Machine learning regression approaches for predicting the ultimate buckling load of variable-stiffness composite cylinders

• Published in: Acta mechanica Vol. 232; no. 3; pp. 921 - 931
• Main Authors: Kaveh, A., Dadras Eslamlou, A., Javadi, S. M., Geran Malek, N.
• Format: Journal Article
• Language: English
• Published: Vienna Springer Vienna 01.03.2021; Springer; Springer Nature B.V
• Subjects: Algorithms; Analysis; Aspect ratio; Buckling; Classical and Continuum Physics; Control; Cost analysis; Cylinders; Decision trees; Dynamical Systems; Engineering; Engineering Fluid Dynamics; Engineering Thermodynamics; Finite element method; Heat and Mass Transfer; Machine learning; Mathematical analysis; Metamodels; Methods; Nonlinearity; Optimization; Original Paper; Regression analysis; Regression models; Solid Mechanics; Steering; Stiffness; Theoretical and Applied Mechanics; Vibration
• ISSN: 0001-5970; 1619-6937
• DOI: 10.1007/s00707-020-02878-2

Due to the high computational cost of finite element analyses, particularly in optimization tasks, establishing a high-fidelity surrogate model is of immense importance to engineers. Analyses of fiber steering composite cylinders are costly and often involve high nonlinearity. Among different metamodeling approaches, machine learning techniques provide effective models that can capture high nonlinearity. Therefore, in this study, different machine learning techniques are employed to establish a relationship between the fiber angle and buckling capacity of the cylinders under bending-induced loads. The utilized data set contains a total of 11,000 cases, including seven attributes (i.e., fiber angles) for 11 different aspect ratios ( L / R ). The numerical results demonstrate that compared with the Random Forest Regressor, Decision Tree algorithm, and Multiple Linear Regression, the Deep Learning model obtains stable results with smaller errors and higher generalization.