FRP筋和箍筋加固混凝土梁的抗剪强度预测是结构工程应用中最复杂的问题之一。已经进行了许多实验和理论研究,以建立抗剪承载力与设计变量之间的关系。然而,由于剪切机制的复杂性,现有的半经验模型无法提供精确的预测。为了提供更准确和可靠的模型,采用机器学习(ML)技术研究了FRP筋和箍筋加固混凝土梁的抗剪性能。从报道的文献中汇编了由120个测试样本组成的数据库。实现了人工神经网络(ANN)以及ANN与遗传优化算法(GA-ANN)的组合,以开发ML模型。通过神经解释图(NID),关键的设计因素,即,光束宽度和有效深度,剪切跨度与深度比,混凝土抗压强度,FRP纵向配筋率,FRP抗剪配筋率,FRP纵向钢筋和FRP箍筋的弹性模量,被识别并确定为模型的输入参数。通过将模型预测与可用的测试结果进行比较,验证了所提出模型的准确性。GA-ANN模型的应用提供了更好的统计结果(平均值Vexp/Vpre等于0.99,R2为0.91,RMSE为22.6kN),并通过将R2值提高18.2%和RMSE值提高52.5%而优于CSAS806-12预测。此外,特别注意设计参数对抗剪承载力的耦合影响,这在文献中的模型和可用的设计指南中没有得到合理的考虑。最后,基于数据驱动回归分析方法,建立了考虑耦合效应的ML回归方程。分析结果表明,预测结果与试验结果一致,具有合理的准确性,该模型可以有效地应用于FRP筋和箍筋混凝土梁受剪承载力的预测。
The shear strength prediction of concrete beams reinforced with
FRP rebars and stirrups is one of the most complicated issues in structural engineering applications. Numerous experimental and theoretical studies have been conducted to establish a relationship between the shear capacity and the design variables. However, existing semi-empirical models fail to deliver precise predictions due to the intricate nature of shear mechanisms. To provide a more accurate and reliable model, machine learning (ML) techniques are adopted to study the shear behavior of concrete beams reinforced with
FRP rebars and stirrups. A database consisting of 120 tested specimens is compiled from the reported literature. An artificial neural network (ANN) and a combination of ANN with a genetic optimization algorithm (GA-ANN) are implemented for the development of an ML model. Through neural interpretation diagrams (NID), the critical design factors, i.e., beam width and effective depth, shear span-to-depth ratio, compressive strength of concrete,
FRP longitudinal reinforcement ratio,
FRP shear reinforcement ratio, and elastic modulus of
FRP longitudinal reinforcement rebars and
FRP stirrups, are identified and determined as input parameters of the models. The accuracy of the proposed models has been verified by comparing the model predictions with the available test results. The application of the GA-ANN model provides better statistical results (mean value Vexp/Vpre equal to 0.99, R2 of 0.91, and RMSE of 22.6 kN) and outperforms CSA S806-12 predictions by improving the R2 value by 18.2% and the RMSE value by 52.5%. Furthermore, special attention is paid to the coupling effects of design parameters on shear capacity, which has not been reasonably considered in the models in the literature and available design guidelines. Finally, an ML-regression equation considering the coupling effects is developed based on the data-driven regression analysis method. The analytical results revealed that the prediction agrees with the test results with reasonable accuracy, and the model can be effectively applied in the prediction of shear capacity of concrete beams reinforced with FRP bars and stirrups.