Machine Learning Predictions for the Comparative Mechanical Analysis of Composite Laminates with Various Fibers

This article addresses the complex behavior of composite laminates under varied layer orientations during tensile tests, focusing on carbon fiber and epoxy matrix composites. Data characterizing the mechanical load behavior are obtained using twelve composite laminates with different layer orientations and the DIGIMAT-VA software (version 2023.3). First, these data were used to elaborate a complex comparative analysis of composite laminates from the perspective of materials science. Composite laminates belong to three classes: unidirectional, off-axis oriented, and symmetrically balanced laminates, each having a specific behavior. From the perspective of designing a new material, a prediction model that is faster than the finite element analysis is needed to apply this comparative analysis’s conclusions. As a novelty, this paper introduces several machine learning prediction models for composite laminates with 16 layers arranged in different orientations. The Regression Neural Network model performs best, effectively replacing expensive tensile test simulations and ensuring good statistics (RMSE = 34.385, R2 = 1, MAE = 19.829). The simulation time decreases from 34.5 s (in the case of finite element) to 0.6 s. The prediction model returns the stress–strain characteristic of the elastic zone given the new layer orientations. These models were implemented in the MATLAB system 2024, and their running proved good models’ generalization power and accuracy. Even specimens with randomly oriented layers were successfully tested.