Image-based Monitoring of Bolted Connections in Steel Structures via Deep Leaning and Hough Transform
- Abstract
- The goal of this study is to design a framework for monitoring bolted connections in steel structures based on the deep learning and Hough line transform (HLT) techniques. To achieve the goal, firstly, a two-phase framework based on Region-based Convolutional Neural Networks (RCNN) and HLT techniques is proposed to detect steel bolts, which are clean and rusted bolts, and loosened bolts in bolted connections. Secondly, a modified framework using synthetic data and the Mask RCNN technique for training a bolt detector is newly designed to improve the detectability of image-based bolt monitoring and to reduce computation cost. Thirdly, a series of test scenarios on a lab-scale bolted connection of H-girder are conducted to evaluate the feasibility of the framework. Identification of bolts and their angles are carried out under environmental effects on image quality, including perspective distortions, capture distances, and light intensity. Lastly, the evaluation of the bolt detector based on synthetic data and Mask RCNN is also conducted on the new lab-scale bolted connection. Segmentation of clean and rusted bolts and rotation angles are also examined for various environmental effects. The results show that the bolt detection framework is promising for the monitoring health condition of the bolted connection in steel structures.
- Issued Date
- 2021
- Awarded Date
- 2021. 2
- Type
- Dissertation
- Publisher
- 부경대학교
- Affiliation
- 부경대학교 대학원
- Department
- 대학원 해양공학과
- Advisor
- KimJeongTae
- Table Of Contents
- 1. INTRODUCTION 1
1.1. Background 1
1.2. Objective and Scope 3
1.3. Organization of the thesis 3
2. FRAMEWORK OF DAMAGED BOLT MONITORING VIA DEEP LEARNING AND HOUGH TRANSFORM 4
2.1. Introduction 4
2.2. Design of Bolt Monitoring Framework 4
2.2.1. Overall of Bolt Monitoring Techniques 4
2.2.2. Framework of RCNN-based Bolt Detection and HLT-based Bolt Angle Estimation 6
2.2.3. Framework of Mask RCNN-based Bolt Segmentation and HLT-based Bolt Angle Estimation 8
2.3.Method of RCNN-based Bolt Detection 10
2.3.1. Training Data Generation 10
2.3.2. Training a RCNN-based Deep Learning Model 11
2.4. Method of Mask RCNN-based Bolt Segmentation 15
2.4.1. Synthetic Data Generation 15
2.4.2. Training a Synthetic Image-based Bolt Detector 19
2.5. Method of HLT-based Bolt-loosening Monitoring 22
2.5.1. Homographic Transformation 22
2.5.2. Canny Edge Detector 23
2.5.3. HLT-based Bolt Angle Monitoring 26
2.5.4. Damage Classification based on Upper Control Limit (UCL) 28
3. DAMAGED BOLT ASSESSMENT VIA RCNN AND HOUGH TRANSFORM TECHNIQUE 29
3.1. Introduction 29
3.2. Experimental Setup of Steel Girder Connection for Damaged Bolt Assessment 29
3.2.1. Test Set-up 29
3.2.2. Test Scenarios 30
3.3. Feasibility of Proposed Framework for Damage Assessment 34
3.3.1. RCNN-based Bolt Identification 34
3.3.2. Hough Line Transform-based Bolt Angle Monitoring 37
3.4 Summary and Results 41
4. DAMAGED BOLT ASSESSMENT USING MASK RCNN AND HOUGH LINE TRANSFORM TECHNIQUES 43
4.1. Introduction 43
4.2. Experimental Setup of Steel Girder Connection for Damaged Bolt Assessment 43
4.2.1. Test Set-up 43
4.2.2. Test Scenarios 44
4.3. Feasibility of Proposed Framework for Damage Assessment 45
4.3.1. Mask RCNN-based Bolt Segmentation 45
4.3.2. Hough Line Transform-based Bolt Angle Monitoring 49
4.4. Summary and Discussion 52
5. SUMMARY AND CONCLUSION 53
REFERENCES 55
ACKNOWLEDGMENTS 58
- Degree
- Master
- Files in This Item:
-
-
Download
Image-based Monitoring of Bolted Connections in Steel Structures via Deep Leaning and Hough Transfo.pdf
기타 데이터 / 4.46 MB / Adobe PDF
-
Items in Repository are protected by copyright, with all rights reserved, unless otherwise indicated.