Estimation of concrete strength based on deep learning-based image segmentation coupled with IR thermal extraction
- Issued Date
- 2022
- Abstract
- 본 연구에서는 임의의 시간 간격으로 콘크리트의 강도를 예측하는 방법을 제안하였다. 이 방법은 크게 두 가지 방식으로 구성된다. 첫 번째는 촬용된 모든 이미지에 대해 신경망 아키텍처를 사용하여 콘크리트 영역을 정확하게 분류하는 심화학습 기반의 이미지 분할 기술이다. 연구 대상에 가장 적합한 CNN (Convolutional Neural Network) 백본 모델을 찾기 위해 여러 환경에서 다양한 가상의 계수를 사용하여 훈련, 테스트 및 평가되었습니다. 그런 다음 이 방법을 열적 정보 추출과 함께 사용하여 콘크리트의 강도를 예측했다. 열화상 카메라를 적용하여 콘크리트 표면 온도를 감지하였다. 방법의 신뢰성은 일부 환경변수를 변경하여 사전에 조사하였다. 그런 다음 건설 현장에서 열화상 카메라를 사용하여 분할된 이미지에서 구체적인 온도 데이터를 수집하였다. 이를 성숙도(Maturity)식을 사용하여 콘크리트의 온도, 재령, 강도 간의 관계를 해석했다. 실험결과 이 방법이 콘크리트의 강도를 범용적으로 추정할 수 있음을 확인하였다. 뿐만 아니라 시간적, 경제적 측면에서도 공사의 효율성을 높이는 데 큰 도움이 될 수 있음을 알 수 있었다.|In this research, a method was proposed that is used to predict the strength of concrete at any time interval. This method is composed of two major schemes. The first one is deep learning-based concrete image segmentation technique which employs state-of-the-art neural network architectures to accurately classify the concrete area from any captured image. Segmentation models with a convolutional neural network (CNN) backbone were trained, tested and evaluated by varying different hyperparameters with in the training environment to come up with a model that best fit the target of the research. Then, this method together with thermal extraction was used to predict the strength of the concrete. The second scheme is the application of a thermal imaging (IR) camera to detect the concrete's surface temperature. The method's reliability was first investigated by varying some parameters. Then the IR camera is used on the construction site to collect concrete temperature data from the segmented image. Maturity method was then used to interpret the relationship between temperature, age, and strength of concrete.
The result shows that the method can estimate the strength of concrete at any time. In addition to this, it is also very helpful in increasing the efficiency of construction projects in regards to both time and economic standpoint.
- Alternative Title
- 심층학습 기반 화상 분할과 연동된 열화상 정보를 활용한 콘크리트의 강도예측
- Alternative Author(s)
- Minwuye Mesfin, Woldeamanuel
- Affiliation
- 조선대학교 일반대학원
- Department
- 일반대학원 건축공학과
- Advisor
- 김형기
- Awarded Date
- 2022-02
- Table Of Contents
- LIST OF FIGURES iv
LIST OF TABLES vii
ABSTRACT viii
한 글 요 약 ix
1. INTRODUCTION 1
1.1. Background 1
1.2. Objective of the research 3
1.3. Outline of the research 3
2. DEEP LEARNING: A LITERATURE REVIEW 5
2.1. Introduction 5
2.2. Deep Learning (DL) 6
2.3. Deep Learning Neural Networks 8
2.3.1. Artificial Neural Networks (ANNs) 8
2.3.2. Recurrent Neural Networks (RNNs) 9
2.3.3. Generative Adversarial Networks (GANs) 10
2.3.4. Convolutional Neural Networks (CNNs) 11
2.4. Architecture of a CNN 12
2.5. State-of-the-art CNN architectures 17
2.6. Transfer Learning 22
2.7. Semantic Segmentation 24
3. APPLICATION OF DEEP LEARNING MODELS IN CONCRETE AREA SEGMENTATION 29
3.1. Introduction 29
3.2. Experimental Methodology 29
3.2.1. Data Acquisition and Labeling 29
3.2.2. Data training 33
3.2.3. Evaluation and Validation 36
3.3. Results and Discussion 37
4. INFRARED THERMOGRAPHY (IRT) AND APPLICATION ON THE CONCRETE INDUSTRY 44
4.1. Introduction 44
4.2. Basic Principles and Features of Thermal imaging (IRT) camera 45
4.3. Investigation on the reliability of IRT in extracting temperature from concrete 50
4.3.1. Type of equipment used 51
4.3.2. Effect of distance range from ROI 55
5. CORRELATION OF DEEP LEARNING AND IRT IN PREDICTING STRENGTH DEVELOPMENT OF CONCRETE: PROPOSED MODEL 61
5.1. Introduction 61
5.2. Concrete Maturity 62
5.3. Strength-Maturity Relationships 64
5.4. Experimental Methodology 67
5.4.3. Laboratory-Based investigation 67
5.4.4. Field-Based Investigation 72
5.5. Result and Discussion 75
5.5.1. Comparison of thermocouples and thermal imaging camera 75
5.5.2. Combined effect of concrete area prediction and Strength Prediction based on maturity method 79
5.5.3. Comparison in formwork removal time 82
6. CONCLUSION 87
6.1. Conclusion 87
6.2. Recommendations 88
6.3. Future Studies 89
REFERENCES 91
PUBLICATIONS 98
ACKNOWLEDGEMENTS 99
- Degree
- Master
- Publisher
- 조선대학교 대학원
- Citation
- 울드엠마뉴엘 메스핀. (2022). Estimation of concrete strength based on deep learning-based image segmentation coupled with IR thermal extraction.
- Type
- Dissertation
- URI
- https://oak.chosun.ac.kr/handle/2020.oak/17250
http://chosun.dcollection.net/common/orgView/200000590180
- Appears in Collections:
- General Graduate School > 3. Theses(Master)
- Files in This Item:
-
-
Download
Estimation of concrete strength based on deep learning-based image segmentation coupled with IR ther.pdf
기타 데이터 / 9.87 MB / Adobe PDF
-
Items in Repository are protected by copyright, with all rights reserved, unless otherwise indicated.