Dynamic Blast Response Prediction of Assembled Structures Based on Machine Learning

This study proposed an innovative assembled blast-resistant composite structure integrating ultra-high performance concrete plates and ceramic foam layers, designed to enhance blast protection for a power valve hall hole blocking system. Based on the full-scale blast test and numerical simulation, the dynamic response of the structure under blast load was revealed. The parametric studies showed that when the thickness of the UHPC ribbed plate was increased from 30 mm to 40 mm, the maximum displacement at the edge of the hole was reduced by 60.9%. However, a further increase in thickness to 50 mm led to an increase in the inertia effect due to the high stiffness, resulting in a reduction in the maximum displacement value by only 8.61%. In addition, a machine learning framework combining generative adversarial networks (GANs) and Extremely Randomized Trees (ERT) model was constructed to predict the maximum displacement of the structure under blast loading. Furthermore, interpretability analysis by the (SHapley Additive exPlanations) SHAP algorithm verified the consistency of the decision logic of the ERT model with the physical mechanism of the explosion. This study established a full-chain design framework of structural design, mechanism research and intelligent prediction, which provided theoretical support and an intelligent tool system for protection engineering.