Long-term Deflection Prediction of Reinforced Concrete Structures Based on Sparse Polynomial Chaos Expansion

Long-term deflection prediction of reinforced concrete flexural structures is important for evaluating their serviceability and safety throughout their life cycle. Since it is difficult to consider all the influencing factors by empirical methods, in order to be able to accurately predict the long-term deflection of reinforced concrete structures. In this paper, we propose to predict the long-term deflection of reinforced concrete structures by using a sparse polynomial chaos expansion (PCE) model and perform global sensitivity analysis of the parameters affecting the deflection of the structure. Sparse PCE models are built and evaluated by using experimental datasets, compared with common surrogate models (RBF, SVR and Kriging) and common machine learning model (BP neural network). The models are trained and tested by using a ten-fold cross-validation algorithm. The results show that the sparse PCE model outperforms both common surrogate models and machine learning models in predicting the long-term deflection of reinforced concrete structures, with correlation coefficients R², relative average absolute error (RAAE), relative maximum absolute error (RMAE), and root-mean-square error (RMSE) of 0.970, 0.108, 0.537, and 0.062, respectively. Moreover, the sparse PCE model's RMSE value is much better than the empirical method. Finally, the parameters affecting structural deflection were ranked in order of importance based on the results of global sensitivity analysis of the sparse PCE, where the instantaneous or immediate measured deflection a(i) , span-to-depth ratio l/h , and age t ' concrete strength f_c ' are more important, and successively decreasing. The sparse PCE model can be used for long-term deflection prediction of reinforced concrete structures and can be evaluated to quantify the key factors affecting the long-term suitability of reinforced concrete structures.