Review article
https://doi.org/10.21857/ygjwrce11y
On artificial neural networks application in solid mechanics as an alternative to conventional finite element modelling
Jurica Sorić
orcid.org/0009-0005-5960-4831
; University of Zagreb, Faculty of Mechanical Engineering and Naval Architecture
Abstract
Due to the complexity of engineering problems and the advances in computer performance, a novel computational strategy employing artificial neural networks has recently arisen as an alternative to numerical modelling by conventional finite element application. Neural networks are the core technology in the framework of machine learning, which is a subfield of artificial intelligence, and have been adopted for solving computational mechanics problems, especially in the field of solid mechanics. In the present paper, a short review of the neural networks is given, while the feedforward neural network and the physics-informed neural network are presented and discussed in more detail. In the framework of the physics-informed neural network formulations, both the governing partial differential equations and the energy functional are employed in the loss functions. The feedforward neural network approach is tested by linear elastic analysis, while the efficiency of the physics-informed neural network is demonstrated by modelling of elastoplastic structural responses and two-dimensional crack propagation using phase-field theory. All results are compared by the finite element solutions. It is shown that the neural network algorithms reproduce the finite element results correctly, and that they have an advantage in computational efficiency
Keywords
artificial neural networks; feedforward neural network; physics-informed neural network; linear elastic analysis; elastoplastic analysis; crack propagation
Hrčak ID:
343850
URI
Publication date:
27.1.2026.
Visits: 367 *