Comparison of artificial neural network and response surface methodology prediction in key performance of two-component grout material in shield tunneling

In this study, response surface methodology (RSM) and artificial neural network (ANN) techniques were employed to predict the key performance of a two-component grout material used in shield tunneling, considering the 28 d compressive strength, gel time, initial and final setting times, and water-to-land compressive strength ratio. A Box-Behnken design consisting of 17 mix combinations was used to investigate the effects of three key variables: water-to-binder ratio, water-to-bentonite ratio, and the volume ratio of component A to B. The results indicated that RSM provided an interpretable polynomial model but tended to oversimplify nonlinear interactions, whereas ANN captured complex multivariate relationships more accurately, thus yielding higher predictive precision and better adaptability to local variations. Specifically, ANN achieved a higher coefficient of determination (R2) and lower prediction errors for all target indicators. These findings confirm the superior modeling capability of ANN and provide practical guidance for the performance-driven mix design of two-component grout materials in tunneling applications.