A mask-guided multi-LSTM-CNN architecture for rapid seismic analysis of existing RC frames

This paper presents an efficient Mask-Guided Multi-LSTM-CNN network designed for the seismic analysis of existing reinforced concrete (RC) frames. In contrast to traditional non-linear static analyses, such as pushover analysis, which can be time-consuming and computationally expensive, the proposed neural network offers a rapid and accurate alternative for predicting the capacity of structures, identifying potential weaknesses, and guiding seismic upgrading interventions. The model integrates information from three complementary sources: structural topology, global parameters describing the building as a whole and local parameters characterizing each column. A key feature of the proposed approach is its adaptability to buildings with varying configurations, particularly regarding the number of storeys and column layout. This demonstrates its ability to generalize across diverse structural conditions and proves its robustness in real-world applications. Furthermore, the model significantly reduces computation time compared to traditional Finite Element Method (FEM) simulations, making it a practical tool for large-scale seismic assessments. The results highlight the model’s efficiency and flexibility and, in so doing, it aims at contributing to the growing interest demonstrated by the Structural Engineering community towards the application of ML tools as possible alternative to more traditional FEM-based mechanistic approaches.