سيده مريم فرشباف شكيب

To ensure the efficiency and continuous operation of structures throughout their useful life, the ability to detect early damages of structures is one of the main concerns of civil engineers. Various methods have been proposed for identifying the presence of damage, its location, and severity, all of which are known as structural health monitoring. In recent years, access to various types of data has increased with the advancement of sensors. Also, with the increasing advancement of computer science and data science, the use of machine learning and deep learning for better eva‎luation of this huge amount of information is increasing day by day. Machine learning is a subfield of artificial intelligence that can create algorithms using input data that can make changes to achieve the desired output without human intervention. Deep learning is a subfield of machine learning whose generated algorithms have similar performance to machine learning algorithms, but with the difference that more layers are used in deep learning, each with a different interpretation than other layers, relative to input data. In this thesis, two image captioning algorithms will be presented for identifying the type and extent of damage. Image captioning refers to the process of generating textual descriptions from an image. In fact, image captioning creates a link between machine vision and natural language processing. In the introduced algorithms, sensor outputs are converted to time-frequency domain and transformed into images, and a textual description is considered for each image that represents the type and extent of damage identified in the structure. The first algorithm has a classic encoder-decoder structure for image captioning, and in the second algorithm, transformers structure have been used, which is an innovation of this research. Transformers create a relationship between each word of the description and a part of the image that corresponds to the desired word and also reduce computational costs. Finally, two algorithms are presented that have been validated on some civil structures and eva‎luated with BLEU1 and BLEU2 coefficients. Keywords: Structural Health Monitoring, Damage Detection, Deep Learning, Transformers, Image Captioning,

پايش سلامت سازه‌ها , شناسايي آسيب , يادگيري عميق , توصيف تصوير , ترنسفورمرها

Structural Health Monitoring , , Damage Detection , Deep Learning , Image Captioning , Transformers

Maryam Farshbaf Shakib

Golamreza Ghodrati Amiri