基于CNN-BiLSTM-Triplet的C919液压系统状态监测方案

张振良; 何荣荣; 周相志; 崔哲; 董杰; 何修齐

doi:10.13224/j.cnki.jasp.20240716

基于CNN-BiLSTM-Triplet的C919液压系统状态监测方案

doi: 10.13224/j.cnki.jasp.20240716

张振良^{1, 2,},
何荣荣^1,*, ,,
周相志¹,
崔哲¹,
董杰¹,
何修齐³

1.
鄂尔多斯应用技术学院大飞机学院，内蒙古鄂尔多斯 017000
2.
内蒙古工业大学理学院，呼和浩特 010051
3.
中国民用航空飞行学院绵阳分院，四川绵阳 621000

基金项目: 鄂尔多斯应用技术学院校级科研项目（KYYB2020006）；内蒙古自治区高等学校科学研究项目（NJZY222230）；民航飞行技术与飞行安全重点实验室飞行技术专题项目（FZ2022ZX37）

详细信息

作者简介:
张振良（1995-），男，讲师，硕士生，主要研究方向为健康监测、深度学习

通讯作者:
何荣荣（1993-），女，助理实验师，主要研究方向为复杂装备可靠性。E-mail：1031026203@qq.com

中图分类号: V240.2
计量
- 文章访问数: 1090
- HTML浏览量: 358
- PDF量: 34
- 被引次数: 0
出版历程
- 收稿日期: 2024-10-17
- 网络出版日期: 2025-08-07

Condition monitoring scheme of C919 hydraulic system based on CNN-BiLSTM-Triplet

ZHANG Zhenliang^{1, 2
,},
HE Rongrong^{1,*
, ,},
ZHOU Xiangzhi¹,
CUI Zhe¹,
DONG Jie¹,
HE Xiuqi³

1.
Comac Aviation Colleage，Ordos Institute of Technology，Ordos Inner Mongolia 017000，China
2.
College of Science，Inner Mongolia University of Technology，Hohhot 010051，China
3.
Civil Avation Flight University of China Mianyang Flight College，Mianyang Sichuan 621000，China

摘要

摘要:
针对液压系统故障模式复杂难以识别、国产飞机液压系统状态监测困难等问题，提出了基于三通道注意力机制的卷积双向长短时记忆网络（CNN-BiLSTM-Triplet）来进行液压系统故障识别及特征分析。对数据进行降维并重塑成三维格式，利用卷积网络获取特征图。通过旋转操作和残差变换分别计算三通道注意力并传输到网络中来强化特定数据特征，使用BiLSTM学习提取后的特征，引入Dropout层以及L2正则化提升鲁棒性；并通过液压系统故障数据集验证了模型性能及其泛化能力。利用机器学习解释模型进行特征分析，并可视化特定特征和错分类样本的模型表现来确定状态监测指标及其优先性。实验证明：所提方法对液压系统5种故障模式的识别率均高于95.77%，并确定了对故障影响最大的特征为压力及流量，并结合C919客机液压系统的具体结构新增了压力脉动传感器完善液压系统状态监测方案。
- 状态监测 /
- 液压系统故障 /
- 故障识别 /
- 卷积神经网络 /
- 双向长短时记忆 /
- 三通道注意力
Abstract:
To address the challenges posed by complex fault modes in hydraulic systems and the difficulty of status monitoring in domestically manufactured aircraft, convolutional neural networks-bidirectional long short-term memory-triplet triplet attention （CNN-BiLSTM-Triplet） model was proposed for hydraulic system fault identification and feature analysis. Initially, the data were dimensionally reduced and reshaped into a three-dimensional format. A convolutional network was then employed to extract feature maps. Triplet attention was calculated using rotation operations and residual transformations, and subsequently transmitted to the network to enhance specific data features. BiLSTM was employed to capture and learn the extracted features, with a Dropout layer and L2 regularization incorporated to mitigate overfitting and enhance model robustness. The model’s performance and generalization ability were validated through fault datasets of hydraulic systems. Model interpretation method was applied for model feature analysis. Furthermore, the behavior of specific features and misclassified samples was analyzed to determine the status monitoring indicators and their relative priorities. Experimental results demonstrated that the proposed method achieved a recognition rate above 95.77% for all five fault modes in the hydraulic system, and identified pressure and flow as the most significant features for faults. Additionally, based on the specific structure of the C919 aircraft hydraulic system, a pressure pulsation sensor was added to improve the condition monitoring scheme.
- condition monitoring /
- fault of hydraulic system /
- fault identification /
- convolutional neural networks /
- bidirectional long short-term memory /
- triplet attention

HTML

图 1 民用飞机液压系统循环简化图

Figure 1. Schematic diagram of the hydraulic system circulation of civil aircraft

下载: 全尺寸图片幻灯片

图 2 CNN-BiLSTM-Triplet模型结构示意图

Figure 2. Schematic diagram of the CNN-BiLSTM-Triplet model structure

下载: 全尺寸图片幻灯片

图 3 液压试验台示意图

Figure 3. Hydraulic system diagram

下载: 全尺寸图片幻灯片

图 4 SE主成分载荷分布图

Figure 4. Distribution diagram of SE principal component loadings

下载: 全尺寸图片幻灯片

图 5 SE主成分累计方差解释率

Figure 5. Cumulative variance explained rate diagram of SE principal components

下载: 全尺寸图片幻灯片

图 6 液压蓄能器故障的训练损失和准确率变化图

Figure 6. Training loss and accuracy variation chart of hydraulic accumulator faults

下载: 全尺寸图片幻灯片

图 7 液压蓄能器故障诊断结果

Figure 7. Diagnosis results of hydraulic accumulator faults

下载: 全尺寸图片幻灯片

图 8 阀门故障诊断结果

Figure 8. Diagnosis results of valve faults

下载: 全尺寸图片幻灯片

图 9 液压泵泄漏故障诊断结果

Figure 9. Diagnosis results of pump leakage fault

下载: 全尺寸图片幻灯片

图 10 冷却系统故障诊断结果

Figure 10. Diagnosis results of hydraulic cooling system faults

下载: 全尺寸图片幻灯片

图 11 系统稳定状态诊断结果

Figure 11. Diagnosis results of system steady state

下载: 全尺寸图片幻灯片

图 12 特征摘要图

Figure 12. SHAP feature summary plot

下载: 全尺寸图片幻灯片

图 13 numpy数组格式的特征重要性总结图

Figure 13. Feature importance summary plot in numpy array

下载: 全尺寸图片幻灯片

图 14 各个特征对不同功率模式下模型预测的贡献度

Figure 14. Contribution of each feature to model predictions under different power modes

下载: 全尺寸图片幻灯片

图 15 C919客机液压系统传感器布置方案

Figure 15. Sensor layout scheme of the hydraulic system for C919 aircraft

下载: 全尺寸图片幻灯片

表 1 故障数据集特征数据描述

Table 1. Description of fault dataset feature data

特征代码	测量值	降维前特征量	降维后主成分数	标准单位
PS1	测点P1压力	6000	5	Pa
PS2	测点P2压力	6000	6	Pa
PS3	测点P3压力	6000	9	Pa
PS4	测点P4点压力	6000	2	Pa
PS5	测点P5压力	6000	2	Pa
PS6	测点P6点压力	6000	2	Pa
EPS	电动机功率传感器	6000	5	W
FS1	测点P7流量	600	7	m³/s
FS2	测点P8流量	600	2	m³/s
TS1	测点P9温度	60	1	K
TS2	测点P10温度	60	1	K
TS3	测点P11温度	60	1	K
TS4	测点P12温度	60	1	K
VS	振动	60	7	m/s
SE	效率因子	60	5	%
CE	虚拟冷却效率	60	2	%
CP	虚拟冷却功率	60	2	W
降维后合计特征			60

下载: 导出CSV

表 2 故障数据集标签数据描述

Table 2. Description of fault dataset label data

故障代码（分类）	冷却效率	阀门故障	泵泄漏	液压蓄能失效	系统稳定状态
0	完全有效	流畅切换	无泄漏	无故障	稳定
1	效率降低	轻微滞后	轻微泄漏	压力轻微降低	尚未稳定
2	基本失效	严重滞后	严重泄漏	压力严重降低
3		基本失效		完全故障

下载: 导出CSV

表 3 模型参数设置

Table 3. Model parameter settings

参数	设定值	参数	设定值
输入层结构	3×20	批处理数量	128
卷积层输出通道数	256	迭代次数	500
卷积层卷积核	3～5	测试集占比	0.15
池化窗口	1	验证集占比	0.15
BiLSTM层输出特征	128	优化器	Adam
全连接层激活函数	Softmax	损失函数	交叉熵损失
数据输入格式	3维	失活率	0.4

下载: 导出CSV

表 4 不同故障下的多模型性能对比

Table 4. Comparison of multi-model performance under different faults

故障模式	模型	准确率/%	精确率/%	召回率/%	F1/%	漏检率/%	MCC/%
冷却效率	CNN-BiLSTM-Triplet	100.00	100.00	100.00	100.00	0	100.00
	CNN-BiLSTM-Attention	100.00	100.00	100.00	100.00	0	100.00
	CNN-BiLSTM	100.00	100.00	100.00	100.00	0	100.00
	CNN-LSTM	100.00	100.00	100.00	100.00	0	100.00
	Random Forest	95.69	96.08	94.28	95.09	1.36	90.34
阀门故障	CNN-BiLSTM-Triplet	95.77	94.73	94.69	94.69	0.91	93.78
	CNN-BiLSTM-Attention	94.56	93.51	92.57	92.99	1.21	91.66
	CNN-BiLSTM	92.74	90.48	89.65	90.01	1.21	88.85
	CNN-LSTM	92.76	92.26	92.89	92.47	2.11	89.69
	Random Forest	89.12	89.14	86.85	87.80	1.81	75.96
泵泄漏	CNN-BiLSTM-Triplet	98.19	97.92	97.62	97.76	0.45	96.96
	CNN-BiLSTM-Attention	96.83	95.66	95.90	95.77	0.68	94.49
	CNN-BiLSTM	95.69	93.86	95.23	94.47	0.91	92.85
	CNN-LSTM	93.88	91.24	91.11	91.01	2.27	89.50
	Random Forest	86.40	82.18	78.64	77.82	3.17	77.77
液压蓄能器失效	CNN-BiLSTM-Triplet	98.79	98.68	98.63	98.65	0.60	98.32
	CNN-BiLSTM-Attention	95.92	94.64	95.31	94.80	0.60	94.41
	CNN-BiLSTM	93.20	91.89	91.42	91.64	0.60	89.73
	CNN-LSTM	91.61	88.47	90.29	89.34	0.60	87.27
	Random Forest	86.88	84.02	84.43	83.82	0.60	82.07
系统稳定状态	CNN-BiLSTM-Triplet	97.28	97.85	95.91	96.80	2.49	93.79
	CNN-BiLSTM-Attention	95.69	96.08	94.28	95.09	1.81	90.34
	CNN-BiLSTM	95.01	94.99	93.68	94.29	1.36	88.66
	CNN-LSTM	93.88	93.19	93.32	93.25	0.91	86.51
	Random Forest	92.74	93.97	89.98	91.56	2.27	83.85

下载: 导出CSV

参考文献(25)

[1]	冯东洋, 姜春英, 鲁墨武, 等. 基于TSFFCNN-PSO-SVM的飞机起落架液压系统故障诊断[J]. 航空动力学报, 2024, 39(1): 20220111. FENG Dongyang, JIANG Chunying, LU Mowu, et al. Fault diagnosis of aircraft landing gear hydraulic system based on TSFFCNN-PSO-SVM[J]. Journal of Aerospace Power, 2024, 39(1): 20220111. (in Chinese FENG Dongyang, JIANG Chunying, LU Mowu, et al. Fault diagnosis of aircraft landing gear hydraulic system based on TSFFCNN-PSO-SVM[J]. Journal of Aerospace Power, 2024, 39(1): 20220111. (in Chinese)
[2]	ZHANG Kaigan, XIA Tangbin, XU Yuhui, et al. A hybrid prognostic & health management framework across multi-level engineering systems with scalable convolution neural networks and adjustable functional regression models[J]. Advanced Engineering Informatics, 2024, 61: 102566.
[3]	LI Weidong, JIANG Heping, FENG Chunhua, et al. A hybrid intelligent diagnostic approach for spool jamming faults of hydraulic directional valves[J]. Measurement, 2025, 241: 115706. doi: 10.1016/j.measurement.2024.115706
[4]	LIU Yijian , FANG Yanjun , ZHU Xuemei. Modeling of hydraulic turbine systems based on a Bayesian-Gaussian neural network driven by sliding window data[J]. Journal of Zhejiang University-Science C (Computer & Electronics), 2010, 11(1): 56-62.
[5]	LIU Huizhou, YAN Shibo, HUANG Mengxing, et al. A fault diagnosis method for hydraulic system based on multi-branch neural networks[J]. Engineering Applications of Artificial Intelligence, 2024, 137: 109188. doi: 10.1016/j.engappai.2024.109188
[6]	张立杰, 鄂东辰. 基于自适应MPCA的翻车机液压系统状态监测与故障诊断[J]. 振动与冲击, 2018, 37(8): 245-250, 256. ZHANG Lijie, E Dongchen. Car dumper hydraulic system state monitoring and fault diagnosis based on adaptive MPCA[J]. Journal of Vibration and Shock, 2018, 37(8): 245-250, 256. (in Chinese ZHANG Lijie, E Dongchen. Car dumper hydraulic system state monitoring and fault diagnosis based on adaptive MPCA[J]. Journal of Vibration and Shock, 2018, 37(8): 245-250, 256. (in Chinese)
[7]	YAN Wei, SHI Yu, JI Zengyan, et al. Intelligent predictive maintenance of hydraulic systems based on virtual knowledge graph[J]. Engineering Applications of Artificial Intelligence, 2023, 126: 106798.
[8]	GUO Dexuan, TANG Juan, XIE Zongfa, et al. A twice-open control method for a hydraulic variable valve system in a diesel engine[J]. Processes, 2024, 12(7): 1526-1526.
[9]	LI Feng, LIU Shiheng, WANG Tianhu, et al. Optimal planning for integrated electricity and heat systems using CNN-BiLSTM-Attention network forecasts[J]. Energy, 2024, 309: 133042. doi: 10.1016/j.energy.2024.133042
[10]	戚晓利, 毛俊懿, 王兆俊, 等. 基于SConvNeXt-ECMS与DBO-RELM模型的滚动轴承故障诊断方法[J]. 航空动力学报, 2025, 40(5): 20230678. QI Xiaoli, MAO Junyi, WANG Zhaojun, et al. Rolling bearing fault diagnosis method based on SConvNeXt-ECMS and DBO-RELM model[J]. Journal of Aeros pace Power, 2025, 40(5): 20230678. (in Chinese QI Xiaoli, MAO Junyi, WANG Zhaojun, et al. Rolling bearing fault diagnosis method based on SConvNeXt-ECMS and DBO-RELM model[J]. Journal of Aeros pace Power, 2025, 40(5): 20230678. (in Chinese)
[11]	姜建国, 杨效岩, 毕洪波. 基于VMD-FE-CNN-BiLSTM的短期光伏发电功率预测[J]. 太阳能学报, 2024, 45(7): 462-473. JIANG Jianguo, YANG Xiaoyan, BI Hongbo. Photovoltaic power forecasting method based on VMD-FE-CNN-BiLSTM[J]. Acta Energiae Solaris Sinica, 2024, 45(7): 462-473. (in Chinese JIANG Jianguo, YANG Xiaoyan, BI Hongbo. Photovoltaic power forecasting method based on VMD-FE-CNN-BiLSTM[J]. Acta Energiae Solaris Sinica, 2024, 45(7): 462-473. (in Chinese)
[12]	WANG Rui, SHI Xiaoshuang, PANG Shuting, et al. Cross-attention guided loss-based deep dual-branch fusion network for liver tumor classification[J]. Information Fusion, 2025, 114: 102713. doi: 10.1016/j.inffus.2024.102713
[13]	KUMAR A, GLOWACZ A, TANG Hesheng, et al. Knowledge addition for improving the transfer learning from the laboratory to identify defects of hydraulic machinery[J]. Engineering Applications of Artificial Intelligence, 2023, 126: 106756. doi: 10.1016/j.engappai.2023.106756
[14]	ZHANG Bo, XIONG Yujie, XIA Chunming, et al. Transformer-based end-to-end attack on text CAPTCHAs with triplet deep attention[J]. Computers & Security, 2024, 146: 104058.
[15]	民用航空器维修基础系列教材编写委员会. 民用航空器维修基础系列教材: 第 3 册飞机结构和系统[M]. 北京: 中国民用航空维修协会, 2020. Civil Aircraft Maintenance Fundamentals Series Textbook Compilation Committee. Civil aircraft maintenance fundamentals series textbook: Volume 3 aircraft structures and systems [M]. Beijing: China Civil Aviation Maintenance Association, 2020. (in Chinese Civil Aircraft Maintenance Fundamentals Series Textbook Compilation Committee. Civil aircraft maintenance fundamentals series textbook: Volume 3 aircraft structures and systems [M]. Beijing: China Civil Aviation Maintenance Association, 2020. (in Chinese)
[16]	冯蕴雯, 潘维煌, 路成, 等. 基于故障逻辑的民机液压状态监控与故障诊断[J]. 系统工程与电子技术, 2025, 47(3): 842-854. FENG Yunwen, PAN Weihuang , LU Cheng, et al. Fault-logic-based hydraulic status monitoring and fault diagnosis for civil aircraft[J]. System Engineering and Electronics Technology, 2025, 47(3): 842-854. (in Chinese FENG Yunwen, PAN Weihuang , LU Cheng, et al. Fault-logic-based hydraulic status monitoring and fault diagnosis for civil aircraft[J]. System Engineering and Electronics Technology, 2025, 47(3): 842-854. (in Chinese)
[17]	谷立臣, 刘沛津, 陈江城. 基于电参量信息融合的液压系统状态识别技术[J]. 机械工程学报, 2011, 47(24): 141-150. GU Lichen, LIU Peijin, CHEN Jiangcheng. State recognition technique of hydraulic system based on electrical parameters information fusion[J]. Journal of Mechanical Engineering, 2011, 47(24): 141-150. (in Chinese doi: 10.3901/JME.2011.24.141 GU Lichen, LIU Peijin, CHEN Jiangcheng. State recognition technique of hydraulic system based on electrical parameters information fusion[J]. Journal of Mechanical Engineering, 2011, 47(24): 141-150. (in Chinese) doi: 10.3901/JME.2011.24.141
[18]	欧阳小平, 杨华勇, 郭生荣, 等. 现代飞机液压技术[M]. 杭州: 浙江大学出版社, 2016. OUYANG Shaoping, YANG Huayong, GUO Shengrong, et al. Modern hydraulics for aircrafts[M]. Hangzhou: Zhejiang University Press, 2016. (in Chinese OUYANG Shaoping, YANG Huayong, GUO Shengrong, et al. Modern hydraulics for aircrafts[M]. Hangzhou: Zhejiang University Press, 2016. (in Chinese)
[19]	李浩文. 基于CNN-BiLSTM-Attention的雨雪条件下高速公路短时交通流预测研究[D]. 哈尔滨: 哈尔滨工业大学, 2023. LI Haowen. Study on short-term traffic flow prediction of expressway under rain and snow conditions based on CNN-BiLSTM-Attention[D]. Harbin: Harbin Institute of Technology, 2023. (in Chinese LI Haowen. Study on short-term traffic flow prediction of expressway under rain and snow conditions based on CNN-BiLSTM-Attention[D]. Harbin: Harbin Institute of Technology, 2023. (in Chinese)
[20]	GRAVES A, SCHMIDHUBER J. Framewise phoneme classification with bidirectional LSTM and other neural network architectures[J]. Neural Networks, 2005, 18(5/6): 602-610.
[21]	骆东松, 王春梅, 王毅科. 基于CNN-Attention-BiLSTM的变工况滚动轴承故障诊断[J]. 控制工程, 2022, 33(11): 6789-6801. LUO Dongsong, WANG Chunmei, WANG Yike. Fault diagnosis of rolling bearings under variable operating conditions based on CNN-Attention-BiLSTM [J]. Control Engineering, 2022, 33(11): 6789-6801 (in Chinese). LUO Dongsong, WANG Chunmei, WANG Yike. Fault diagnosis of rolling bearings under variable operating conditions based on CNN-Attention-BiLSTM [J]. Control Engineering, 2022, 33(11): 6789-6801 (in Chinese).
[22]	HUANG Keke, WU Shujie, LI Fanbiao, et al. Fault diagnosis of hydraulic systems based on deep learning model with multirate data samples[J]. IEEE Transactions on Neural Networks and Learning Systems, 2022, 33(11): 6789-6801. doi: 10.1109/TNNLS.2021.3083401
[23]	REHMAN A, KHAN A, ALI M A, et al. Performance analysis of PCA, sparse PCA, kernel PCA and incremental PCA algorithms for heart failure prediction[C]//2020 International Conference on Electrical, Communication, and Computer Engineering. Piscataway, US: IEEE, 2020: 1-5.
[24]	ALOMARI Y, ANDÓ M. SHAP-based insights for aerospace PHM: temporal feature importance, dependencies, robustness, and interaction analysis[J]. Results in Engineering, 2024, 21: 101834. doi: 10.1016/j.rineng.2024.101834
[25]	葛薇, 王少萍. 航空液压泵磨损状况预测[J]. 北京航空航天大学学报, 2011, 37(11): 1410-1414. GE Wei, WANG Shaoping. Wear condition prediction of hydraulic pump[J]. Journal of Beijing University of Aeronautics and Astronautics, 2011, 37(11): 1410-1414. (in Chinese GE Wei, WANG Shaoping. Wear condition prediction of hydraulic pump[J]. Journal of Beijing University of Aeronautics and Astronautics, 2011, 37(11): 1410-1414. (in Chinese)