Study On Damage Identification Based On Optimal Sensor Placement And Ensemble Learning For Long-span Bridges

In the long-term service of the bridge,due to the external environmental factors such as environmental corrosion,vehicle load and the increase of traffic flow,certain damage may occur.if not found and repaired in time,it will affect its normal operation function.Therefore,it is particularly important to establish a bridge health monitoring system and timely conduct a safety assessment of the structural condition.Among them,the sensor subsystem for data acquisition,the damage identification and early warning system for data analysis are the core components of health monitoring.However,for long-span Bridges that generate large-scale data in real time online health monitoring,there are few researches on deep mining of monitoring data with a limited number of sensors.In order to obtain the overall health status information of the bridge as much as possible,how to optimize the number and location of sensors and use the mass monitoring data collected by a limited number of sensors to judge and evaluate the health status of the bridge is still a hot and difficult issue all the time.This paper takes the main bridge of the Wuhai Yellow River Extra Large Bridge in Inner Mongolia—the low tower cable-stayed bridge as the supporting project,and carried out study on the damage identification method of ensemble learning and the optimized layout of health monitoring sensors for long-span Bridges.The main research contents are as follows:(1)Aiming at the problem of sensor optimal layout,a new optimization algorithm is proposed to solve the problem of large number of degrees of freedom in sensor arrangement,the exponential explosion of the solution range and the concentration of extreme values for long-span bridges--Adaptive annealing chaos genetic algorithm,and the effectiveness of the proposed algorithm is proved through two sets of benchmark function extreme value optimization and classic TSP combination optimization problem.(2)The finite element model was established and the dynamic modal analysis was carried out based on the low tower cable-stayed bridge.According to the displacement modal characteristics,the number and location of the main beam acceleration sensors were optimized by the new optimization algorithm,and the specific sensors layout scheme for main beam was proposed.In order to facilitate the practical engineering application,combined with the algorithm parameter sensitivity analysis of the supporting engineering,the acceleration sensor optimal layout software suitable for the long-span bridge is developed by MATLAB.(3)The main beam of the low tower cable-stayed bridge is divided into several sub-areas,and the stiffness reduction is used to simulate the damage of different degrees.The moving vehicle load model is established by applying the moving vehicle load.The obtained sensor data is used to construct the damage index and the damage sensitivity analysis is performed to establish the damage sample set.(4)This paper introduced the more mainstream ensemble learning algorithms(Random Forest,XGBoost,etc.)in the field of machine learning,explored the research ideas of ensemble learning algorithms in damage recognition,and proposed a method of damage identification for large-span bridge based on multi-sensor features and ensemble learning,and applied to the low tower cable-stayed bridge.The results show that: the new optimization algorithm proposed in this paper has better global optimization characteristics,fast convergence characteristics,and higher robustness than traditional genetic algorithms,which can better solve the problem of optimal layout of long-span bridge sensors;the wavelet packet energy index constructed by using the optimal placement sensor monitoring data has strong sensitivity,thus the multi-sensor feature vector is taken as the fusion feature index for the identification of the damage area;Based on the multi-sensor feature vector and ensemble learning algorithm,the damage area of the low tower cable-stayed bridge was located and the recognition accuracy was more than 80%,which has a better recognition effect,so this method has a good application prospect in online real-time monitoring and damage early warning of long-span bridge health monitoring system.However,the ensemble learning method is relatively new and its application in the field of damage identification still needs to be further explored.