Research On Parameters Of The RC Beam-Column Lumped Plastic Hinge Model Based On Neural Network

To ensure the accuracy of structural elastoplastic analysis,it is necessary to establish a structural analysis model that can accurately reflect the mechanical properties of components.The lumped plastic hinge model is often used to simulate RC beams and columns.This method can effectively characterize the elastic-plastic response of members under cyclic loading with high efficiency.The parameters of the constitutive model directly affect the calculation accuracy,so it is necessary to calibrate the parameters through the experimental data.At present,empirical formulas are often used to select the parameters of lumped plastic hinge model.However,there is a large error in the calculation results due to the lack of sufficient generalization ability of the empirical formula.The Artificial Neural Network（ANN）algorithm with strong nonlinear approximation ability and self-learning ability can solve this problem.It is suitable for establishing the relationship between model parameters and component characteristic parameters.In this paper,based on a large number of test data of RC beam and column under cyclic loading,the ANN model is implemented to associate the Pinch-IMK model parameters with components.A hysteretic loop prediction model（HLPM）is proposed and used to predict the Pinch-IMK model parameters to improve the accuracy of elastoplastic analysis of frame structures.The main research work and achievements of this paper are as follows:（1）Establish the test database.The digital database constructed by 91 RC beams and 243 RC columns under cyclic loading,which covers the common scope of engineering design and has strong applicability.The database contains 1)basic information: failure shape,loading mode,material strength,geometric information and reinforcement information;2)Test results:hysteresis curve and skeleton curve.（2）The specimens in the database are adopted to identify the values of Pinch-IMK model parameters,and the data sets which can be used for ANN training and testing are obtained.Four key hysteretic parameters are selected by parameter influence analysis.A differential algorithm fitness function considering the restoring force,hysteretic energy dissipation and reloading section stiffness of hysteretic curve is proposed,which is also used as a comprehensive error index to evaluate the fitting effect of hysteretic curve.The visualization program of parameter identification is developed,the geometric algorithm is used to identify the skeleton parameters,and the differential evolution algorithm is used to identify the key hysteretic parameters.（3）Evaluate the calculation accuracy of the existing methods.The skeleton parameters of all specimens in the database are calculated by empirical formula and section analysis,and compared with the identified skeleton parameters.In addition to the high average accuracy of peak strength calculated by section analysis,the error of existing methods in calculating skeleton parameters is generally large.The calculation results of empirical formula often overestimate the bearing capacity of RC beams and underestimate the bearing capacity of RC columns;the calculation results of section analysis for RC beams can’t simulate the strength decline segment,and the ductility is not consistent with the test.For specimens with large axial compression in section analysis,the ductility of specimens is often underestimated and strength softening occurs prematurely.（4）A hysteretic loop prediction model（HLPM）is proposed and the accuracy of the model is evaluated.The ANN model is implemented to associate the Pinch-IMK model parameters with RC beams and columns.Using cross-verification method and grid search to determine the super-parameters of neural network.The input parameters of ANN are screened based on the existing empirical formula and the disturbance analysis of input variables.By comparing the parameters of skeleton feature points predicted by HLPM with the results of section analysis and the calculation results of existing empirical formulas,it can be seen that the prediction results of HLPM have higher accuracy.By comparing the hysteresis curve calculated based on HLPM prediction parameters with that of IMK model based on empirical formula and fiber model,it is found that the comprehensive error of HLPM is the smallest and is in a small value.The hysteretic curve predicted by HLPM is more accurate,which can better represent the bearing capacity,strength degradation,stiffness degradation and pinch effect of RC beams and columns.