DAMAGE DETECTION OF TRUSS STRUCTURES BY APPLYING MACHINE LEARNING ALGORITHMS

Abstract

Infrastructures including bridges constructed in the period of high economic growth are getting
older. For the damage detection of truss structures, this study assumes to utilize vibration signals obtained from
sensors installed into the bridges. By preparing damaged and non-damaged bridge structures, large quantities
of response data are generated. AR (Auto-Regressive) model is then applied to the time signals to extract the
structure’s soundness characteristics. Here, AR coefficients are values in which damaged structural
characteristics are reflected. Then, the machine learning technique is applied to the AR coefficients to classify
the structures into damaged and non-damaged ones. Results showed that the machine learning method
successfully detected the damage of truss members. This kind of SHM (Structural Health Monitoring)
technology is expected to contribute to early damage detection and preventive maintenance of bridges leading
to increase the accuracy of the damage detection of truss structures with low costs and fewer efforts for
maintenance.