The science of structural health monitoring (SHM) has received great attention in recent years and has reached a major milestone in its evolution. In this research, a new methodology for detecting damage in steel beams using machine learning will be introduced. Ease of use, high accuracy and reduction of monitoring costs have been among the presuppositions considered for presenting the new method. In this research, both the laboratory model and the analytical model have been used. At first, a 14 IPE steel beam was subjected to a dynamic load called an impact load, and the response signals were recorded in the undamaged state and under damage scenarios by acceleration mapping sensors. Then, the response signals were processed with quadratic time-frequency function and Smoothed Pseudo-Wigner-Ville function and the dynamic characteristics of the beams were extracted and prepared for machine learning training in order to detect damage. By using these data as machine learning input and determining the proportional outputs relative to the percentages of damage of each scenario, XGBOOST and multi task elastic net (MTEN) machine learning algorithms were trained. The K-fold index was used to evaluate the performance of machine learning models and algorithms, and the calculation results showed that the XGBOOST algorithm has a higher detection accuracy than the Multi-Task Elastic Net (MTEN) algorithm. In order to evaluate, validate and ensure the performance of the proposed method, an apple scenario was used to use both machine learning algorithms to predict damage. While the results showed the high detection accuracy of both methods and the lowest error rate in determining the location and severity of the damage, the high accuracy of the XGBOOST algorithm in identifying the severity and location of the damage was confirmed.
