Prediction Of Blast Loads And Rapid Evaluation Of Disaster Effects In Complex Urban Environments

Due to the complexity of urban environment,terrorist attacks or accidental explosions will cause damage and even continuous collapse of buildings within the radius of several kilometers,resulting in serious casualties and property losses.Rapid and accurate prediction of blast loads in complex urban environments can provide reliable basis for structural damage assessment,and visualization of disaster effects can provide guidance for pre-disaster defense and post-disaster rescue.The irregular nature of buildings in urban environments means that the simulation of blast wave propagation is beyond the scope of semi-empirical methods.Although CFD can accurately capture the characteristics of shock wave propagation in complex environments,but the downside is the high associated computational cost while ensuring the accuracy of the predictions,especially for the city-scale blast scenarios.Based on machine learning algorithm,this paper proposes a rapid prediction method of blast loads in complex urban environment,and establishes the process framework of explosion disaster simulation including the rapid modeling and the disaster visualization,providing guidance for urban blast disaster rescue.The main research contents and conclusions are as follows:(1)The open source software blastFoam was used to establish numerical models of different scaled distances.The mesh sensitivity analysis was conducted based on Richardson Extrapolation method,through which the appropriate mesh sizes were obtained.Two different experiments were compared with numerical models.The results show that the numerical simulation method and mesh size adopted in this paper can effectively simulate the propagation of blast wave in complex environment,and the numerical results are in good agreement with the experimental data.(2)The datasets including multiple explosion scenarios were generated by validated high-fidelity CFD models,and a rapid prediction method of blast loads based on machine learning is proposed.The blast loads obtained by coarse mesh models were set as the inputs and the real blast loads were set as outputs.The performance of artificial neural network(ANN)and Extreme Gradient Boosting trees(XGBoost)algorithms were compared.Subsequently,the better performing XGBoost is utilized with K-Means clustering algorithm to develop data-driven models.The clustering pretreatment significantly improves the accuracy of the model.At last,the models are evaluated using unseen input data to show that the proposed method has strong generalization capability.(3)The whole process simulation framework including regional building modeling,multi-stage remapping and result visualization was proposed.Based on this framework,a community was taken as the research object to simulate different blast scenarios.The damages were visualized combining with different damage criteria.High fidelity visualization results of shock wave propagation were obtained with satellite map.According to the simulation results,the influences of different building layouts and different explosive equivalents on the propagation of blast wave and damages of building structures were compared and analyzed.