Research On Safety Assessment Method Of Post-earthquake RC Structures Based On Computer Vision

China as a country with frequent earthquakes,the post-earthquake damage investigation of post-earthquakes is of great significance to the recovery and reconstruction.At present,the common method of post-earthquake investigation of structural damage is visual inspection,which requires inspectors to determine the damage severity of the structure following standards.Although proven effective,the evaluation procedure is time-consuming,high-risk,and evaluation accuracy relies on inspectors’ professional knowledge and subjective experience.In addition,structural health monitoring(SHM)can be applied to monitor the real-time damage of the structures and generate reliable data,but it is generally used to monitor the dynamic characteristics of early damage to the structure while damage investigation is often aimed at assessing damage to post-earthquake structures.For realizing the structural safety assessment,a novel approach based on computer vision was proposed.The main research contents are as follows.The damage failure modes and damage region of the reinforced concrete(RC)components are studied.Using ABAQUS to simulate the low-cycle loading test of the components of reinforced concrete(beams,columns,and shear walls),the simulation results are compared and analyzed from the load-displacement skeleton curve according to different seismic parameters.The damage failure process of the components is studied,and the damage failure region is divided according to the compression damage contour plot under different failure modes.It is the foundation of the damaged location of the components.A damage quantification method based on RC components and a structural safety assessment method is proposed.According to the five-level damage states,the relationship between the skeleton curve and damage index es is established through the modified Park-Ang two-parameter damage model.The damage states of the components are quantified through the damage index.The visible damage is extracted by studying the damage failure process of the component.The relationship between the damage state and damage index is obtained,laying the foundation for the damage detection method based on computer vision.In addition,the structural safety assessment based on the number of damaged components and the damage index is proposed respectively based on the standards to achieve the cross-scale assessment from component to structure.A multicategory damage detection and safety assessment system for post-earthquake reinforced concrete structures is established.The post-earthquake damage images of RC structures are collected on the post-earthquake investigation in Beichuan of China and copyright-free images from the internet.Five damage classes are detected using transfer learning,including check crack,wide crack,concrete spalling,rebar exposure,and rebar buckling.The network’s performance from the input image size,optimizer,etc.,was optimized and compared with the current popular single-stage object detection networks in terms of detection speed and accuracy to verify the practicability of YOLOv4.The visible damage,damage state,damage index,and damage failure mode are integrated using the Py Qt5 module.A system based on YOLOv4 damage detection and safety assessment is established to realize the safety assessment of post-earthquake structural damage.