Compressive Strength Prediction of Concrete Under Sulfate Attack Using Coupled Machine Learning Methods

One of the most significant factors affecting the durability of concrete is sulfate attack. In this paper, to predict the compressive strength (CS) of concrete under sulfate attack, three coupled machine learning methods (SSA-BP, PSO-BP and NGO-BP) were developed by coupling BP neural networks (BPNN) with three swarm intelligence algorithms, which are sparrow search algorithm (SSA), particle swarm optimization algorithm (PSO) and northern goshawk optimization algorithm (NGO), respectively. Twelve influencing factors related to material composition, erosion medium and exposure conditions are chosen as inputs, and the CS of concrete subject to sulfate attack is selected as the output. The database of 591 samples collected from published literatures is divided into three parts. Performance indexes are used to evaluate the three coupled models and BP independent model. Finally, the influence of each input on the CS of concrete under sulfate attack is examined using the Grey relational analysis approach. The following findings are reached: (1) all coupled models can predict the CS of concrete under sulfate attack with higher accuracy and achieve better performance than BP independent model, and the best one is SSA-BP model. Benefitted both from the strong nonlinear mapping ability of BPNN and from the global search and fast convergence ability of SSA, SSA-BP model has strong potential in predicting the CS of sulfate attack concrete. (2) Grey relational analysis shows that, among the twelve inputs considered, the initial compressive strength of concrete has the highest correlation (almost one) with the CS of concrete under sulfate attack. The robustness of the suggested model is confirmed by the relational analysis of all input parameters. (3) In addition, this model can provide an innovative way to assess the durability of concrete under complex or harsh environmental conditions.