A machine learning approach to damage detection of bridges

This study proposes a method for damage detection on bridges using its operational vibrations under a moving vehicle. The method uses signal energy of acceleration responses as the damage sensitive feature. Outlier analysis using one class support vector machines are used to classify the data and detect the presence of damage in the system. Factors influencing the accuracy of damage identification are investigated. It is observed that using adequate training data and enough number of sensors, a reasonable estimate on the presence of damage can be obtained. The method can handle the operational variability of the vehicular traffic such as mass and velocity of the vehicle. The method is illustrated using simple numerical simulations.