387 0

딥러닝을 활용한 콘크리트 강도 예측 및 배합 산정에 관한 연구

Title
딥러닝을 활용한 콘크리트 강도 예측 및 배합 산정에 관한 연구
Other Titles
A study on the Prediction for Compressive Strength and Mixing Proportions of Concrete Using Deep Learning
Author
최주희
Advisor(s)
이한승
Issue Date
2023. 2
Publisher
한양대학교
Degree
Master
Abstract
Concrete mix design is conducted to calculate the ratio of concrete mixing proportions. This process is conducted to satisfy the requried performance of compressive strength to for constructability and durability fo the structure. The existing concrete mix design process accompanies confirmation work throgh repetitive test, but this process has the disadvantage of being repetitive and time consuming. Recently, in order to reduce the experimental process of the mix design process, research is being conducted to calculate the concrete mixing proportions through artificial intelligence technology such as deep learning based on the publicly available concrete mixture data. However, compared to researches that predict concrete strength or chloride ion diffusion of concrete through artificial intelligence, researches on prediction of mixing proportions are still lacking. The applications of deep learning technology for automatic extraction and judgement in order to extract and apply information to various data in the industrial field are in the limelight. On the other hand, in the case of reseraches incorporating existing deep learning technology, they showed problems with lower performance compared to machine learning models. Therefore, in this study, DNN model that can comprehensively predict the compressive strength and mixing proportions of concrete was designed to supporting concrete mix design. The performance of the model for learning and prediction was evaluated using three performance evaluation methods to ensure reliability of model. In order to confirm the elementary performance of DNN model, the prediction performance for compressive strength was compared with 4 machine learning models. DNN model overall satisfied higher prediction performance compared to machine learning model. DNN model was separated according to the mixing of admixtures to OPC and ternary concrete. Thus, DNN model was modified and tuned to improve the overall performance of DNN model, by applying the performance improvement techniques. Prediction performance for concrete compressive strength and mixing proportions was evaluated. In the case of the compressive strength, the input value and the predicted value were similar, and in the case of the mixing proportions, the error was different depending on the mixing of the admixture. The prediction performance of each mixing proportion was evaluated for the ternary concrete DNN model. The binder factor, which showed a relatively high error compared to other factors, has a wide variance on the data set. Therefore, it is required to collect approporiate data for learning DNN model. As the model became more complex, the prediction performance of DNN model was higher, on the other hand, the time consumption for learning also increased. Accordingly, model structure has to be optimized by considering the learning time and performance. It will be possible to improve the performance of DNN model through extensive data collection. In addition, Appropriate data refienement and in-depth analysis of data should be accompanied.|콘크리트 배합설계는 시공성, 내구성과 같은 구조물의 성능 확보를 목적으로 콘크리트의 압축강도 등의 요구성능을 만족하는 콘크리트 사용재료의 비율을 산정하기 위하여 시행된다. 기존의 콘크리트 배합설계 과정은 반복적인 시험배합을 통한 확인 작업을 동반하나, 이러한 배합설계 과정은 반복적이고 시간 소모적이라는 단점을 가진다. 이에 최근에는 배합설계 과정의 실험적인 과정을 축소하기 위해 공개된 콘크리트 배합 데이터를 기반으로 딥러닝 등의 인공지능 기술을 통해 콘크리트 배합을 산정하기 위한 연구가 진행되고 있다. 반면 인공지능 기술을 통해 콘크리트 강도나 염소이온 확산계수를 예측하는 연구에 비하여 콘크리트 소요압축강도를 만족시키기 위한 배합 인자 및 사용 재료 예측에 관한 연구는 여전히 부족한 실정이다. 또한 최근 산업 분야에서 다양한 데이터에 대해 정보를 추출하고 이를 적용하기 위하여 자동적인 추출 및 판단을 위한 딥러닝 기술의 접목이 주목받고 있는 반면, 배합 요소 예측을 위해 인공지능 기술 중 딥러닝 기술을 접목하는 연구는 현저히 적은 실정이다. 특히 기존의 딥러닝 기술을 접목한 연구들의 경우 머신러닝 모델에 비하여 낮은 성능을 보이는 문제점을 나타내었다. 따라서 본 연구에서는 인공지능 기술 중 딥러닝 기술을 접목하여 콘크리트의 압축강도와 배합 인자의 동시적인 예측을 통해 콘크리트 배합 산정이 가능한 DNN 모델을 제시하고자 한다. 개발한 DNN 모델과 머신러닝 모델과의 압축강도 예측에 대한 성능 비교를 통한 모델의 개선을 수행하여 머신러닝에 상응하는 성능을 확보하도록 하였으며, DNN 모델의 은닉층 수 및 은닉뉴런 수를 다르게 하여 모델 구조에 따른 압축강도 및 배합 요소에 대한 학습 및 예측 성능을 평가하여 DNN 모델의 신뢰도 확보 및 가장 높은 성능을 확보하기 위한 방법론을 제시하고자 하였다.
URI
http://hanyang.dcollection.net/common/orgView/200000654110https://repository.hanyang.ac.kr/handle/20.500.11754/180139
Appears in Collections:
GRADUATE SCHOOL[S](대학원) > DEPARTMENT OF SMART CITY ENGINEERING(스마트시티공학과) > Theses (Master)
Files in This Item:
There are no files associated with this item.
Export
RIS (EndNote)
XLS (Excel)
XML


qrcode

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

BROWSE