An evaluation of effective design parameters on earthquake performance of RC buildings using neural networks

The extent of building damage is related to the features of the structural system, which have many parameters. In particular, it is difficult to determine the extent to which structural parameters affect structural performance to identify the main parameters that may cause damage. In the present study, changes in the quality of a load-bearing system and reinforced concrete (RC) structure materials during an earthquake were determined. The related structural parameters were determined by considering the structural damage parameters observed in earthquakes: concrete compressive strength, yield and ultimate strength of steel, transverse reinforcement, infill wall ratio, short column, strong column–weak beam, and shear wall ratio. A total of 256 RC buildings with between 4 and 7 floors were modeled, and pushover analysis was applied to each to obtain the building capacity curves. A performance assessment was performed predicated on the basic criteria of the Turkish Earthquake Code (TEC-2007), which was revised in parallel with FEMA-356. In addition, the influence of the structural parameters was determined using a set of Artificial Neural Network (ANN) algorithms, and a parametric study was performed accordingly. The load-bearing system and material were discussed by matching the findings obtained from the study with the documented damage from previous earthquakes. The effect of each parameter tested in this study had various affecting ratios on the earthquake performance of the structure. It was found that shear wall ratio and short column formation are the most significant structural components that affect performance. The compressive strength of concrete and transverse reinforcement were determined to be the least significant parameters. In addition, the ANN determined the structural performance with quite satisfactory rate. The earthquake performance estimation percentages of the selected ANN algorithms varied between 91.68% and 98.47% depending on the algorithm type and other parameters of the ANN modeling.