engleski

Implementing neural networks for FE post-processing

In recent years machine learning has become more popular especially since open source libraires such as TensorFlow have become available. Neural networks have been used in computational mechanics for a long time [1] but were a fringe area. However, with an increase in computational power and availability of easy to implement libraries such as Keras, machine learning is quickly gaining more traction [2, 3]. In this paper the possibility of using neural networks to post-process finite element displacements is investigated and proven to be a viable application. This proof of concept was done on 2D membrane finite elements and a reduction in computational time was observed.

Deep learning; finite elements; data driven modelling

nije evidentirano