Detecção de fissuras em micro tomografia de concreto reforçado por fibras utilizando redes neurais profundas

Abstract

Fiber-reinforced concrete is a crucial material for civil construction, and monitoring its healh is an important task for preserving structures and preventing accidents and financial losses. Among the various non-destructive concrete health monitoring methods, Computed Microtomography (Micro-CT) image analysis stands out as an inexpensive method free from noise and external interference. However, manual inspection of cracks in Micro-CT images is subjective and requires significant human effort. Deep learning algorithms have been used extensively for concrete cracks automatic detection, but only in surface regions. Detecting cracks all fiber-reinforced concrete structure, including inside in high-resolution Micro-CT volumes, remains a challenge. Therefore, this work proposes a framework for automatic crack detection explicitly for fiber-reinforced concrete Micro-CT images, combining super-resolution-based preprocessing, Detection Transformers (DETR), and committee-based post-processing to improve crack detection reliability and reduce false positives. The model was trained on a new publicly available dataset, the FIRECON dataset, which consists of 4,064 images annotated by an expert, achieving metrics of 86.098% Intersection over Union (IoU), 89.37% Precision, 83.26% Recall, 84.99% F1-Score, and 44.69% Average Precision (AP). This framework, therefore, significantly reduces analysis, human effort, and improves consistency compared to manual methods used in previous studies. The results demonstrate the potential of deep learning to aid image analysis in concrete health assessments, understanding of the cracking mechanisms of fiber-reinforced concrete, and provides a valuable tool for the development of durable, high-performance engineering materials.