Neural networks-based line element method for large deflection frame analysis

The line finite element method (LFEM) is the predominant simulation method in structural design due to its robustness in large-scale structural analysis. However, it sometimes suffers from the tedious computational process due to its fine-mesh requirement to ensure accuracy. The machine learning (ML) technique provides an efficient mesh-free alternative but necessitating tremendous training datasets for modeling large-scale structural systems. In this paper, a novel numerical framework, named the neural networks-based line element (NNLE) method, synergizing the unique advantages of the finite element method and ML technique, is proposed and presented within the context of large deflection frame analysis. The neural networks (NN) model is only trained for modeling single components, thereby significantly diminishing the model scale and the required training dataset. Then, the NN model is used to formulate a new NNLE and implemented within the existing LFEM framework to simulate the entire structural system. Extensive examples are performed to demonstrate the accuracy, efficiency, compatibility, and flexibility of the proposed NNLE method compared with the conventional LFEM and ML techniques. It is convinced that the proposed NNLE method will offer new insights into the combination of the traditional finite element method and the emerging ML approach.