Zongchao Liu

This paper proposes a slope stability prediction model based on deep learning and digital twinning methods. To establish a reliable slope database, 30 actual slopes were collected, and 100 digital twin (DT) models were generated for each actual slope by fine-tuning the slope profiles. The safety factors of all slope samples were calculated using the Limit Equilibrium Methods (LEMs). A convolutional neural network (CNN) regression model was established, and the root mean square error (RMSE) was used as the evaluation indicator. In order to find an excellent CNN model, the K-fold (K = 10) cross-validation was used to compare the predictive effect of 1D CNN and 2D CNN on the slope safety factor. On this basis, CNN models with different network depths were compared. The results showed that the 2D CNN model with six convolutional layers had the best network prediction effect for the slope dataset. To validate the generalization ability of the model, an actual slope was input into the CNN model; its prediction result was 1.0229, and the absolute error with its real safety factor (1.0197) was 0.0032. With the slope stability prediction model proposed in this paper, the safety factor of slopes can be obtained from their geological and physical data, which greatly simplifies the calculation of the safety factor and has great engineering significance.

Unsteady aerodynamic forces are significantly critical to the safety and stability of trains traveling in high winds. This paper describes a study into unsteady crosswind forces of a moving train subjected to a three-dimensional stochastic wind field with longitudinal, lateral, and vertical turbulences. Initially, a three-dimensional computational fluid dynamic (CFD) model is established to calculate the aerodynamic coefficient of a moving train, and then the wind velocity time histories at the position of the train are generated. Finally, the quasi-steady theory and weighting function method are used to model the unsteady crosswind forces of a moving train in a three-dimensional turbulence field. The results demonstrate that a generalized sine form is useful for predicting the aerodynamic coefficient that varies with the resultant wind yaw angle, and an adequate modeling of unsteady crosswind forces with complete wind turbulences can produce a greater force fluctuation and peak. Particularly when the flow direction of crosswind deviates from 90°, consideration of only a portion of the turbulence components may underestimate the dynamic response of trains.

Automatic bridge surface defect detection is of wide concern; it can save human resources and improve work efficiency. The object detection algorithm, especially the You Only Look Once (YOLO) series of networks, has important potential in real-time object detection because of its fast detection speed, and it provides an efficient and automatic detection method for bridge surface defect detection. Hence, this paper employs an improved YOLOv3 network for detecting bridge surface defects (cracks and exposed rebar) and compares the effects of the advanced YOLOv2, YOLOv3 and faster region-based convolutional neural network (Faster RCNN) in bridge surface defect detection, and then two approaches (transfer learning and data augmentation) are used to improve the YOLOv3. The results confirm that, by combining high- and low-resolution feature images, the YOLOv3 improves the detection effect of the YOLOv2 (using single-resolution feature images); the average precision (AP) value of the improved YOLOv3 (0.9–0.91) is 6–10% higher than that of the YOLOv2 (0.83–0.86). Then, the anti-noise abilities of the YOLOv2 and YOLOv3 are studied by introducing white Gaussian noise, and the YOLOv3 is better than the YOLOv2. Simultaneously, the YOLO series of detectors perform better in detection speed; the detection speed of the improved YOLOv3 (FPS (frames per second) = 23.8) is 103 times that of the Faster RCNN (FPS = 0.23) with comparable mAP values (improved YOLOv3 = 0.91; Faster RCNN = 0.9). It is demonstrated that, in consideration of detection precision and speed, the proposed improved YOLOv3 is a decent detector for fast and real-time bridge defect detection.

Timely crack detection plays an important role in building damage assessment. In this study, an automatic crack detection method based on image registration and pixel-level segmentation (improved DeepLab_v3+) is proposed. Firstly, the moving images are calibrated by image registration, and the similarity method is adopted to evaluate the calibrated results. Secondly, the DeepLab_v3+ is improved and used to segment the fixed images and the calibrated images. Finally, the difference of crack pixels between the fixed and calibrated images is estimated, and the key parameter is investigated to find the optimal optimizer and learning rate. The results illustrate that: (1) the image registration technology shows excellent calibration achievement and the average error is only 4%; (2) with the resnet50 being selected as the backbone network of improved Deeplab_v3+, the automatic detection method proposed in this study is more efficient in comparison with other common pixel-level segmentation algorithms; (3) the best network optimizer of improved Deeplab_v3+ and learning rate of crack segmentation task are sgdm and 0.001, respectively. The crack detection method proposed in this study can significantly improves the technical level of crack detection in practical projects.