Structural damage identification incorporating uncertain boundary flexibility by ensemble couple sparse coding classification method

Abstract

Transfer function (TF) represents the ratio of structure’s response to the loading and can be utilised for the identification of damage in a structure. The capability of distinguishing between changes of the TF due to damage in the existing structure and due to idealised boundary conditions of the analytical model (such as soil-foundation effects or spring bearings) is noteworthy. However, in many cases, having precise knowledge of boundary characteristics is infeasible. In this regard, a number of couple sparse coding (CSC) learnings are utilised with TFs of the structure with various probable boundary conditions as input parameters to reduce the uncertainty of idealised boundary conditions. The system of CSCs termed ensemble to enhance efficiency of the method. Principle component analysis (PCA) technique is applied to avoid over fitting problem due to the large size of TF data in machine learning training. In numerical verifications, the damage scenarios are considered different from the trained data. A 2 D frame structure is presented as numerical example to demonstrate the adequacy of the proposed method for damage detection under complicated effects of soil foundation. A beam which is suspended using soft rubber bands with uncertain stiffness is considered for experimental validation of the method. Results indicate that the proposed method is capable of identifying damage properties with acceptable precision.