http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Deep Learning-based Analysis on Physical Factors of Bar Warping in Rolling Process
Jong Hwan Lee(이종환),Eon Ho Im(임언호),Hyuck Cheol Kwon(권혁철),Jea Sook Chung(정제숙),Seungchul Lee(이승철) 대한기계학회 2021 대한기계학회 춘추학술대회 Vol.2021 No.4
The bar warping phenomenon in the rolling process negatively decreases both productivity and product quality in steel manufacturing. However, it is difficult to predict and analyze root causes of bar warping phenomenon with conventional methods such as a finite element method (FEM). Therefore, it has been a chronic problem in the rolling process of steel-making. In this paper, we propose a data-driven method to predict the bar warping phenomenon using a deep learning that takes into account physical factors such as difference in speed of the upper/lower rollers, initial thickness, initial width, and heating time, etc. In addition, LRP (Layer-wise Relevance Propagation), which is one of the explainable AI methods, is applied to numerically compute feature importance among such physical factors. We validated this approach with steel making’s actual rolling process data. As a result, we achieved high predictive performance with about 90% accuracy in predicting the bar warping events. In particular, we identified a new physical factor, width reduction, which turns to be a major effect on the bar warping phenomenon.