometry, and the structural theory generalized variables. To process this information, two types of Neural Networks (NN) are employed: a Convolutional Neural Network (CNN) extracts geometric features from cross-section images, e.g., number and position of stiffeners, and a Deep Neural Network (DNN) combines these features with material and geometric data to predict the accuracy of each theory. As a result, the framework effectively predicts the most accurate and computationally efficient theories and their associated generalized variables.