역전파신경회로망을 이용한 결함탐상모델링에 관한 연구 = A Study on Defect Detection Modelling using Back propagation Neural Networks

다국어 초록 (Multilingual Abstract)

Quantitative nondestructive evaluation(QNDE) gets hold of position and shape of defects in mechanical components and estimates safety of structure. The ultrasonic defect detection out of QNDE is based on refraction of ultrasonic wave by defects in solid material. Neural network is information processing system modeling human brain and has an application to defect identification with ultrasonic defect detection. The present method consists of three process. The first process gets sample data on defect and ultrasonic wave parameters according to various defect patterns. In the second process, Neural networks is learned using sample data. In third process, neural networks predicts defect patterns by unlearned data This processes are applied to the identification of size and location of defects hidden in SM45C.

목차 (Table of Contents)

• 1. 서론
• 2. 계층형 신경 회로망
• 2.1 신경회로망의 구조
• 2.2 학습 알고리즘
• 3. 역문제 해석의 원리와 최적화 방법
• 4. 복수표면결함의 동정(idenfication)
• 4.1 동정
• 4.2 신경회로망의 학습
• 5. 실험결과 및 고찰
• 5.1 계층형 결함동정시스템
• 5.2 결함 수의 동정
• 6. 결론