인공신경망 이론을 이용한 2단계 탐지 기법에 의한 3층 철근콘크리트 골조 구조물의 손상탐지

Abstract

최근 들어 인공신경망(Artificial Neural Network)을 이용한 손상탐지 연구는 해석모델이나 축소모형, 부재단위의 실험체를 이용한 연구가 주로 이루어져 왔다. 일반적인 철근콘크리트 골조 구조물에 대한 손상탐지 연구는 한정적으로 수행되어 왔고, 철근콘크리트 골조 구조물의 손상위치와 정도를 정확히 예측하기 위해서는 추가적인 손상탐지 기법에 대한 연구가 필요하다. 본 연구는 철근콘크리트 골조 구조물의 손상탐지 정확성을 높일 수 있는 방법을 제안하고자 한다. 실험구조물의 동특성 데이터를 얻기 위하여 진동실험을 실시하였고, 진동실험을 통하여 얻은 실험구조물의 응답은 구조물식별기법(Structural System Identification)을 통하여 구조물의 동특성을 구하였다. 인공신경망 학습데이터를 생성하기 위해서 유한요소해석프로그램을 사용하였고 실험구조물의 동특성과 가장 유사한 기본해석모델을 만든 후 이 기본해석모델을 이용하여 손상과 동특성이 쌍을 이루는 200개의 학습데이터를 생성하였다. 기존 인공신경망을 이용한 손상탐지의 정확도를 개선하고자 학습데이터를 분석하였다. 효과적인 손상탐지를 위하여 학습데이터를 가공하였고 동특성별 특징을 고려하여 단계별로 제한된 학습데이터를 사용하여 손상탐지를 실시하였고 정확성을 검증하였다.

본 논문은 총 4장으로 구성되며 그 내용은 다음과 같다.

1장에서는 연구배경과 기존연구, 연구범위에 관해 설명했다.

2장에서는 실험에 사용된 구조물과 진동실험에 대하여 소개하였고, 진동실험을 통하여 구한 실험구조물의 응답을 구조물식별기법을 사용하여 구조물의 동특성을 추출하였다.

3장에서는 인공신경망의 학습과정을 설명하고, 유한요소모델해석 결과를 이용하여 인공신경망 학습에 필요한 학습데이터 작성 방법을 소개하였다. 학습데이터 가공방법을 설명한 후 기존 방법에 따라서 실험 구조물의 손상탐지를 실시하였다. 손상탐지 기법의 정확도를 개선하기위해 기존 학습데이터 가공방법을 분석하고 새로운 데이터 가공법 및 단계별 손상탐지를 제안하였다. 제안된 방법을 사용하여 동일한 구조물에 손상탐지를 실시하였고 손상탐지 결과 기존 손상탐지 방법보다 높은 정확도를 보였다.

4장에서는 제안된 방법의 결과를 바탕으로 결론을 나타냈다.|A recent study on the damage detection by the ANN(Artificial Neural Network) have been targeted at analytical model, scaled model or element-level. The damage detection of a reinforced concrete structure has been limitedly studied, and in order to evaluate the location and quantification of the severity of the damage for a reinforced concrete structure precisely, additional research on damage detection method is required. In this study, a method that enhance the accuracy of the damage detection was proposed. The acceleration responses of the reinforced concrete structure are acquired by vibration test. And using SSI(Structural System Identification), the dynamic properties of the structure are calculated. After constructing baseline model which is the most similar with the dynamic properties of the real structure by finite element analysis program, SAP2000, the training data are generated by the baseline model. In this study, in order to improve the existing damage detection by ANN, training data are analyzed, and effectively processed. Using processed training data, staged damage detection was performed, better results than existing method for damage detection was induced. This paper consists of four chapters and its contents are as follows.

In chapter 1, background, the existing research regarding and research scope are explained. In chapter 2, structure used in experiment and vibration test were introduced. And modal properties of the structure were extracted by structure's acceleration response which was applied by SSI.

In chapter 3, training procedure of ANN was explained. By using results of the finite element modeling analysis, processing procedure of training data, which were need to training ANN, was illustrated. Processing method of training data was explained, and then damage detection of the test structure was carried on by existing methods. In order to improve the accuracy of the damage detection, existing processed data were analyzed and both new data processing method and staged damage detection were proposed. Using proposed method, damage detection of the structure was conducted, and the proposed damage detection method was more precise than the existing damage detection method.

In chapter 4, the conclusions are presented based on the proposed method and results of the validation.; A recent study on the damage detection by the ANN(Artificial Neural Network) have been targeted at analytical model, scaled model or element-level. The damage detection of a reinforced concrete structure has been limitedly studied, and in order to evaluate the location and quantification of the severity of the damage for a reinforced concrete structure precisely, additional research on damage detection method is required. In this study, a method that enhance the accuracy of the damage detection was proposed. The acceleration responses of the reinforced concrete structure are acquired by vibration test. And using SSI(Structural System Identification), the dynamic properties of the structure are calculated. After constructing baseline model which is the most similar with the dynamic properties of the real structure by finite element analysis program, SAP2000, the training data are generated by the baseline model. In this study, in order to improve the existing damage detection by ANN, training data are analyzed, and effectively processed. Using processed training data, staged damage detection was performed, better results than existing method for damage detection was induced. This paper consists of four chapters and its contents are as follows.

In chapter 1, background, the existing research regarding and research scope are explained. In chapter 2, structure used in experiment and vibration test were introduced. And modal properties of the structure were extracted by structure's acceleration response which was applied by SSI.

In chapter 3, training procedure of ANN was explained. By using results of the finite element modeling analysis, processing procedure of training data, which were need to training ANN, was illustrated. Processing method of training data was explained, and then damage detection of the test structure was carried on by existing methods. In order to improve the accuracy of the damage detection, existing processed data were analyzed and both new data processing method and staged damage detection were proposed. Using proposed method, damage detection of the structure was conducted, and the proposed damage detection method was more precise than the existing damage detection method.

In chapter 4, the conclusions are presented based on the proposed method and results of the validation.