Predicting reinforcing bar development length using polynomial chaos expansions

The bond stress of a reinforcing bar in a cementitious matrix varies along the bar length and is difficult to quantify. Thus, design code provisions refer to the concept of bar development length and rely on statistical analysis of rebar-pull-out test results. In the present study, a novel data-driven predictive model based on Polynomial Chaos Expansions (PCE) was developed to predict the reinforcing bar development length using 534 experimental results of simple pull-out tests on short unit bar lengths. The predictive capability of PCE was compared to that of other data-driven models, namely the Response Surface Method (RSM) and Artificial Neural Networks (ANN). Moreover, predictions of the PCE, RSM and ANN were further compared with calculations of three commonly used design code formulas (i.e., ACI 318-14, ACI 408R-03, and Eurocode 2) and predictions of two existing empirical models (i.e. Model Code 2010 and Hwang et al. model). A parametric study was conducted to explore the sensitivity of the proposed model to influential input parameters. It was found that the Polynomial Chaos Expansions model offers a powerful predictive tool for reinforcing bar bond strength. The model was able to capture trends that differ from that of existing models that assume unrealistic uniform bond stress along the rebar. This flexible and data intensive model for predicting rebar bond stress and full embedment length could offer an intelligent platform for accommodating new bar materials, new test data, and calibrating existing design provisions to keep design codes relevant.