Chinese Journal of Ship Research, Volume. 17, Issue 5, 289(2022)

Fault diagnosis of marine diesel engines based on graph convolutional network under unbalanced datasets

Ruihan WANG1,2, Hui CHEN1,2, Cong GUAN1,2, and Mengzhuo HUANG1,2
Author Affiliations
  • 1Key Laboratory of High Performance Ship Technology of Ministry of Education,Wuhan University of Technology, Wuhan 430063, China
  • 2School of Naval Architecture,Ocean and Engery Power Engineering, Wuhan University of Technology, Wuhan 430063, China
  • show less
    Figures & Tables(15)
    [in Chinese]
    [in Chinese]
    [in Chinese]
    [in Chinese]
    [in Chinese]
    [in Chinese]
    [in Chinese]
    • Table 1. The technical parameters of 7K98MC two-stroke marine diesel engine

      View table
      View in Article

      Table 1. The technical parameters of 7K98MC two-stroke marine diesel engine

      技术指标数值
      缸径/mm980
      行程/mm2 660
      活塞面积/m20.754 3
      整机重量/t2 100
      最大功率/kW40 055
      柴油机最大额定转速/(r·min−1)94
      气缸最大平均指示压力/bar18.2
      气缸最高爆发压力/bar140.1
      涡轮增压器3×TPL85-B11
      发火顺序1−7−2−5−4−3−6
    • Table 2. Simulation dataset(case-1)

      View table
      View in Article

      Table 2. Simulation dataset(case-1)

      柴油机工况特征个数样本数据总量训练样本测试样本
      正常工况(G1)151 000700300
      压缩机故障(G2)1530030270
      空冷机故障(G3)1520020180
      喷油定时错误(G4)1520020180
    • Table 3. Precision rates and recall rates of different classifiers(case-1)

      View table
      View in Article

      Table 3. Precision rates and recall rates of different classifiers(case-1)

      分类器精确率召回率
      G1G2G3G4G1G2G3G4
      SVM0.980.770.770.460.970.810.710.33
      BPNN0.990.760.830.500.960.810.750.35
      RF10.770.870.5110.830.780.55
      SMOTE-SVM10.780.790.7810.770.780.73
      SMOTE-BPNN10.830.880.7710.880.830.75
      SMOTE-RF10.850.880.8210.870.810.81
      本文GCN10.9211110.961
    • Table 4. Precision rates of different classifiers(case-2)

      View table
      View in Article

      Table 4. Precision rates of different classifiers(case-2)

      柴油机工况训练集样本数据量
      单缸失火(G1)28
      喷油时间异常(G2)35
      针阀磨损(G3)87
      活塞环磨损(G4)102
      排气阀漏气(G5)89
      空气滤清堵塞(G6)87
      正常工况(G7)1 848
    • Table 5. Statistical parameters of vibration signals in time and frequency domains

      View table
      View in Article

      Table 5. Statistical parameters of vibration signals in time and frequency domains

      序号特征值序号特征值序号特征值
      1均值10波形指标19幅值极差
      2均方根值11峰值指标20功率谱方差指标
      3方根幅值12脉冲指标21幅值最大值
      4偏度13裕度指标22谱原点矩
      5峭度14偏度指标23功率谱重心指标
      6方差15峭度指标24平均幅值
      7最大值16均值频率25功率谱方差
      8最小值17频率中心26相关因子
      9峰值18幅值最小值27谐波因子
    • Table 6. The average accuracy under different activation functions

      View table
      View in Article

      Table 6. The average accuracy under different activation functions

      序号第1层第2层平均准确率
      1SigmoidSigmoid0.924 2
      2TanhTanh0.925 1
      3ReLUReLU0.945 8
      4TanhReLU0.947 5
      5SigmoidReLU0.935 2
      6TanhSigmoid0.934 8
    • Table 7. Parameter setting of GCN

      View table
      View in Article

      Table 7. Parameter setting of GCN

      网络结构参数参数设置
      隐含层层数/个2
      输入层神经元个数/个27
      隐含层神经元个数/个{64,38}
      输出层神经元个数/个5
      激活函数{Tanh,ReLU}
      随机失活[0.25,0.25]
      学习率0.01
      优化器Adam
      损失函数交叉熵损失函数
    • Table 8. Classification results of different classifiers

      View table
      View in Article

      Table 8. Classification results of different classifiers

      模型召回率精确率准确率
      SVM0.5220.4170.413
      BPNN0.8420.8380.843
      RF0.8930.8890.886
      CNN0.7830.7820.775
      GCN0.9520.9450.947
    Tools

    Get Citation

    Copy Citation Text

    Ruihan WANG, Hui CHEN, Cong GUAN, Mengzhuo HUANG. Fault diagnosis of marine diesel engines based on graph convolutional network under unbalanced datasets[J]. Chinese Journal of Ship Research, 2022, 17(5): 289

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Marine Machinery, Electrical Equipment and Automation

    Received: Apr. 17, 2022

    Accepted: --

    Published Online: Mar. 26, 2025

    The Author Email:

    DOI:10.19693/j.issn.1673-3185.02859

    Topics