inis 24(2): e1

Research Article

Early State Prediction Model for Offshore Jacket Platform Structural Using EfficientNet-B0 Neural Network

Download47 downloads
  • @ARTICLE{10.4108/eetinis.v11i2.4740,
        author={Le Anh-Hoang Ho and Viet-Dung Do and Xuan-Kien Dang and Thi Duyen-Anh Pham},
        title={Early State Prediction Model for Offshore Jacket Platform Structural Using EfficientNet-B0 Neural Network},
        journal={EAI Endorsed Transactions on Industrial Networks and Intelligent Systems},
        volume={11},
        number={2},
        publisher={EAI},
        journal_a={INIS},
        year={2024},
        month={3},
        keywords={Identify damage, Offshore Jacket Platforms, Vibration assessment, Wavelet transformer, CNN, Confusion matrix},
        doi={10.4108/eetinis.v11i2.4740}
    }
    
  • Le Anh-Hoang Ho
    Viet-Dung Do
    Xuan-Kien Dang
    Thi Duyen-Anh Pham
    Year: 2024
    Early State Prediction Model for Offshore Jacket Platform Structural Using EfficientNet-B0 Neural Network
    INIS
    EAI
    DOI: 10.4108/eetinis.v11i2.4740
Le Anh-Hoang Ho1, Viet-Dung Do1, Xuan-Kien Dang1,*, Thi Duyen-Anh Pham1
  • 1: Ho Chi Minh City University of Transport
*Contact email: kien.dang@ut.edu.vn

Abstract

Offshore Jacket Platforms (OJPs) are often affected by environmental components that lead to damage, and the early detection system can help prevent serious failures, ensuring safe operations and mining conditions, and reducing maintenance costs. In this study, we proposed a prediction model based on Convolutional Neural Networks (CNNs) aimed at determining the early stage of the OJP structure’s abnormal status. Additionally, the EfficientNet-B0 Deep Neural Network classifies normal and abnormal states, which may cause problems, by using displacement signal analysis at specific areas taken into account throughout the test. Displacement data is transferred to a 2D scalogram image by applying a continuous Wavelet converter that shows the state of the work. Finally, the scalogram image data set is used as the input of the neural network, and feasibility experimental results compared with other typical neural networks such as GoogLeNet and ResNet-50 have verified the effectiveness of the approach.