Forecasting the deterioration of cement-based mixtures under sulfuric acid attack using support vector regression based on Bayesian optimization

Immersion test is time-consuming and labor-intensive in evaluating the resistance of concrete against sulfuric acid attack. An alternative way for acid resistance evaluation is to predict the sample deterioration through machine learning methods using a currently available database. However, current predictive models have failed to include testing conditions, which limits their applications to certain testing conditions. Accordingly, predictive models need to be developed to include parameters of both mixture design and testing conditions in the forecasting of deterioration of mortar under sulfuric acid attack. In this study, such predictive models were established using the Bayesian optimized-support vector regression (BO-SVR) algorithm. Prediction errors were calculated, and a superiority test was conducted to evaluate the performance of the proposed BO-SVR models. It was found that the proposed BO-SVR model outperformed the other models in predicting the mass change and the compressive strength. This provides a new way of evaluating the acid resistance of cement-based materials.