Predicting the Response of Laminated Composite Beams: A Comparison of Machine Learning Algorithms

Abstract

A comparative study of machine learning regression algorithms for predicting the deflection of laminated composite beams is presented herein. The problem of the scarcity of experimental data is solved by ample numerically prepared data, which are necessary for the training, validation, and testing of the algorithms. To this end, the pertinent geometric and material properties of the beam are discretized appropriately, and a refined higher-order beam theory is employed for the accurate evaluation of the deflection in each case. The results indicate that the Extra-Trees algorithm performs best, demonstrating excellent predictive capabilities.