Collapse Vulnerability of Reinforced Concrete Buildings Using Neural Networks

In this study, an Artificial Neural Network (ANN) analytical method has been developed for evaluation the collapse vulnerability (earthquake performance) of reinforced concrete (RC) buildings . In this study, collected total of 260 reinforced concrete buildings with 4 storey, that were chosen to represent the existing RC buildings. The commercial program Sta4CAD is used for modeling and analysing these buildings. The performance analysis of these 260 RC buildings have been used for training neural networks. The parameters that affect on earthquake performance represent the input and the performance represent the output. In this study 16 parameters have been thought to be effective on the performance of RC buildings were considered: Torsional Irregularity (A1), Slab Discontinuities (A2), Projections in Plan (A3), Weak Storey (B1), Soft Story (B2), Discontinuity of Vertical Structural Elements (B3), Weak Column – Strong Beam (C2), Stirrup Spacing (cm), Average Shear Wall Ratio, Average Column Ratio (CA) , Concrete Compression Strength (C), Type of Steel (Fy), Soil Type (Z), Turkish Earthquake Code (1975– 1997- 2007), Earthquake Zone (EZ) and Importance Factor (I). The output parameters are the Structural Performance (S1-S4) was obtained based on the 4 performance levels in Turkish Earthquake Code-2007 (TEC-2007). The performance analysis of RC buildings was performed according to both the linear performance analysis and nonlinear (static pushover analysis) procedures as specified in TEC-2007. iv 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 is the most significant structural components that affect. The projections in plan and slab discontinuities were determined to be the least significant parameters. According to the study, the prediction accuracy of ANN has been found 90% accuracy for nonlinear (pushover analysis method) and about 89% accuracy for linear performance analysis method. Keywords: Artificial neural network, collapse vulnerability, earthquake performance based design.