As the number of years of operation of an in-service tunnel increases, typical apparent diseases such as cracks and water leakage, occurs, affecting the safe operation of the tunnels. How to identify various diseases through effective technical means and feedback the results to the maintenance unit is particularly important. Through the investigation of tunnel disease detection technologies and methods at home and abroad, the principle of using machine vision technology to measure typical apparent diseases of tunnels is analyzed. Combined with computer technology, an intelligent identification system for typical apparent diseases of tunnels based on zoom cameras is developed to realize the appearance of tunnel linings. Intelligent identification and monitoring of diseases, and the use of calibration plates to simulate cracks and water seepage in solid tunnels. After experimental tests, the monitoring system can realize 7×24 hours real-time monitoring of tunnel diseases. Under the condition of excellent shooting environment, the recognition rate of tunnel apparent diseases can reach 95%. The minimum width of monitoring cracks is 0.2mm, and the minimum leakage area is 50 cm². That is to say, the system can realize the identification and all-weather monitoring of multiple diseases in multiple locations and types.

The prediction of joint opening-closing deformations helps improve the operational safety of immersed tunnel systems. Addressing issues such as the poor accuracy and limited applicability of existing models, a novel method for predicting the joint opening-closing deformations of immersed tunnels is proposed. Firstly, the Sparrow Search Algorithm (SSA) is used to adaptively optimize the hyperparameters of the Long Short-Term Memory (LSTM) network, constructing an SSA-LSTM model to achieve feature learning and preliminary prediction of the deformation information. Based on this, the Ljung-Box (LB) test is employed to analyze the residual sequences, and a Wavelet Neural Network (WNN) is introduced to further extract effective information from the residual sequences, obtaining corrected values for the residual sequences. Finally, the prediction results from the SSA-LSTM model and the corrected residual sequence results are superimposed and reconstructed to obtain the predicted values of the joint opening-closing deformations. The proposed method is validated based on the immersed tube tunnel project of the Hong Kong-Zhuhai-Macao Bridge. The results show that: The proposed model exhibits excellent overall prediction performance, effectively mining useful information from the deformation sequences and having an advantage in considering the local features of the deformation sequences. The final prediction achieve an R~2 of 0.9994, RMSE of 0.007 2 mm, and MAE of 0.006 6 mm, which are higher in accuracy and stability compared to LSTM, SSA-LSTM, and traditional SVR, BP, and XGBoost models. This method can serve as a means to deeply explore the development and variation patterns of joint opening-closing deformations.