A machine learning framework for the shear limit state reliability of existing FRC beams without stirrups

Abstract A machine learning framework based on Gaussian process regression (GPR) for designing and conducting the shear reliability assessment of exisitng fibre reinforced concrete (FRC) beams without stirrups is presented. The GPR model has good predictive capabilities in terms R2, IA, RMSE, and MAE. Sobol-based Analysis of Covariance (ANCOVA) reveals the tensile strength of fibres fuf, beam width bw, and shear span to effective depth ratio a/d to be the main contributors to the variability of the shear capacity. Reliability assessment considered in a case study of an existing FRC beam designed using the present GPR confirms the structural reliability of the underlying GPR model.