A machine learning-based time-dependent shear strength model for corroded reinforced concrete beams

Abstract

Shear strength of corroded reinforced concrete (CRC) beams is a key concern in the design and/or retrofit processes for an RC structure during its life-cycle. In this paper, we develop a machine learning (ML)-based approach for predicting the residual shear strength of CRC beams at different service times. To achieve this goal, we collected 158 shear tests of CRC beams and adopted one of the most representative ensemble ML algorithms, i.e., the gradient boosting regression tree (GBRT), to establish a predictive model for the shear strength. Six dimensionless variables indicating geometric dimensions, material properties, reinforcing details and the corrosion extent of the beam are set as the inputs, while the shear strength is set as the output. 70% of data are used as training set and the remaining 30% is used to evaluate the model performance. Results indicate that the model produces excellent predictions with an average R2 over 0.9. Moreover, five empirical shear strength models for CRC beams are also used for verifying the proposed GBRT model, which demonstrates that the proposed model has a distinct superior performance. Finally, a mechanical model is used to calculate the corrosion extent of reinforcements and thus a time-dependent shear strength prediction approach for CRC beams is developed based on the ML-based model. The proposed time-dependent prediction model is capable to provide the shear strength of CRC beams with any given service time.