Lightweighting CNN Models via Dual-Phase Pruning for Motor Fault Diagnosis in Small and Medium-Sized Vessels 


Vol. 40,  No. 4, pp. 13-24, Aug.  2025
10.14341/JKOSOS.2025.40.4.13


PDF
  Abstract

In Korea's maritime industry, many marine accidents occur due to engine failures. In particular, pump-motor system failures significantly contribute to vessel engine damage. Although deep-learning-based fault diagnosis models have been extensively studied, obtaining various training datasets remains highly challenging. While transfer learning models mitigate this limitation, their size makes them impractical for deployment on low-spec hardware in small- and medium-sized vessels. Consequently, model compression is essential for their practical application. The proposed approach in this study employs a dual-phase pruning (DPP) technique that combines unstructured pruning based on first-order Taylor expansion, with structured pruning to optimize model size. This hybrid approach significantly minimizes model size while preserving diagnostic performance. The proposed method could achieve a substantial reduction in model size— from 158.5 MB to 7.89 MB—and in FLOPs—from 48.42 MFLOPs to 4.3 MFLOPs—without significant performance degradation. Overall, the proposed dual-phase pruning technique offers a lightweight and efficient fault diagnosis solution that can achieve substantial model compression without compromising performance, making it well-suited for low-spec hardware in small and medium-sized vessels.

  Statistics
Cumulative Counts from November, 2022
Multiple requests among the same browser session are counted as one view. If you mouse over a chart, the values of data points will be shown.


  Cite this article

[IEEE Style]

이진열, 김하람, 박교식, 최종선, "Lightweighting CNN Models via Dual-Phase Pruning for Motor Fault Diagnosis in Small and Medium-Sized Vessels," Journal of the Korean Society of Safety, vol. 40, no. 4, pp. 13-24, 2025. DOI: 10.14341/JKOSOS.2025.40.4.13.

[ACM Style]

이진열, 김하람, 박교식, and 최종선. 2025. Lightweighting CNN Models via Dual-Phase Pruning for Motor Fault Diagnosis in Small and Medium-Sized Vessels. Journal of the Korean Society of Safety, 40, 4, (2025), 13-24. DOI: 10.14341/JKOSOS.2025.40.4.13.

Search
(IN TITLE, AUTHOR, ABSTRACT,KEYWORDS)

Advanced Search ▼

POPULAR KEYWORDS
(TOP 10 KEYWORDS)

Recent Publications
(LAST 3 YEARS)

pISSN : 1229 - 9197
eISSN : 1875 - 0052

Impact Factor(2022): 2.5