Artificial Neural Networks Applied in Mechanical Structural Design

Artificial Intelligence has brought many new problem-solving approaches to society in the last few years. In Artificial Intelligence domain, the Machine Learning techniques are being very successful mainly because of their ability to learn. The main objective of this article is the research, development and application of Artificial Intelligence techniques for mechanical structural design. Selected method is based on the Artificial Neural Networks, that allows the prediction of a certain variable based on a given set of data. The applied artificial network was developed in Python with TensorFlow library and the database was prepared using the
finite element software, Ansys, with parametric language for automatic data extraction. A classic solid mechanics case study, comprising a plate with a central hole subjected to uniaxial remote stress, is explored. It is intended to obtain the stress distribution for a plate with hole radius between 25 and 50 mm. With Artificial Neural Networks, a substantial reduction in the simulation time is observed, being, approximately, 79 times faster when compared to the solution time of the conventional finite element approach. The developed neural network has a relative average error of about 4.57%, which is considered satisfactory given that it is a
first application of these networks in this domain. In conclusion, with this work it is possible to highlight the potential advantages of Artificial Neural Networks in applied to stress/strain calculation in solid mechanics: shorter response time, less computational resources and problem simplification, in detriment of a lower resolution/accuracy of structural behaviour. Optimization procedures and digital-twin concepts can take advantage of these benefits, enabling near real-time calculations