Fiber reinforced plastics (FRP) are often used to enhance the capacity of the reinforced concrete beam (RC beam). However, the debonding failure is often observed due to the effect of the complex environment and the random loads. The debonding failure includes three types: plate end interfacial debonding (PE debonding), critical diagonal crack-induced debonding (CDC debonding), and intermediate flexural crack-induced interfacial debonding (IC debonding). In order to investigate the IC debonding strain of RC beam strength by FRP, this paper proposed some data-driven models to explore the IC debonding strain, based on the machine learning approaches The concrete strength, shear span proportion, the proportion of anchorage length to shear width, tensile reinforcement proportion, steel yield strength, stirrup reinforcement ratio, FRP stiffness, and the proportion of sheet span to beamwidth were regarded as the input parameters. The IC debonding strain was regarded as the output. It was found that the BP model can predict the IC debonding strain well. However, the BP data-driven model is easy to fall into a local minimum, and it is very difficult to converge, which has a negative effect on the accuracy of the model. The Sparrow Search Algorithm (SSA) was proposed to update it. The results indicated that the neural network optimized by SSA with lowest relative error, which can predict the IC debonding strain well. In addition, a study on the importance of each input found that the concrete strength, shear span proportion, and reinforcement yield strength will have a big impact on the IC debonding strain.
Read full abstract