RESEARCH ON PARAMETERS OF THE RC BEAM LUMPED PLASTIC HINGE MODEL BASED ON DEEP LEARNING

The lumped plastic hinge model is often used in component-based elasto-plastic analysis of structures. The calculation accuracy depends on the selection of model parameters, but empirical formulas are difficult to describe the complex nonlinear relationship between the component mechanical characteristics and model parameters. It collects the experimental results of RC beam under revered cyclic loadings as the database. The pinch Ibarra-Medina-Krawinkler (Pinch-IMK) model was used as the basic hysteretic model of RC beams, and a lumped-plastic-hinge-model parameter prediction model of RC beam is proposed based on deep learning. By the grounds of experimental data, the Trilinear skeleton feature point parameters and hysteresis parameters are identified. 182 sets of skeleton feature point parameters and 91 sets of hysteresis parameters are obtained. Take the component feature parameters as inputs, the skeleton feature point parameters and the hysteresis parameters as outputs to establish a Pinch-IMK parameters deep learning prediction model of RC beam, named HDLM. By comparing the results of fiber moment-curvature analysis and that of existing empirical formulas, it is obvious that HDLM predicts the skeleton feature point parameters more accurately. By comparing the hysteretic curve calculated based on HDLM prediction parameters with that of IMK model based on empirical formula, it can be seen that HDLM is more accurate in predicting hysteretic curve and can better represent the strength degradation, stiffness degradation and pinch effect of RC beams.