Abstract: Sampling tests on the selfcompacting concrete mixture performance after onsite pumping of steel shell immersed pipe (pipes E1～E4) are conducted to ensure the quality of the self-compacting concrete in the steel-shell immersed tube in the warehouse. The sensitivity of key parameters affecting the post-pump performance of self-compacting concrete mixture in steel shell immersed tube is evaluated based on the grey relational analysis, support vector machine (SVM), and Bayesian inference. In addition, prediction models for the postpump performance of self-compacting concrete mixture are established. The results show the following: (1) The postpump performance of selfcompacting concrete mixture in steel shell immersed tube (i.e., molding temperature, slump flow, V-funnel flow time, L-type instrument H2/H1, and air content) is correlated with pumping distance, number of elbows, delivery time, and ambient temperature. Further, the sensitivity order for each factor is ambient temperature > delivery time > number of elbows > pumping distance. (2) The engineering practice shows that the established SVM nonlinear and Bayesian linear probabilistic prediction models have a high prediction accuracy and excellent robustness. (3) The SVM nonlinear prediction model has a higher prediction accuracy than that of the Bayesian linear probabilistic prediction model; however, the Bayesian explicit probabilistic model is more practical. Overall, the combined application of the two prediction model types directs the quality control of subsequent selfcompacting concrete pipes in steelshell immersed tubes in the ShenzhenZhongshan Link.

Key words: Shenzhen-Zhongshan Link, self-compacting concrete, mixture performance, grey correlation analysis, support vector machine, Bayesian inference