Artificial Neural Network Modeling to Predict Corrective Stress of a Two-layer Composite Plate under Fully Reversed Cyclic Loading Using the Finite Element Method & Morrow Method

Abstract

In this study, a single-hidden-layer feedforward Artificial Neural Network was developed to predict the maximum stress of a two-layer composite plate based on the Morrow correction method under fully reversed cyclic loading. A rectangular two-layer plate made of Epoxy Carbon Woven (230 GPa) was analyzed using the Finite Element Method for various fiber orientations of each layer (0°, 15°, 30°, 45°, 60°, 75°, and 90°) relative to the transverse axis. The results indicate that a 0° fiber orientation in both layers produced the maximum stress, potentially weakening the plate under tensile load, whereas a 90° orientation minimized stress and enhanced tensile strength. Increasing the second-layer angle while keeping the first layer fixed reduced stress, whereas decreasing the first-layer angle for a fixed second-layer angle increases stress. Maximum stress regions shifted from localized points to linear distributions, sometimes moving from uniform edge distributions to mid-edge concentrations. When both layers had identical angles, the stress magnitude increased, and the maximum stress shifted from the loaded edge to a corner, indicating stress concentration and a potential reduction in lifespan. The Artificial Neural Network demonstrated excellent predictive performance, achieving an optimal validation Mean Squared Error of 3.2266 × 10⁻⁴ at iteration 22, with a minimum overall Mean Squared Error of 6.5566 × 10⁻⁴ across all datasets. The correlation coefficients for the training, validation, test, and entire datasets were 0.99508, 0.98649, 0.99484, and 0.99485, respectively, indicating a strong agreement between the Mean Squared Error predictions and Finite Element Method-simulated values.