Fast simulation of bolted joint using machine learning.

What is it about?

We often encounter large composite structures in fields such as aerospace and offshore wind energy turbines. These structures require effective joining techniques, with bolts being one of the preferred methods. However, creating high-fidelity Finite Element (FE) models of composite bolted joints is computationally expensive, making it difficult to simulate large structures with numerous bolts. In such cases, a novel method that enables faster simulation of bolted joints significantly improving efficiency compared to traditional high-fidelity methods- presents a promising research avenue. This approach makes it feasible to study large structures with many bolts in the future. As an initial step, we employed a submodeling approach using shell and beam elements in FE models, which accelerates the simulation process. Finally, the behavioral response of individual FE models is input into a Feed Forward Neural Network (FFN), and by incorporating this FFN into a 2-noded user element, we simulate the non-linear response of the bolted joints.

Featured Image

Photo by Charlie Wollborg on Unsplash

Photo by Charlie Wollborg on Unsplash

Why is it important?

The proposed method enables the development of an abstract model of a bolted joint that can simulate nonlinear behavior under finite strain conditions much faster than a high-fidelity model. This makes it possible to simulate large structures containing field of bolts and exploit the full potential of composite material.