Research On Aseismic Behavior Of Corroded RC Columns Based On Machine Learning

The offshore bridge under sea erosion environment is prone to corrosion damage in the long service.For the offshore bridge which is sited in the seismic area,the structural performance degradation caused by corrosion will seriously affect its seismic performance.For the study of seismic behavior of reinforced concrete columns in corrosive environment,the traditional research method is mainly based on experiment,and on the finite element simulation,supplementarily.This paper makes a new attempt to solve this problem by means of machine learning based on statistical learning theory.The main research contents include:Firstly,an OpenSees finite element model was developed for a quasi-static test of reinforced concrete in a corrosive environment.By comparing the finite element simulation results with the test results,the validity and accuracy of the modeling method based on corrosive materials constitutions modified in this paper was verified.Secondly,adopting finite element simulation as the data source,we selected 10 basic characteristics of corroded RC columns as input parameters,and the shear force values of characteristic points on the skeleton curve of RC column as output.The skeleton curve prediction model of corroded reinforced concrete column based on BP neural network was established.Neural network used mean square error as performance function,LM algorithm as training function,and gradient descent method to optimize network parameters.Besides,genetic algorithm was adopted to optimize the network initial weights and thresholds.The results showed that the skeleton curve model based on BP neural network has excellent prediction performance on the test set.Then,support vector machine(SVM)algorithm was adopted.10 basic characteristics of the column were still selected as inputs,and the cycle control points,peak points and shear residual points on the hysteretic curve of RC column were taken as outputs.A hysteretic curve prediction model of corroded RC column based on support vector regression machine was established.RBF function was selected as the kernel function in the SVR model,and the model parameters were optimized by grid search and K-fold cross-validation methods.The results showed that the SVR model was effective in predicting the characteristic points of hysteretic curves of RC columns,but the model did not take into account the unloading section of hysteretic curves.Finally,the prediction model was used to analyze the influence of immersion time in NaCl solution and longitudinal bar mass loss rate on seismic performance of RC columns.The results showed that the bearing capacity and ductility of RC columns were reduced under corrosive environment.At the same time,the influences of different mechanical parameters such as axial compression ratio,shear span ratio and reinforcement ratio on seismic performance of columns were analyzed.The results showed that the seismic behavior of RC columns obtained by machine learning model is consistent with that obtained by traditional research methods,basically.