Sensor equipped vehicles for bridge structural monitoring
Sensor equipped vehicles for bridge structural monitoring
Embedded Files
Over the course of three years I've developed two methods for the structural health monitoring of bridges, based on the principle of vehicle bridge interaction. A vehicle, equipped with vertical sensors recording the vertical acceleration of the vehicle as it crosses the bridge can be used to extract information on the state of the bridge, the methods are briefly described below.
The programs were implemented using MATLAB and consists of several codes:
Finite elements analysis
Ensemble Kalman filter for both input and output predictions
Genetic algorithm
Method 1: unsupervised learning based approach for bridge damage detection based on the data collected from multiple vehicles
Method 1: unsupervised learning based approach for bridge damage detection based on the data collected from multiple vehicles
Vehicle response data is collected from multiple vehicles crossing the bridge. The collection is done under multiple temperatures and traffic conditions which allows for a robust prediction of the bridge condition even under variations in the environmental and operational conditions. PCA is applied on the collected data and the mapping matrix is used to compute the loss in information during the PCA transformation is applied.
The PCA transformation matrix only accounts for the environmental and operational variations and hence unless the health of the bridge is affected the error should be small. Hence by constantly keeping track of the values of the error it's possible to detect the presence of damage on the bridge structure.
The following example is considered. A bridge made out of concrete and steel is modeled by a beam structure.
The properties of the steel and the concrete vary with respect to the temperature.
Data from vehicles are collected over the year, under different traffic conditions and different temperature conditions. Assuming that no damage is initially present a baseline is established
Damage is simulated by a reducing the stiffness of a beam element. The results are shown below.
2400 vehicle runs are simulated. The first batch is used to construct the baseline (training), the second batch is used to validate the PCA (testing) and the last batch of vehicles crossings are simulated under damage conditions.
The method clearly works at detecting damage in the bridge.
Method 2: Bridge numerical model updating using vehicle collected data
Method 2: Bridge numerical model updating using vehicle collected data
The proposed method rely on vehicle collected data to extract an input-output relationship that can be used to update a numerical model of the bridge. The method is not affected by the presence or the variability of road roughness. The input-output relationship is extracted from the vehicle acceleration data through an ensemble Kalman filter and the properties of the bridge are identified through a genetic algorithm optimization procedure.
The procedure is summarized in the flowchart. A numerical example is carried out to prove the efficiency of the method. The following bridge is considered
A bridge with multiple cross sections is considered.
The following results are obtained using a Genetic Algorithm. We observe a very good match between the real and the predicted bridge properties. Which proves the efficiency of the method.
Page updated
Google Sites
Report abuse