Determination of the efflorescence defects on concrete and plaster surfaces using a novel deep-learning model


Creative Commons License

Nayır S., Yılmaz Y.

Journal of Structural Engineering & Applied Mechanics (Online), vol.7, no.2, pp.123-135, 2024 (Peer-Reviewed Journal)

Abstract

Efflorescence is a common problem in concrete and masonry structures. It occurs when soluble salts in the material are dissolved by water, and then carried to the surface where the water evaporates, leaving behind the salt deposits. In recent years, the construction industry has seen the development of deep learning methods to detect defects and damages such as efflorescence during the service life of structures. These advanced techniques offer more efficient and accurate ways to identify and address issues, improving the overall quality and durability of structures. This study aims to identify efflorescence damage using an innovative deep-learning model. In this scope, 617 different images were taken showing efflorescence defects on concrete and plaster surfaces. A custom dataset was created by labeling the efflorescence regions from the images taken and this dataset was trained using the YOLOv8x model. The accuracy, recall, precision, mAP50, and mAP50-95 performance measures obtained from the YOLOv8x model are 0.962, 0.902,0.991, 0.851, and 0.6334 respectively. According to these performance metrics, efflorescence damage was considerably detected. Finally, the model was tested by taking efflorescence damage images from different concrete and plaster surfaces. The study indicated that detecting efflorescence defects on light-colored surfaces in the model was difficult, with a low detectability rate. Algorithms can be developed to detect efflorescence damage more effectively on light-colored surfaces.