A Data-Driven Machine Learning Model for Predicting Demand-to-Capacity Ratio for RC Jacketing of Seismically Deficient Buildings

Existing reinforced concrete (RC) buildings are often vulnerable to earthquakes for various reasons, including noncompliance with seismic design standards, irregularities in mass and architectural plans, and construction defects. The columns of these buildings are not designed to withstand seismic forces; instead, they are meant mainly to support only gravity loads, making them susceptible to partial damage or system collapse during earthquakes. To enhance the capacity of such columns against seismic force, they are typically retrofitted using an RC jacketing system. The parametric modelling and design of any RC jacketed structure using computer-aided analysis and design can involve significant effort with time consumption. Therefore, to achieve a reasonable accuracy in design with lesser computational time, this work proposes a novel data-driven method to predict the demand-to-capacity ratio (DCR) of RC jacketed columns using an artificial neural network (ANN). To predict the DCR, various design variables associated with RC column jacketing are considered, like retrofit placement, concrete compressive strength, cross-sectional characteristics, jacket thickness, longitudinal reinforcement ratio, transverse reinforcement area, and slenderness ratio. In this study, a finite element (FE) program is used to generate datasets of jacket design to train a multilayer feedforward neural network (MLFFNN). Performance evaluations of the developed model are made by using four statistical measures - mean absolute error (MAE), mean absolute percentage error (MAPE), mean squared error (MSE), and the coefficient of determination (R2). The results show that the model successfully predicts DCR values with less than 7% variation compared to the FE solution, demonstrating the model's reliability, accuracy, and computation cost saving in predicting the DCR of seismically deficient columns in an RC building.