A review on the applications of artificial neural network techniques for accelerating finite element analysis in the civil engineering domain

Abstract

Finite element (FE) modelling is widely recognised as the most powerful and foremost computational technique for analysing complex structural systems due to its highly efficient modelling and simulation capabilities. Despite the strengths, its computational demands restrict its ability of performing instantaneous computations and present substantial challenges for achieving real-time analyses results. However, the integration of artificial neural networks (ANNs) with FE modelling offers a simplified calculation process whilst accelerating the computational time substantially. In this light, current study conducts a worth and timely investigation on application of ANNs as a surrogate for real-time FE modelling in the civil engineering domain. ANNs can be integrated into the computational process of FE modelling in numerous ways and selecting the type of the neural network architecture for surrogate modelling and integrating it to represent the FE model are crucial. Even though that depends on the complexity of the model that is to be analysed, the graph neural networks (GNNs) showed an excellent performance in both analysing and visualising mesh-based FE modelling as it incorporates the underlying relationship among different elements in a FE mesh. ANNs have the potential to significantly accelerate forward calculation process and also, provide a more efficient approach to inverse analysis. However, such applications are seldom in the existing body of knowledge. It has been identified that predicting stress or displacements was the most common primary goal of most studies with a particular emphasis on addressing static structural behaviour.