The invention relates to the technical field of ship power device fault diagnosis, in particular to a marine main engine small sample fault diagnosis method based on a deep concentric twin network, which can overcome the defect of limited fault data volume of an existing ship power device and remarkably improve the fault diagnosis accuracy of a marine main engine. The DCSN adopts a sample pairing mode to expand training samples, the risk of deep model overfitting is reduced by balancing the number of positive pairs and negative pairs, meanwhile, the DCSN adopts concentric loss as an optimization function, inter-class distinguishable features can be automatically learned in end-to-end training, and the accuracy of deep model overfitting is improved. According to the method, the effectiveness of the DCSN is verified on a marine main engine fault data set, and the influence of related parameters on DCSN diagnosis performance is discussed, including values of inner and outer boundaries and twin network output feature dimensions.
本发明涉及船舶动力装置故障诊断技术领域,具体的说是一种能够克服现有船舶动力装置故障数据量有限的不足,显著提高船舶主机故障诊断准确性的基于深度同心孪生网络的船舶主机小样本故障诊断方法,DCSN采用样本配对的方式扩充训练样本,并通过平衡正对和负对的数量,来减小深度模型过拟合的风险,同时DCSN采用同心损失作为优化函数,有助于在端到端的训练中自动学习到类间可区分性特征,本发明在船舶主机故障数据集上验证了所提DCSN的有效性,并讨论了相关参数对DCSN诊断性能的影响,包括内外边界的取值、孪生网络输出特征维度。
Marine main engine small sample fault diagnosis method based on deep concentric twin network
基于深度同心孪生网络的船舶主机小样本故障诊断方法
2024-07-09
Patent
Electronic Resource
Chinese
Fault diagnosis for the Space Shuttle main engine
NTRS | 1992
|Fault diagnosis for the Space Shuttle main engine
AIAA | 1992
|Marine main engine oil supply system and marine main engine oil supply method
European Patent Office | 2024
|