Machine Learning Based Seismic Damage Assessment Of Reinforced Concrete Columns

Reinforced concrete(RC)columns are the most basic vertical load-bearing members in various engineering structures such as buildings,Bridges and hydraulic structures.The damage and performance degradation of RC columns during earthquake action affect the local and overall safety of structures.Damage assessment is an important means to study the mechanism of disaster evolution of soil-reinforced concrete columns under earthquake action.At present,scholars at home and abroad generally establish specific damage models through pseudo-static tests to quantify the seismic damage of reinforced concrete columns,but it is not realistic to carry out relevant experimental research after the actual earthquake test.In addition,the current intelligent seismic damage assessment methods only focus on the qualitative classification of damage types of reinforced concrete columns and the identification and location of damage,which cannot solve the quantitative assessment problem of seismic damage degree of members,and the postearthquake assessment is not really intelligent and automated.To solve the above problems,this paper studies the intelligent assessment method of earthquake damage degree of reinforced concrete columns based on computer vision and machine learning,and the classification assessment results of component performance degradation based on K-means clustering algorithm.An improved Park damage model considering the nonlinear accelerated accumulation effect of earthquake damage is established.An intelligent evaluation method for earthquake damage degree of reinforced concrete columns based on appearance damage parameters and component information is proposed.The main contents include:Firstly,an improved PARK model considering the nonlinear acceleration accumulation effect of reinforced concrete column damage during quasi-static tests was proposed.The performance leveling division method of members based on strength,ductility and stiffness was studied,and a quantitative evaluation method of component performance leveling was established based on the improved PARK damage model.According to the 95% guarantee rate,the range of damage indexes under different performance levels is determined.Based on the above model for reinforced concrete rectangular section 124 column quasi static test results of earthquake damage assessment,the results show that the proposed improved Park damage model compared with the original model has higher accuracy and universality,for most of the components can be well characterized its nonlinear acceleration accumulation laws of development of earthquake damage,And it has a uniform and clear upper and lower bound.Secondly,a method for evaluating the degree of seismic damage of reinforced concrete columns based on appearance damage parameters and component information is proposed.The horizontal,vertical length,number,and concrete spalling area of concrete cracks are extracted from the target detection frame of the appearance disease image of the pseudo-static test of reinforced concrete column components.Appearance damage parameters such as the exposed area of steel bars,and design parameters such as component damage parameters and shear span ratio,axial compression ratio,and reinforcement ratio as input to the network.The damage index calculated by the improved Park model is used as the actual value of the network output,and the appearance damage parameters are established The deep neural network model of the mapping relationship between the component information and the damage index;based on the above method,the damage prediction of the image sequence of the pseudo-static test is performed.The result shows that the intelligent evaluation model can accurately reflect the damage development trend,and the predicted value of the damage index on the verification set is compared with The true value correlation coefficient is above 0.98.Finally,the deep neural network structure parameters and training hyperparameter optimization and generalization performance research were carried out,and the machine learning model with the best damage index prediction performance was obtained.The average accuracy rate of the damage assessment of reinforced concrete columns with different design parameters was more than 75%,The average correct rate of evaluating the damage degree of reinforced concrete under different earthquakes is more than 88%.