Assessment of Damage in Composite Beams with Wavelet Packet Node Energy Features and Machine Learning

Abstract

Composite structures are commonly used in bridge and building construction. A typical example is a bridge deck comprising of concrete slab and steel beam girders, with shear connectors linking the concrete slab and the steel beams to form the composite effect. The behaviour of a composite structure is dependent not only on the condition of the main constituent components, namely concrete slab and the steel beams, but also on the effectiveness of the shear connectors. In fact, the influences of damage to the main components (herein called flexural damage) and damage to the shear connectors on the overall structural performance are distinctively different. Therefore it is important that damages in a composite structure be distinguished between these two types of parameters. However, not much attention has been paid in differentiating the flexural and shear connector damages in the existing damage identification literature. In this paper we will provide an overall discussion on the distinctive effects of flexural and shear-connector damages to the rigidity and its distribution in a composite beam. A vibration based approach is then presented for the identification of the mixed presence of flexural and shear-link damage parameters by means of supervised machine learning. The wavelet packet node energy (WPNE), which has been shown to be sensitive to structural changes in previous studies, is chosen as input features. Appropriate selection of the wavelet packet transformation levels in the case using a single measurement sensor, as well as using multiple sensors, are discussed. Results demonstrate that with WPNE features combined with supervised machine learning, it is possible to differentiate and identify flexural and shear-connector damages, and hence the actual structural condition, of a composite beam.