فاطمه منصوري هانيس

Structural health monitoring is a significant research field in engineering, aiming to detect damage, locate it, assess its severity in civil structures and infrastructure such as high-rise buildings, bridges, dams, and more, and ultimately estimate the remaining useful life of the structure. Data-driven and machine learning-based techniques are widely used and efficient fields in structural health monitoring. Nowadays, the use of semi-supervised machine learning algorithms has become common in research, particularly in structural health monitoring, due to their ability to reduce the need for large volumes of labeled data, which is costly, time-consuming, and sometimes impossible to obtain, as well as their high prediction accuracy. In this research, a novel and efficient method is presented using a semi-supervised graph-based machine learning algorithm called Local and Global Consistency (LGC), applied for the first time in structural health monitoring, alongside a supervised algorithm called Random Forest, which is a type of ensemble learning algorithm, to detect damage in structures. The proposed method is validated by detecting damage in a numerical bridge model at the University of Florida, a laboratory model simulating a stadium grandstand at the Qatar University laboratory, and the Yonghe River cable-stayed bridge in China. The results indicate the high prediction accuracy of this method (ranging from 88% in the Qatar University laboratory model to 100% in the Yangtze bridge) in detecting damage conditions in these structures. Acceleration responses collected from the health monitoring systems installed on these structures were used to predict the presence and location of damage. First, the feature extraction process was performed in both the time and frequency domains using Fast Fourier Transform (FFT) from the raw data. In the next step, the semi-supervised machine learning algorithm LGC was trained with a limited number of labeled data and a large amount of unlabeled data to label the unlabeled data. Finally, the performance of the semi-supervised algorithm was eva‎luated by the supervised Random Forest classifier algorithm.

پايش سلامت سازه ها , تشخيص آسيب , يادگيري ماشين , يادگيري نيمه‌نظارت‌شده , يادگيري با سازگاري محلي و كلي , الگوريتم جنگل تصادفي , تبديل فوريه سريع

Structural Health Monitoring (SHM) , Damage Detection , Machine Learning (ML) , Semi-Supervised Learning (SSL) , Learning with Local and Global Consistency (LGC) , Random Forest , Fast Fourier Transform (FFT)

Fateme Mansouri Hanis

Dr. Gholamreza Ghodrati Amiri