Machining learning-based damage diagnosis in beam-like structures using the spectral correlation

Abstract

The field of structural damage assessment within the operational phase of structures continues to evolve, presenting a captivating challenge for construction engineering. This challenge is driven by the practical need for real-time insights into the current condition of structures. Output-based structural damage identification methods have emerged as a result of this challenge. This approach holds immense promise as it circumvents the necessity of precise excitation source information collection, making it highly adaptable to real structures. This study introduces an innovative approach to damage assessment, employing supervised Machine Learning techniques. It leverages the correlation of spectral signals as input features for artificial neural networks (ANN) and decision trees. The machine learning algorithms are designed to provide crucial insights, including the detection of new damage, assessment of damage severity, and pinpointing the precise location of structural degradation. In order to validate the effectiveness of this methodology in the realm of engineering structures, an experimental model was constructed using a supported beam model. Two distinct machine learning algorithms were applied to ascertain the relevance and viability of the proposed feature. In conclusion, this study establishes a standardized approach for solving the problem of damage identification based on spectral correlation in structural engineering.