基于多源局部响应信号的钢筋混凝土梁承载性能评估OACHSSCD

As a key load-bearing component of the structure,the bearing performance evaluation of reinforced concrete beams is directly related to the safe operation of the overall structure.This study takes reinforced concrete beams as the research object,and implements batch modeling and calculation of reinforced concrete beam models with random initial damage based on ABAQUS and Py-thon secondary development scripts.Multi-source response database such as strain,acceleration,and strain cloud maps are constructed.Based on multi-source data of reinforced concrete beams,a two-stage bending capacity evaluation model was proposed using deep learning methods:(1)For the pre-diction of elastic stage bearing capacity,a multi-channel convolutional neural network was used for damage rate identification,and a relationship model between damage rate and elastic stage bearing ca-pacity was constructed.The results show that the combination of dynamic acceleration and strain cloud maps had a good prediction effect on damage rate,with an accuracy of 91.3％;(2)For the prediction of ultimate bending capacity,a prediction model for ultimate bearing capacity was construc-ted using long short-term memory network based on the temporal characteristics of strain signals and the spatial characteristics of cloud maps,with an accuracy rate of 98.6％.The deep learning network prediction in this paper was validated using indoor RC beam four-point bending tests.The results show that the identification error of damage rate was less than 0.94％,and the prediction error of ultimate bearing capacity was less than 2 kN.