Scholarlink

This study presents an innovative approach leveraging deep learning techniques, specifically the latest iteration of the YOLO (You Only Look Once) algorithm, YOLOv8, to identify and predict the cracking locations within Ultra-High-Performance Fiber-Reinforced Concrete (UHPFRC). Recognizing the challenge posed by the unpredictable distribution of fibers in UHPFRC during the casting process, which significantly affects the material’s mechanical performance and durability, this research aims to enhance the predictability and reliability of UHPFRC structures in civil engineering applications. Utilizing a dataset of CT scan images marked with defective fiber distributions, a model was trained to detect the defective fiber distributions within UHPFRC specimens. By statistically analyzing the detection results of defective fiber distributions, Prediction Success Rate, μ, was introduced to evaluate the prediction results. The calculated results show that the μ values are all greater than 0.64, which means that the method accurately predicts the cracking locations in untested UHPFRC specimens before four-point bending tests. This outcome confirms the effectiveness of the proposed method and opens new avenues for the application of AI in predictive maintenance and the design of more resilient infrastructural components.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Utilization of One-Stage Detection Algorithm to Predict UHPFRC Cracking Locations Through Fiber Distribution Analysis

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Xin Luo,
  • Takashi Matsumoto

摘要

This study presents an innovative approach leveraging deep learning techniques, specifically the latest iteration of the YOLO (You Only Look Once) algorithm, YOLOv8, to identify and predict the cracking locations within Ultra-High-Performance Fiber-Reinforced Concrete (UHPFRC). Recognizing the challenge posed by the unpredictable distribution of fibers in UHPFRC during the casting process, which significantly affects the material’s mechanical performance and durability, this research aims to enhance the predictability and reliability of UHPFRC structures in civil engineering applications. Utilizing a dataset of CT scan images marked with defective fiber distributions, a model was trained to detect the defective fiber distributions within UHPFRC specimens. By statistically analyzing the detection results of defective fiber distributions, Prediction Success Rate, μ, was introduced to evaluate the prediction results. The calculated results show that the μ values are all greater than 0.64, which means that the method accurately predicts the cracking locations in untested UHPFRC specimens before four-point bending tests. This outcome confirms the effectiveness of the proposed method and opens new avenues for the application of AI in predictive maintenance and the design of more resilient infrastructural components.