Health status monitoring of ship rotating machinery based on sliding window and long-short term generative adversarial network

Abstract

Introduction: Ship rotating machinery is the core of the ship’s power system, and its health is crucial to the safety and economy of ship operation. However, the traditional operation and maintenance model is difficult to meet the intelligent needs of modern ships.

Methods: This study proposes an integrated solution of “data preprocessing-fault diagnosis-life prediction” to finish real-time monitoring and accurate prediction of the health condition of ship rotating machinery. First, the solution uses the Mobile Grid-Based Density Clustering (MGBDC) algorithm for data preprocessing, and then integrates VMD, Autoencoder (AE), and BiGRU to establish a FD model. The model innovatively integrates SW and LSGAN to enhance the prediction precision of RUL, and finally builds a full-process intelligent monitoring system.

Results: Experiments showed that the mean fault recognition rate of the FD model reached 97.7%, which was 4.2%–6.4% higher than the comparison models. The designed algorithm selected MSE, RMSE, and MAE for remaining service life prediction as low as 0.017, 0.130, and 0.098, and the correlation coefficient reached 0.986. In practical applications, the system FD accuracy increased to 97.7%, the remaining service life prediction MAE was only 1.2 days, and the annual operation and maintenance cost of a single ship was reduced by 48.6%.

Discussion: The system provides full-process support for shipping companies and promotes the intelligent transformation of rotating machinery operation and maintenance.