Deformation Response Characteristics of Steel-Plate-Strengthened Shield Tunnel Linings and a Rapid Assessment Method for Strengthening Effectiveness

During operation, shield tunnel linings commonly develop ovalization and cracking. Steel-plate strengthening is an effective measure for delaying structural performance degradation and extending the service life of shield tunnel linings. However, the effects of steel-plate strengthening parameters on the deformation response of shield tunnel linings are not yet fully understood. The associated control mechanisms remain unclear. Moreover, rapid and accurate evaluation of strengthening effectiveness is not achievable with existing methods. To address these issues, a rapid evaluation method integrating refined numerical simulation with the Bayesian optimization–random forest (BO-RF) model was proposed. A refined three-dimensional numerical model was established. The tracking element method was used to ensure continuity of the structural state of the shield tunnel linings throughout the steel-plate strengthening process. The reliability of the model was verified against previously reported full-scale model test results. On this basis, the deformation response of the shield tunnel linings under the reference load was predicted. The initial ovality of the shield tunnel linings and the steel-plate strengthening parameters were used as input variables to evaluate the deformation-control performance. Strengthening time, represented by the initial ovality before strengthening, had the greatest influence on the deformation response and deformation-control performance of the shield tunnel linings. Steel plate thickness had the second-greatest influence, whereas that of steel plate width was relatively limited. Within the parameter range investigated in this study, the deformation-control rate ranged from 16.53% to 91.02%. Four prediction models were constructed, and their hyperparameters were optimized using Bayesian optimization (BO). The following model types were considered: extreme gradient boosting (XGBoost), a one-dimensional convolutional neural network (1D-CNN), a support vector machine (SVM), and a random forest (RF). The corresponding models were denoted as BO-XGBoost, BO-1D-CNN, BO-SVM, and BO-RF, respectively. The best overall performance in terms of prediction accuracy and computational efficiency was achieved by the BO-RF model. The coefficient of determination (R2), root mean square error (RMSE), and mean absolute error (MAE) were 0.9817, 2.8704, and 1.7669, respectively. The model was trained in 206.6 s. Compared with numerical simulation alone, the computational time was reduced by 31.38 h. This reduction corresponded to a 99.62% reduction in computational time. Mechanistic analysis showed that tensile stress on the inner surface of the shield tunnel linings was redistributed through interfacial force transfer at the steel plate-lining interface. Local stress concentration and expansion of the damage zones were thereby inhibited. The steel plate was installed on the inner surface of the shield tunnel linings. Consequently, damage evolution was suppressed to a significantly greater extent on the inner surfaces than on the outer surfaces.