A Data-Driven Approach for Normalizing Environmental and Operational Effects Using Unsupervised and Deep Learning Methods

Bridges are critical components of transportation networks and play a key role in supporting economic and social development; however, they are continuously subjected to various forms of deterioration. Failure to detect damage in a timely manner and to implement appropriate maintenance measures can lead to the accumulation of structural defects, which may progress gradually or occur suddenly depending on their nature. Numerous monitoring and diagnostic techniques have been developed to enable rapid and reliable damage detection, most of which rely on analyzing the structure’s dynamic responses. Nevertheless, structural damage is not the only factor influencing these responses—variations in environmental and operational conditions can also significantly affect the measured data. In recent years, unsupervised machine learning algorithms have been increasingly employed to address these challenges in structural health monitoring. This paper reviews the most widely used anomaly detection methods within the category of unsupervised learning, examining their underlying assumptions and limitations, particularly in isolating and removing environmental effects from damage-sensitive features. Finally, recent advances aimed at overcoming these limitations are discussed, highlighting the growing transition of data-driven damage detection methods from academic research to industrial applications and field implementations.

کلیدواژه ها:

structural health monitoring ، Damage detection ، Unsupervised Machine Learning ، Anomaly Detection ، Environmental Variations