Reinforced Concrete (RC) frame buildings are prevalent worldwide due to their seismic performance and the broad availability of materials. Pushover analysis is regarded as one way to assess the seismic performance of these buildings, which is simple to interpret. The only downside of this method is that it requires developing nonlinear models, which can be time-consuming and computationally expensive. Recently, Machine Learning (ML) has emerged as an alternative to approximate the pushover response of RC frame buildings. However, there are still improvements in several areas, such as including relevant and easily obtainable parameters in structural design (e.g. reinforcement ratios), considering a wide range of building configurations, and comparing different ML techniques. This paper presents the development of two models based on Random Forest (RF) and Artificial Neural Network (ANN) to predict the pushover response of low-rise RC frame buildings, supported by a user-friendly graphical interface. First, a dataset of over 138000 Pushover analyses was performed on flexure-controlled RC frames ranging from two to five stories using OpenSeesPy, subsequently used to develop predictive ML models. These models are designed to offer a trilinear approximation of a building's Pushover curve, and they require input information that is available from the customary building design. The pushover predictions obtained with RF and ANN take 0.12 % and 3.05 % of the computing time compared to the OpensSeesPy nonlinear model. The proposed models demonstrate a high level of accuracy, with a coefficient of determination (R²) of over 0.88 and 0.81 in testing datasets for RF and ANN models, underscoring their potential to significantly streamline the seismic performance assessment process by providing quick Pushover curve estimations.
Read full abstract