• Transactions of the Korean Society of Mechanical Engineers A
• /
• v.28 no.12
• /
• pp.2004-2011
• /
• 2004

The applicability of the ultrasonic C-scan inspection is restricted due to the deterioration of mechanical properties of specimen during the test. Therefore, the aim of this research is applied to Eddy Current (EC) test substitute for the C-scan inspection in CFRP tube containing defects. This research is to evaluate the EC signals for the inspection of CFRP tube containing various circular hole defects (20% to 100% depth to the specimen thickness) using the unloading specimen and radial loading specimen. This study was considered the following points; 1) Analysis of EC signals for the inspection of saw-cut defect and circular hole defect, 2) The evaluation of defect depths and EC signals relationship. 3) Variation of EC signal owing to the radial load. In conclusions, the high frequency such as 300∼500 kHz made it possible to the inspection of 40% to 100% defects. Particularly, in case of 20% defect, the EC signal was not detected due to the noise of micro-crack and delamination. While the depth of the hole defects were decreasing, the difference of the phase angle between unloading specimen and radial loading specimen was gradually increasing.

• Annual Conference on Human and Language Technology
• /
• 2019.10a
• /
• pp.8-11
• /
• 2019

앙상블 기법은 여러 모델을 종합하여 최종 판단을 산출하는 기계 학습 기법으로서 딥러닝 모델의 성능 향상을 보장한다. 하지만 대부분의 기법은 앙상블만을 위한 추가적인 모델 또는 별도의 연산을 요구한다. 이에 우리는 앙상블 기법을 교차 검증 방법과 결합하여 앙상블 연산을 위한 비용을 줄이며 일반화 성능을 높이는 교차 검증 앙상블 기법을 제안한다. 본 기법의 효과를 입증하기 위해 MRPC, RTE 데이터셋과 BiLSTM, CNN, BERT 모델을 이용하여 기존 앙상블 기법보다 향상된 성능을 보인다. 추가로 교차 검증에서 비롯한 일반화 원리와 교차 검증 변수에 따른 성능 변화에 대하여 논의한다.

• Annual Conference on Human and Language Technology
• /
• 2019.10a
• /
• pp.465-469
• /
• 2019

자연어 처리를 위해서 두 문장의 의미 유사성을 분석하는 것은 아주 중요하다. 이 논문은 패러프레이즈 검출 태스크를 수행하기 위한 Paraphrase-BERT를 제안한다. 우선 구글이 제안한 사전 학습된 BERT를 그대로 이용해서 패러프레이즈 데이터 (MRPC)를 가지고 파인 튜닝하였고 추가적으로 최근에 구글에서 새로 발표한 Whole Word Masking 기술을 사용하여 사전 학습된 BERT 모델을 새롭게 파인 튜닝하였다. 그리고 마지막으로 다중 작업 학습을 수행하여 성능을 향상시켰다. 구체적으로 질의 응답 태스크와 패러프레이즈 검출 태스크를 동시에 학습하여 후자가 더 잘 수행될 수 있도록 하였다. 결과적으로 점점 더 성능이 개선되었고 (11.11%의 정확도 향상, 7.88%의 F1 점수 향상), 향후 작업으로 파인 튜닝하는 방법에 대해서 추가적으로 연구할 계획이다.

• Journal of the Korean Society for Nondestructive Testing
• /
• v.28 no.2
• /
• pp.184-190
• /
• 2008

Various ECT techniques have been applied basically to assess the integrity of steam generator tithing in nuclear power plant. Among these techniques, the bobbin probe technique is applied generally to examine the volumetric flaws such as a crack-like defect and wear which is generally occurred on steam generator tubing, and additionally MRPC probe is used to examine closely tile top of tubesheet and bending regions due to the high possibility of cracking. Dent and bulge also may be formed on tube during installation process and operation of steam generator, but the dent and bulge indications greater than specific size criteria are recorded on examination report because these indications are not considered as flaw. These indications can be easily detected with bobbin probe and approximately sized with profile bobbin probe, but the size and shape can not be accurately verified. Accordingly, in this study, the $8{\times}1$ multi-array EC probe was designed to increase the measurement accuracy of the sectional profiling EC testing of tube. As a result, we would like to propose the application of $8{\times}1$ multi-array EC probe for the measurement of size and shape of profile change on steam generator tube in OPR-1000 nuclear power plant.