A Machine Learning Technique to Predict Biaxial Failure Envelope of Unidirectional Composite Lamina

摘要

A machine learning technique was used to predict static, failure envelopes ofunidirectional composite laminas under combined normal (longitudinal or transverse)and shear loading at different biaxial ratios. An artificial neural network was chosen forthis purpose due to their superior computational efficiency and ability to handlenonlinear relationships between inputs and outputs. Training and test data for the neuralnetwork were taken from the experimental composite failure data for glass- and carbonfiberreinforced epoxies provided by the world-wide failure exercise (WWFE) program.A quadratic, stress interactive Tsai-Wu failure theory was calibrated based on thereported strength values, as well as optimized from the experimental failure data points.The prediction made by the neural network was compared against the Tsai-Wu failurecriterion predictions and it was observed that the trained neural network provides abetter representation of the experimental data.