Machine Learning-based Parameter Optimization For Fiber Elements Of SRC Fream Beams And Columns

The reliability of performance-based seismic design method depends on accuracy of seismic performance indexes and precision of structural elastoplastic analysis.At present,elastoplastic analysis of SRC beam-column components is still needed to be improved,and the seismic performances of finite element models are quite different from that of actual components,especially for shear failure members with small shear-span ratio.To improve elastoplastic analysis accuracy of SRC beam-column components,this paper collects experimental data of SRC beam-column specimens,establishes prediction model for fiber element parameter of SRC beam-column member（PMFEP-SRC）using random forest algorithm,which optimizes parameter selection of fiber elements.The main research work and results are as follows:（1）Experimental database of SRC beam-column specimens was established.From relevant literature,low-cycle reciprocating test data of 153 SRC beam-column specimens was collected,whose hysteretic curves were extracted by manual,and then experimental database of SRC beam-column specimens was formed.Through statistics for main parameters of SRC specimens,it is proved that this experimental database can represent SRC beam-column components of actual engineerings.（2）Automatic elastoplastic analysis program for SRC beam-column member（AEAP-SRC）was developed.Taking the OpenSees as analysis program,using Python programming language for pre-processing and post-processing data,AEAP-SRC is developed,which can finish automatic modeling,loading and analysing,data processing and visualizing.On the base of AEAP-SRC,influence of each parameters of various material constitutive relations on elastoplastic analysis results is studied by single factor analysis method.（3）A method was proposed to transform experiment results of SRC beam-column components into label data.Both peak bearing capacity ratio R_F and energy dissipation capacity ratio R_E are taken as target parameters,to quantify the similarity between hysteretic curve of numerical analysis and actual hysteretic curve of SRC specimens.Taking bearing capacity adjustment coefficient C_F and stiffness adjustment coefficient C_S as adjustment parameters,the complex problem of finding optimal parameters for fiber elements is simplified to two independent approximately-linear problems.Combined with mountain-climbing algorithm,automatic solution for optimal parameters is realized.Compared with common fiber elements,it is proved that fiber elements with optimal parameters are more accurate.（4）Prediction model for fiber element parameters of SRC beam-column members（PMFEP-SRC）was established.Transforming control parameters of specimens into 11 representative characteristic parameters,and taking optimal parameters of fiber elements（bearing capacity adjustment coefficient C_F ＆ stiffness adjustment coefficient C_S）as label parameters,PMFEP-SRC is established with the help of random forest algorithm.Taking the retained 27 SRC beam-column specimens as test-set,it is proved that PMFEP-SRC has higher accuracy than other two common fiber elements.The comparison results are as follow.1)All three kinds of fiber elements have good performances on calculating peak bearing capacity of SRC beam-column specimens,in which error of PMFEP-SRC fiber element is 7.17%,error of Steel01 fiber element is 13.6% and error of Steel02 fiber element is 14.78%.2)Through reducing stiffness of fiber element,PMFEP-SRC fiber element can simulate mechanical properties of SRC beam-column specimens with various failure modes,whose error of energy dissipation capacity is 11.46%,while Steel01 fiber element and Steel02 fiber element can’t simulate hysteretic characteristics of SRC specimens with shear failure,whose errors of energy dissipation capacity are 42.11% and 32.36%,respectively.（5）Low-cycle reciprocating tests of 9 SRC columns with shear-span ratioλ=3 were completed,and based on the test data,it showed that PMFEP-SRC fiber element was still more accurate than common fiber elements in simulating SRC components with large shear-span ratio.The results are as follow.1)All three kinds of fiber elements have good performances on calculating peak bearing capacity of SRC specimens.2)Only PMFEP-SRC fiber element can well predict energy dissipation capacity of SRC column specimens,whose error is 5.19%.Both Steel01 fiber element and Steel02 fiber element overestimate energy dissipation capacity of SRC specimens,whose errors are 38.89% and 16.03%,respectively.