Damage identification of steel-concrete composite beams based on modal strain energy changes through general regression neural network

摘要

This paper presents a novel method for damage identification of steel-concrete composite beams based on modal strain energy (MSE) changes through general regression neural network (GRNN). A finite element (FE) model was developed using two Euler-Bernoulli beam elements as steel beam and concrete slab layers which are coupled by incorporating a deformable shear connection layer distributed at their interface. The connection layer was modelled as a uniform spring that enables both longitudinal slip and vertical uplift between the two components. The FE model was validated using experimental results of a full-scale composite beam in the laboratory. Three damage indices were defined as the elemental stiffness reduction in the steel, concrete and shear connection layers, respectively. A damage identification approach was developed to investigate the sensitivity of eigenvalues to damage in the composite interface. Then, MSEs change ratios were selected as the features for identifying structural damage in the composite layers. Principle component analysis was utilised to reduce the dimensionality of the large data features obtained from modal analysis leading to determining the main features for structural damage identification. The low dimensional data were employed as the input for the GRNN. Different damage cases were investigated, and the damage vectors were defined as the outputs of GRNN. The results show that the proposed method is efficient and reliable to identify damage in the composite beam with a few low vibration modes, even though small damage in the composite layers does not significantly affect these modes.