DOI QR코드

DOI QR Code

Predicting Dynamic Response of a Railway Bridge Using Transfer-Learning Technique

전이학습 기법을 이용한 철도교량의 동적응답 예측

  • Minsu Kim (School of Railroad Engineering, Korea National University of Transportation) ;
  • Sanghyun Choi (School of Railroad Engineering, Korea National University of Transportation)
  • 김민수 (한국교통대학교 철도공학부) ;
  • 최상현 (한국교통대학교 철도공학부 )
  • Received : 2022.12.19
  • Accepted : 2022.12.27
  • Published : 2023.02.28

Abstract

Because a railway bridge is designed over a long period of time and covers a large site, it involves various environmental factors and uncertainties. For this reason, design changes often occur, even if the design was thoroughly reviewed in the initial design stage. In particular, design changes of large-scale facilities, such as railway bridges, consume significant time and cost, and it is extremely inefficient to repeat all the procedures each time. In this study, a technique that can improve the efficiency of learning after design change was developed by utilizing the learning result before design change through transfer learning among deep-learning algorithms. For analysis, scenarios were created, and a database was built using a previously developed railway bridge deep-learning-based prediction system. The proposed method results in similar accuracy when learning only 1000 data points in the new domain compared with the 8000 data points used for learning in the old domain before the design change. Moreover, it was confirmed that it has a faster convergence speed.

철도교량의 설계는 장기간에 걸쳐 수행되고 대규모의 부지를 대상으로 하기 때문에 다양한 환경적인 요인과 불확실성을 동반하게 된다. 이러한 연유로 초기 설계단계에서 충분히 검토하였더라도 설계변경이 종종 발생하고 있다. 특히 철도교량과 같은 대규모 시설물의 설계변경은 많은 시간과 인력을 소모하며, 매번 모든 절차를 반복하는 것은 매우 비효율적이다. 본 연구에서는 딥러닝 알고리즘 중 전이학습을 통해 설계변경 전의 학습 결과를 활용하여 설계변경 후의 학습의 효율성을 향상시킬 수 있는 기법을 제안하였다. 분석을 위해 기개발한 철도교량 딥러닝 기반 예측 시스템을 활용하여 시나리오들을 작성하고 데이터베이스를 구축하였다. 제안된 기법은 설계변경 전 기존 도메인에서 학습에 사용한 8,000개의 학습데이터 대비 새로운 도메인에서 1,000개의 데이터만을 학습하여 유사한 정확도를 나타내었고 보다 빠른 수렴속도를 가지는 것을 확인하였다.

Keywords

Acknowledgement

본 연구는 국토교통과학기술진흥원 연구사업의 연구비 지원(과제번호21CTAP-C164360-01)에 의해 수행되었습니다.

References

  1. Chung, S.S., Lee, C.H., Kim, S.B. (2020) CORAL Transfer Learning for Deep Learning-Based Virtual Metrology Modeling, J. Korean Inst. Ind. Eng., 46(3), pp.319~325.
  2. Kim, D.A., Kim, S.T., Kim, K.S., Kim, H.R. (2018) Development of Conceptual Cost Estimating Model for Railway Bridges according to Design Changes - Focused on PSC BEAM, J. Korean Soc. Railw., 21(10), pp.1015~1021. https://doi.org/10.7782/JKSR.2018.21.10.1015
  3. Kim, M., Choi, S. (2022) Running Safety and Ride Comfort Prediction for a Highspeed Railway Bridge using Deep Learning, J. Comput. Struct. Eng. Inst. Korea, 35(6), pp. 375~380. https://doi.org/10.7734/COSEIK.2022.35.6.375
  4. Kim, S.I., Kwak, J.W. (2012) Traffic Safety and Passenger Comforts of Railway Bridges, Mag. & J. Korean Soc. Steel Constr., 24(3), pp.39~46.
  5. Kim, Y.M., Shin, S.J., Cho, H.W. (2020) Predictive Modeling for Machining Power Using Transfer Learning, J. Korean Inst. Ind. Eng., 46(2), pp.94~106.
  6. Korea Rail Network Authority (KRNA) (2014) Railway Design Guidlines and Handbook (Running safety and Ride Comfort review), KR C-08070.
  7. Weiss, K., Khoshgoftaar, T.M., Wang, D. (2016) A Survey of Transfer Learning, J. Big Data, 3(1), pp.1~40. https://doi.org/10.1186/s40537-015-0036-x
  8. Yosinski, J., Clune, J., Bengio, Y., Lipson, H. (2014) How Transferable are Features in Deep Neural Networks?, In Proceedings of the 27th International Conference on Neural Information Processing Systems, 2, pp.3320~3328.
  9. Zhuang, F., Qi, Z., Duan, K., Xi, D., Zhu, Y., Zhu, H., Xiong, H., He, Q. (2020) A Comprehensive Survey on Transfer Learning, Proceedings of the IEEE, 109(1), pp.43~76.