针对船厂物流工程车辆柴油机故障特征提取单一和常规算法健康评估准确率不高的问题，提出一种基于改进图神经网络（Graph Convolution Network，GNN）信息融合采样归纳式算法（Multi－Graph SAmple and aggreGatE，Multi－Graph SAGE）。首先采用样本概率相似性原理，提取柴油机振动信号、滑油特征和瞬时转速与柴油机运行状态建立联系；接着采用特征工程筛选重要指标，通过信息融合图采样式学习深入挖掘潜在信息，构建融合邻接矩阵，同时引用JK-net跳跃知识网络，避免数据过载；最后利用交叉熵损失函数使模型准确性得到了进一步的验证。结果表明所提算法能够将特征进行深入融合，通过柴油机的健康状态识别因子HI得到监测状态。模型监测准确率达到98.54%，多项评价指标均高于99.6%。与随机森林 （Random Forest，RF）、极限学习机（Extreme Learning Machine，ELM）、卷积神经网络（Convolutional Neural Networks，CNN）等常规数据驱动相比较，本方法能够有效提高船厂物流工程车辆柴油机健康状态监测的准确率，具有一定的工程应用价值。
Aiming at the problem that the accuracy of single and conventional algorithms for extracting fault features of diesel engines in shipyard logistics engineering vehicles is not high, a multi-graph SAmple and aggreGatE (Multi-Graph SAGE) algorithm based on improved graph convolution network (GNN) is proposed. Firstly, the sample probabilistic similarity principle is used to extract the vibration signal, lubricating oil characteristics and instantaneous speed of the diesel engine to establish a connection with the operating state of the diesel engine. Then, feature engineering is used to screen important indicators, and the potential information is deeply mined through information fusion graph sampling learning, and the fusion adjacency matrix is constructed, and the JK-net jump knowledge network is referenced to avoid data overload. Finally, the accuracy of the model is further verified by using the cross-entropy loss function. The results show that the proposed algorithm can deeply fuse the features, and the monitoring state can be obtained by the health status recognition factor HI of the diesel engine. The monitoring accuracy of the model reached 98.54%, and many evaluation indicators were higher than 99.6%. Compared with conventional data-driven methods such as Random Forest (RF), Extreme Learning Machine (ELM), and Convolutional Neural Networks (CNN), this method can effectively improve the accuracy of monitoring the health status of diesel engines in shipyard logistics construction vehicles, and has certain engineering application value.
2026,48(4): 89-96 收稿日期：2025-5-26
DOI：10.3404/j.issn.1672-7649.2026.04.014
分类号：U664
基金项目：科技部专项基金资助项目（2022YFB2602303）；交通运输部专项基金资助项目（2021-ZD1-023）；中国远洋海运集团科研项目（2023-2-Z002-07）；开放基金（2022KF0019）
作者简介：马川(1979-),男,工程师,研究方向为机械制冷与低温技术
参考文献：
[1] SU Y L, GAN H B, JI Z G. Research on multi-parameter fault early warning for marine diesel engine based on PCA-CNN-BiLSTM[J]. Journal of Marine Science and Engineering, 2024, 12(6): 965.
[2] KAMINSKI P. Experimental investigation into the effects of fuel dilution on the change in chemical properties of lubricating oil used in fuel injection pump of Pielstick PA4 V185 marine diesel engine[J]. Lubricants, 2022, 10(7): 162.
[3] 宋业栋, 马光伟, 裴国斌, 等. 基于深度迁移学习的柴油机故障诊断研究[J]. 振动与冲击, 2023, 42(21): 219-226.
SONG Y D, MA G W, PEI G B, et al. Research on diesel engine fault diagnosis based on deep transfer learning[J]. Vibration and Shock, 2023, 42(21): 219-226.
[4] 毕凤荣, 郭明智, 毕晓阳, 等. 数据不平衡情况下的柴油机故障诊断方法[J]. 天津大学学报(自然科学与工程技术版), 2024, 57(8): 810-820.
BI F R, GUO M Z, BI X Y, et al. Diesel engine fault diagnosis method under the condition of data imbalance[J]. Journal of Tianjin University(Natural Science and Engineering Technology Edition), 2024, 57(8): 810-820.
[5] 赵军, 徐孙庆. 基于信息融合的舰船装备技术状态评估[J]. 舰船科学技术, 2022, 44(8): 178-181.
ZHAO J, XU S Q. Technical condition assessment of ship equipment based on information fusion[J]. Ship Science and Technology, 2022, 44(8): 178-181.
[6] 贾广付. 船舶柴油机故障实例统计与FMECA分析[J]. 河南科技, 2020(8): 52-54.
JIA G F. Statistics and FMECA analysis of marine diesel engine failure instances[J]. Henan Science and Technology, 2020(8): 52-54.
[7] 邱亚兰, 王建林. 基于多传感器感知的船舶柴油机热力参数监测研究[J]. 舰船科学技术, 2025, 47(6): 106-109.
QIU Y L, WANG J L. Research on thermal parameter monitoring of ship diesel engine based on multi-sensor sensing[J]. Ship Science and Technology, 2025, 47(6): 106-109.
[8] 吴宇飞. 基于多源信息融合的往复压缩机智能诊断技术研究[D]. 沈阳: 东北大学, 2016.
[9] KONSTANTINOS M T, GERASIMOS T, CHAITANYA P, et al. Health assessment framework of marine engines enabled by digital twins[J]. International Journal of Engine Research, 2023, 24(7): 3264-3281.
[10] LI K, WANG H, GONG Q, et al. Iterative generation differential decomposition with multi-component vibration signals for marine diesel engine[J]. Journal of Navigation. 2024, 77(4): 526-542.
[11] 黄英, 李准, 王健, 等. 基于柴油机曲轴瞬时转速信号EEMD分解的失火故障诊断[J]. 北京理工大学学报, 2025, 45(4): 384-390.
HUANG Y, LI Z, WANG J, et al. Misfire fault diagnosis based on EEMD decomposition of instantaneous speed signal of diesel engine crankshaft[J]. Journal of Beijing Institute of Technology, 2025, 45(4): 384-390.
