Abstract
High-formwork support systems (HFSSs) play a crucial role in constructing complex projects. However, insufficient monitoring techniques can sometimes lead to catastrophic failures of these structures. This study presents an integrated framework for monitoring HFSS that combines numerical simulation, deep learning, and a Retrieval-Augmented Generation (RAG) system. A finite element model of HFSS was first developed and optimized using a Genetic Algorithm-Particle Swarm Optimization hybrid algorithm, generating structural response data for three critical conditions: Normal operation, local instability, and global instability. These datasets were then used to train a Recurrent Neural Network Long Short-Term Memory (RNN-LSTM) network, which can correctly predict the working states of the structure. Moreover, the effectiveness of RNN-LSTM and convolutional neural network (CNN) was compared. The results demonstrated that RNN-LSTM has superior performance over CNN in predicting the working status of HFSS. Furthermore, the study developed an RAG system incorporating GPT technology and a domain-specific knowledge graph to automate structural health monitoring (SHM) report generation. Several assessment metrics were involved to evaluate the RAG model's performance. The findings indicate that the RAG model could generate accurate and reasonable SHM reports for HFSS.