Damage detection of mechanical structures using machine learning on frequency response functions

Abstract

This study proposes a damage detection method by machine learning that enables low-cost and highly accurate detection of damage location, which utilizes the changes in the frequency response function of mechanical structures. Effective machine learning model for damage detection has been developed by using the frequency response functions of finite element models of various damage conditions as training data. High accuracy and low cost have been achieved by the machine learning model. To clarify the optimal conditions for damage detection, a comparison of measurement points and several machine learning methods was investigated. In addition, the robustness of the machine learning model was investigated. The results show random forest classifier and extra trees classifier are appropriate for damage detection.