Calibration of models predicting the load-bearing capacity of bonded anchors using a genetic algorithm

The paper deals with the topic of resistance of bonded headless post-installed anchors to uncracked concrete, in particular with the calibration of selected models predicting their ultimate tensile load. In particular, the optimization of two models for predicting the tensile capacity of an anchor is investigated in this paper. First model is based on the determination of the ultimate capacity based on separated failure modes (failure of the extraction of concrete cone and the bond failure). The second optimised model, proposed in this paper, summarizes the effect of both of these parameters into a single exponential function. A model combining the effect of concrete strength and adhesive strength has the potential to better capture the true nature of failure in which both materials are involved. In order to compare these models, an extensive database of experimental results was compiled (including own experiments and also results from various authors). The calibration consisted in finding the most appropriate values of the individual input parameters of the models to fit the experimental results as closely as possible. The models for predicting the tensile capacity of anchors are multiparametric. Therefore, a method using elements of genetic algorithms was used for optimization, suitable for this purpose. Several possible statistical evaluation criteria were used for the evaluation of the fit of the models. The optimization of the models showed that the proposed model combining the effect of concrete strength and bond strength can be optimized to better fit the experimental results. •The model of separate failures is overestimated when the failure is approximately equally affected by both.•The model of combined concrete failure and bond failure expressed by a single function can be easily optimised.•The optimization of multiparametric functions using the genetic algorithm method has proven to be an effective tool.