Damage identification of trusses using limited modal features and ensemble learning

Abstract

A damage diagnosis method for trusses based on incomplete free vibration properties utilizing ensemble learning, e.g. Extreme gradient boosting (XGBoost), is presented in this work. Owing to the lack of measurement sensors, modal features are only measured at master degrees of freedom (DOFs) of a few first models instead of all DOFs of a structural system. Accordingly, a modal strain energy-based index (MSEBI) is employed to determine the most potentially damaged candidates. Then, an XGBoost-driven ensemble learning model is constructed from a finite element method (FEM)-simulated dataset. In which, inputs are eigenvectors corresponding to master DOFs, whilst outputs are damage ratios of suspected members. The accuracy of such a model is continuously enhanced by removing low-risk members via a damage threshold. As a consequence, the present paradigm can reliably detect damage to trusses. All test examples are programmed in Python to illustrate the reliability and efficiency of the proposed methodology.