Prediction of durability of reinforced concrete based on hybrid-Bp neural network

Abstract

Mix proportion design has a significant influence on the durability of reinforced concrete (RC). Conventional simple equations encounter challenges in effectively guiding the enhancement of durability in the design process. To address this issue, a backpropagation neural network (BPNN) model with a topological structure of 6–16–2 is devised to build a prediction model for RC durability, offering both forward design and reverse guidance for determining optimal mix proportion. Moreover, the model is optimized by the Bat Algorithm (BA), Ant Colony Optimization Algorithm (ACO), and Particle Swarm Optimization (PSO) Algorithm. The input layer parameters of the model consist the water-binder ratio and the quantity of cement, coarse aggregate, river sand (RS), fly ash (FA), slag, while the output layer parameters include the failure time for corrosion current density of reinforcement (T1) and the failure time for concrete damage degree (T2). The dataset for the model comprises 100*2 sets, which are divided into 70*2 sets for training data, 15*2 sets for validation data, and 15*2 sets for testing data. The relationship between the experimental raw materials and the durability of RC was determined through correlation analysis, and comparative analysis was conducted on the durability evaluation indices for RC. The results indicate a positive correlation between cement content and the durability of RC, whereas the content of RS, coarse aggregate, FA and slag, and water-binder ratio exhibit negative correlation with concrete durability. Among the RC groups, the A3 group demonstrated the highest performance, and the D group stood out as the optimal mineral admixture group. Utilizing the PSO-BPNN model, the following performance indices were predicted for T1: R2=0.850, MAE=38.14, MAPE=0.094, RMSE=47.218, and for T2: R2=0.872, MAE=34.541, MAPE=0.071, RMSE=42.355. Consequently, the PSO-BPNN model demonstrates the highest accuracy in predicting the correlation between concrete mix proportions and the durability of RC, thereby offering valuable guidance for the mix proportion design of RC.