Abstract This research employs machine learning (ML) algorithms to analyze bond behavior between fiber reinforced polymer (FRP) bar and geopolymer concrete (GPC). A synthetic dataset containing 208 instances, each described by nine independent variables, was assembled to train and evaluate six ML models utilizing Decision Tree, Random Forest, Extremely Randomized Trees, eXtreme Gradient Boosting, K‐Nearest Neighbors, and Artificial Neural Networks algorithms. SHapley Additive exPlanations were employed to unpack each model, revealing how individual models map the influence of each independent variable on five failure modes and bond strength in pull‐out tests. Confusion matrix comparison reveals that except for the ANN model, the remaining five ML classifiers exhibited exceptional predictive capability, achieving accuracy values exceeding 0.95 across the entire datasets. Besides, the KNN, ANN, and XGBoost models emerged as the most reliable for bond strength prediction for their R 2 values surpassing 0.96. An increased bond length and the use of larger‐diameter FRP bars are both associated with an increased likelihood for splitting failure. High strength of GPC or thick concrete cover induces surface damage in CFRP bars. Under conditions of high concrete strength, ribbed bar surfaces, and thick concrete layer, GFRP bars are more likely to rupture compared to CFRP bars. To ensure FRP bars achieve enhanced bond strength, an indent‐winding ribbed surface is suggested over spirally‐glued ribs. Sand‐coated FRP bar surfaces do not consistently benefit bond strength, as their performance is governed by the competing effects of increased surface roughness and interlayer delamination between sandblasted layer and resin matrix. For reliable bond strength assessment in pull‐out tests, a concrete cover thickness exceeding 65 mm is advised for FRP bars.
Zhou et al. (Fri,) studied this question.