Complicated temperature fields and crossing high-speed trains offer new challenges for monitoring girder deflections in operating railway bridges. Behavior analyses and early warning of girder deflections are necessary for the diurnal operation of bridge structures. Regarding the Nanjing Dashengguan Yangtze River Bridge, the present paper identifies the temperature-induced/train-induced deflections of the bridge girder from deflection data obtained from the wavelet transform. The principal components of the bridge temperature field are obtained to verify the temperature sensitivity of the girder deflection. The effect of the temperature-induced girder deformation on train-induced girder deflections is analyzed using the train-bridge dynamic model. The statistical laws of the downward and upward of train-induced girder deflections are described using the t location-scale distribution. Based on the mutual updating of monitoring data and numerical model, the behaviors of the temperature-induced/train-induced deflections of the bridge girder can be captured accurately, and the warning threshold of girder deflections under the coupling effects of temperature and trains is determined rapidly. The results demonstrated that the influence of the temperature-induced girder deformation on the train-induced girder deflection is relatively small when compared with the train-induced girder deflection itself, with the increasing amplitude being no more than 3%. A real-time early warning of girder deflections for the long-span railway bridge can be conducted to indicate the abnormal changes in girder deflection less than 10 s, which is likely due to the deterioration of track irregularity on the bridge or damaged bridge components.

The evolution rule of temperature-induced deflection in main girders is an important index to evaluate the service performance of long-span cable-stayed bridges, which directly reflects the coupling effect between the vertical stiffness of the main girder and the tension of multiple cables. However, temperature-induced deflection is caused by the complex temperature field of the main girder, cable tower and cable, while monitoring data have documented a time-lag effect between the temperature and temperature-induced deflection. Hence, it is difficult to accurately describe and model the behavior of the temperature-induced deflection in a long-span cable-stayed bridge in service. To this end, by utilizing the advantage of long short-term memory (LSTM) network for time series prediction, a digital model in minute scale based on monitoring data and deep learning can be developed to predict temperature-induced deflection, and resolve the low precision caused by the single-point input and time-lag effect. Compared with traditional machine learning algorithm and linear regression, a deep learning LSTM network has the best performance. For the cable-stayed bridge in this paper, the mean absolute error of the LSTM model was even less than 0.5 mm, and with the combined hypothesis test, the early warning accuracy for the abnormal change of temperature-induced deflection could achieve a minimum of 0.3%.

The monitoring data makes it feasible to quickly evaluate the cracking of the prestressed concrete box-girder bridge. The live-load strain can accurately quantify the load effect and cracking of bridges due to its explicit datum point of signal. Based on the live-load strain data from bridge monitoring system, this study develops a comprehensive data-driven method of state evaluation and cracking early warning for the prestressed concrete box-girder bridge. The feature of vehicle-induced strain is extracted using the deep learning and classification of long short-term memory network. The vehicle-induced strain features are clustered via Gaussian mixture model, and the cracking early warning of the bridge is conducted according to the reliability of heavy vehicle clustering data. This method can be used as an indicator for the bridge inspection, truck-weight-limit and reinforcement work. The results demonstrate that (1) using the long short-term memory network, a deep learning model can be trained to intelligently classify the non-stationary and stationary sections of vehicle-induced strains, of which the test accuracy of classification surpasses 99%, and (2) according to the Gaussian mixture model probability distribution of data, the vehicle-induced strain features can be clustered by the corresponding Gaussian mixture model crest, which is the premise for reflecting relational mapping between vehicle loading and strain response.

Long-span bridges are the lifeline throats of urban transportation network. Deflection (i.e., deformation) behavior of long-span bridges is complex. It can be found from long-term monitoring data that there is an obvious time lag effect between quasi-static behavior of deflection and environmental temperature, and abnormal signals such as drift and jump-point appear sporadically in the deflection data. In order to deal with the interference from the data time lag and abnormal signal, this paper adopts the Bayesian multiple linear regression (BMLR) method to establish the mathematical model of bridge deflection based on temperature, other points’ deflection, or cable force data. A new paradigm of the recursive modeling strategy of BMLR for bridge deflection based on short-term data is proposed, which truly realizes the dynamic update ability of Bayes’ theorem in multiple regression modeling. Under the same conditions of modeling, the proposed paradigm performs higher accuracy of prediction and lower space of data storage occupied than the traditional multiple linear regression method, and is less time taken than methods of deep learning. The whole process was validated to be robust to the data time lag and abnormal signal. When faced with the situation of sparse sensing points and not enough long time of monitoring, it is possible to fast predict deflection of new-added/adjusted sensing points using short-term observation data.