Evaluation and Prediction of Response Modification Factor for RC Moment Frames under Critical Consecutive Earthquakes using Artificial Neural Network

A large main shock may consist of numerous aftershocks with a short period. The aftershocks induced by a large main shock can cause the collapse of a structure that has been already damaged by the preceding main shock. These aftershocks are important factors in structural damages. Furthermore, despite what is often assumed in seismic design codes, earthquakes do not usually occur as a single event, but as a series of strong aftershocks and even fore shocks. In other word, structures that are located in seismically active regions often may be subjected to successive earthquakes which occurred with significant PGA in a short time after each other. For this reason, this paper investigates the effect and potential of consecutive earthquakes on the response and behavior of reinforced concrete structures. For this purpose, the response modification factor (R factor) which is one of the significant parameters in the structural design of buildings and decreases the lateral forces induced by earthquakes, is calculated and estimated for reinforced concrete moment frames under critical single and successive earthquakes. Thus, three reinforced concrete moment frames with 5, 7, and 12 stories are designed according to Iranian seismic codes (standard No. 2800) and modeled in Opensees software. After the design of the samples, critical seismic scenarios with/without successive shocks are selected from “PEER” center. Consecutive earthquakes not only occurred in the similar directions and same stations, but also their real time gaps between successive shocks are less than 10 days. In the following, R factors of RC moment frames are calculated from the results of incremental dynamic analysis (IDA(, time history and nonlinear static analysis (pushover). The results show about 20% reduction in the R factor and, also increment of damages under successive earthquakes comparing to the individual one. Finally, the idealized multilayer artificial neural networks, with the least value of Mean Square Error (MSE) and maximum value of regression (R) between outputs and targets were then employed to estimate the R factors. Theses artificial neural networks are designed based on the features of frame properties, successive earthquakes. Comparison of predicted R factors with real values indicates the adequate ability of networks in estimation of the results. So that, the average error for the artificial neural network model for predicting the calculated results from IDA, Pushover and Linear Analysis is less than 4%. To be more specific, more than 73% and 93% of the simulated R factors are within ±5% and ±10% of the real values for artificial neural network model. This is an indication that the networks have learned to generalize the unseen information well and reflects good precision in simulating. Moreover, it can be seen that the values simulated by the artificial neural network model spread around the 45 degree line which implies neither over-estimation nor under-estimation.