吉林大学学报(工学版) ›› 2022, Vol. 52 ›› Issue (9): 2107-2118.doi: 10.13229/j.cnki.jdxbgxb20220062

• • 上一篇    

基于数据驱动的车用燃料电池故障在线自适应诊断算法

王克勇1,2(),鲍大同1,周苏1()   

  1. 1.同济大学 汽车学院,上海 201804
    2.上海捷氢科技股份有限公司,上海 201804
  • 收稿日期:2022-01-13 出版日期:2022-09-01 发布日期:2022-09-13
  • 通讯作者: 周苏 E-mail:wangkeyong@126.com;suzhou@tongji.edu.cn
  • 作者简介:王克勇(1978-),男,高级工程师,硕士. 研究方向:燃料电池系统开发及故障诊断. E-mail:wangkeyong@126.com
  • 基金资助:
    同济大学-AVL李斯特博士后/博士生基金项目(1700165103425110)

Data-driven online adaptive diagnosis algorithm towards vehicle fuel cell fault diagnosis

Ke-yong WANG1,2(),Da-tong BAO1,Su ZHOU1()   

  1. 1.School of Automotive Studies,Tongji University,Shanghai 201804,China
    2.Shanghai Hydrogen Propulsion Technology Co. ,Ltd. ,Shanghai 201804,China
  • Received:2022-01-13 Online:2022-09-01 Published:2022-09-13
  • Contact: Su ZHOU E-mail:wangkeyong@126.com;suzhou@tongji.edu.cn

摘要:

长短期记忆(LSTM)多分类算法可以有效地实现车用燃料电池的在线智能故障诊断,然而在实际应用中,车用燃料电池随着运行时间的增加,内部特性会发生衰退,初始诊断模型可能不能满足长期故障诊断的精度。针对该问题,本文基于AVL CURISE M软件搭建了PEMFC原始和衰退模型,并使用模型产生故障数据。随后设计了自适应算法,并使用模型产生的数据进行自适应训练,使得诊断模型能够适应电堆的衰退,保证了车用燃料电池在线智能诊断的精度。在实际燃料电池系统中对该方案进行实测验证,证明了其有效性,该方案可以基于“车端-云端”平台对燃料电池系统进行诊断,算法权重可以自适应更新,以完成对电堆老化的适应,有较好的应用前景。

关键词: 车辆工程, LSTM多分类, 在线智能诊断, 模型参数, 自适应

Abstract:

Long short-term memory (LSTM) multi-classification algorithm can effectively realize the online intelligent fault diagnosis of vehicle fuel cells. However, in practical applications, the internal characteristics of vehicle fuel cells will decline with the increase of operating time, and the initial diagnosis model may not be able to meet long-term fault conditions. Aiming at this problem, PEMFC original and decay models based on AVL CURISE M software were built, and fault data was generated using the models. Then, an adaptive algorithm was designed, and the data generated by the model was used for adaptive training, so that the diagnosis model can adapt to the decline of the stack and ensure the accuracy of the online intelligent diagnosis of vehicle fuel cells. Based on this scheme, the actual fuel cell system has been tested and verified, which proves the effectiveness of the scheme. This scheme can adaptively update the weight of the diagnosis algorithm of the fuel cell system based on the "vehicle-cloud" platform and complete the aging of the stack. It has a good application prospect.

Key words: vehicle engineering, LSTM multi-classification, online intelligent diagnosis, model parameters, self-adaptation

中图分类号: 

  • TK91

图1

燃料电池系统动态模型"

图2

构建数据集使用的电流工况"

表1

降维结果"

信号相关系数能在实际系统中获得

选择作为

数据集

空压机转速-0.0305
冷却泵转速0.0035
背压阀开度-0.1382
空气计量比-0.1068
电堆电压-0.0221
电堆电流0.0009
氢气流阻0.0775
氢气出压力0.0085
氢气流阻-0.1208
空气出口压力0.0289
空气入口压力0.0074
氢气入口压力0.0082
空气流量0.0006
氢气流量-0.0036

图3

LSTM神经网络单元"

表2

数据划分详情"

工况训练样本数量测试样本数量标签
正常12 3533 0080
轻微膜干11 3932 8481
严重膜干11 3932 8482
轻微水淹12 3533 0083
严重水淹12 3533 0084
轻微欠气11 0392 7595
严重欠气11 0392 7596

表3

网络结构"

结构层网络参数输出
长短时记忆网络层(LSTM)

步长(timestep):100

隐藏节点(hidden size):24

100×24
全连接层1(FC1)2424
激活层(Relu)2424
全连接层2(FC2)77

图4

网络结构"

表4

诊断结果"

精确率召回率耗时/ms存储量/kbit
0.980.98<117

表5

燃料电池物性参数衰退项目"

衰退项目原始模型衰退模型
催化层催化层厚度/mm0.010.009 5
催化层电子迁移率0.7570.719
参考交换电流密度3.3973.227

催化层活化能量/

(J·mol-1

66 00062 700
氧含量0.750.7125
水含量10.95
质子交换膜膜厚度/mm0.0250.023
热导率/[W·(m·K)-10.20.19
电导率(阿伦尼乌斯公式首项)/(S·m-10.510.48
电导率(阿伦尼乌斯公式尾项)/(S·m-110 54210 014.9
气体扩散层气体扩散层厚度/mm0.340.323
气体扩散层相对穿透率10.95
气体扩散层孔隙率0.4470.424
阳极双极板流动效率0.50.475
阳极双极板摩擦因数6.36.615
阴极双极板流动效率0.50.475
阴极双极板摩擦因数15.416.17

图5

自适应算法流程"

表6

自适应训练时候使用的超参数"

超参数数值
批处理数量(Batch size)150
训练轮数(Epochs)4096
最大迭代次数6
自适应置信度0.95
初始学习率(Initial learning rate)5×10-4

图6

自适应诊断模型和原始诊断模型对比"

图7

大功率燃料电池系统"

表7

正常工况的边界条件"

电流密度/(mA·cm-2空气计量比冷却水温度/℃
802.968
1002.868
2002.671
3002.574
4002.378
5002.279
6002.281
7002.183
8001.984
9001.985
10001.986
11001.887

表8

故障嵌入的边界条件"

状态空气计量比冷却水温度/℃
正常--
轻微膜干-+3
严重膜干-+5
轻微水淹--5
严重水淹--10
轻微欠气-0.1-
严重欠气-0.2-

图8

验证使用的燃料电池系统和FCU"

图9

故障试验拉取的电流"

图10

故障试验数据的高频阻抗和空气流量"

图11

最后一轮自适应迭代训练评价指标"

表9

六轮自适应迭代结果"

对比项目精确率召回率迭代后剩余数据
推理用时/ms54.21
迭代1轮0.9800.9786450
迭代2轮0.9820.982740
迭代3轮0.9830.984446
迭代4轮0.9820.981241
迭代5轮0.9770.981101
迭代6轮0.9830.979-

原始诊断模型

推理衰退数据

0.9060.901-
1 中国汽车工程学会. 节能与新能源汽车技术路线图[M]. 北京:机械工业出版社,2021.
2 Benmouna A, Mohamed B, Daniel D, et al. Fault diagnosis methods for proton exchange membrane fuel cell system[J]. International Journal of Hydrogen Energy, 2017, 42(2): 1534-1543.
3 周苏,胡哲,文泽军. 基于K均值和支持向量机的燃料电池在线自适应故障诊断[J]. 同济大学学报, 2019, 47(2): 255-260.
Zhou Su, Hu Zhe, Wen Ze-jun. A K-means/support vector machine based self-adaptive online fault diagnosis method for fuel cell system[J]. Journal of Tongji University, 2019, 47(2): 255-260.
4 Gu X, Hou Z J, Cai J. Data-based flooding fault diagnosis of proton exchange membrane fuel cell systems using LSTM networks[J]. Energy and AI, 2021, 4: No. 100056.
5 Shao M, Zhu X J, Cao H F, et al. An artificial neural network ensemble method for fault diagnosis of proton exchange membrane fuel cell system[J]. Energy, 2014, 67: 268-275.
6 Sanchez D G, Garcia-Ybarra P L. PEMFC operation failure under severe dehydration[J]. International Journal of Hydrogen Energy, 2012, 37(8): 7279-7288.
7 Canut J M L, Latham R, Mérida W, et al. Impedance study of membrane dehydration and compression in proton exchange membrane fuel cells[J]. Journal of Power Sources, 2009, 192(2): 457-466.
8 Hui L, Tang Y H, Wang Z W, et al. A review of water flooding issues in the proton exchange membrane fuel cell[J]. Journal of Power Sources, 2008, 178(1): 103-117.
9 Zhang C Z, Zhou W J, Lan Z, et al. An experimental study on anode water management in high temperature PEM fuel cell[J]. International Journal of Hydrogen Energy, 2015, 40(13): 4666-4672.
10 Shimpalee S, Beuscher U, Zee J W V. Analysis of GDL flooding effects on PEMFC performance[J]. Electrochimica Acta, 2007, 52(24): 6748-6754.
11 O'Rourke J, Ramani M, Arcak M. In situ detection of anode flooding of a PEM fuel cell[J]. International Journal of Hydrogen Energy, 2009, 34(16): 6765-6770.
12 Kim M, Jung N, Eom K S, et al. Effects of anode flooding on the performance degradation of polymer electrolyte membrane fuel cells[J]. Journal of Power Sources, 2014, 266: 332-340.
13 Benmouna A, Becherif M, Depernet D, et al. Fault diagnosis methods for proton exchange membrane fuel cell system[J]. International Journal of Hydrogen Energy, 2017, 42(2): 1534-1543.
14 朱添麟. 基于电压调节的车载燃料电池空气供给系统建模与控制[D]. 长春: 吉林大学通信工程学院, 2021.
Zhu Tian-lin. Modeling and control of vehicle fuel cell air supply system based on voltage regulation[D]. Changchun: College of Communication Engineering, Jilin University, 2021.
15 杨朵. 燃料电池空气供给系统控制与故障诊断策略研究[D]. 合肥: 中国科学技术大学计算机科学与技术学院, 2021.
Yang Duo. Research on control and fault diagnosis strategy fuel cell air supply system[D]. Hefei: School of Computer and Control Engineering, University of Science and Technology of China, 2021.
16 孙田. 重卡用大功率燃料电池发动机空气供给控制策略研究[D]. 北京: 北京交通大学机械与电子控制工程学院, 2020.
Sun Tian. Research on control strategy of high power fuel cell engine air supply for heavy-duty truck[D]. Beijing: School of Mechanical and Electronic Control Engineering, Beijing Jiaotong University, 2020.
17 陈旭, 胡广地, 邓宇翔, 等. PEMFC燃料饥饿控制策略综述[J]. 电池, 2019, 49(2): 163-167.
Chen Xu, Hu Guang-di, Deng Yu-xiang, et al. Review of fuel starvation control strategies for PEMFC[J]. Battery, 2019, 49(2):163-167.
18 黄亮, 彭清, 谢长君, 等. 基于差分进化优化的支持向量机燃料电池故障诊断[J]. 电源技术, 2021, 45(10): 1316-1319.
Huang Liang, Peng Qing, Xie Chang-jun, et al. Fuel cell fault diagnosis for support vector machines optimized based on differential evolution algorithm[J]. Chinese Journal of Power Sources, 2021, 45(10): 1316-1319.
19 张少哲, 戴海峰, 袁浩, 等. 质子交换膜燃料电池电化学阻抗谱敏感性研究[J]. 机械工程学报, 2021, 57(14): 40-51.
Zhang Shao-zhe, Dai Hai-feng, Yuan Hao, et al. Sensiblity study on electrochemical impedance of proton exchange membrane fuel cell[J]. Joural of Mechanical Engineering, 2021, 57(14): 40-51.
20 贾坤晗. 进气相对湿度对车用燃料电池性能影响的研究[D]. 北京: 北京建筑大学机电与车辆工程学院, 2021.
Jia Kun-han. Research on the effects of inlet relative humidity on the performance of vehicle fuel cells[D]. Beijing: School of Mechatronics and Vehicle Engineering, Beijing University of Civil Engineering and Architecture, 2021.
21 何红曈. 质子交换膜燃料电池内部水含量特性分析[D]. 北京: 北京建筑大学机电与车辆工程学院, 2021.
He Hong-tong. Analysis of the variation characteristics of water content in the flow channel of PEMFC[D]. Beijing: School of Mechatronics and Vehicle Engineering, Beijing University of Civil Engineering and Architecture, 2021.
22 卫超强. 不同参数对质子交换膜燃料电池输出特性的影响[D]. 太原: 太原理工大学机械与运载工程学院, 2021.
Wei Chao-qiang. Effects of different parameters on the outpour characteristics of proton exchange membrane fuel cells[D]. Taiyuan: School of Mechanics and Transportation Engineering, Taiyuan University of Technology, 2021.
23 Devlin S J. Pearson correlation coefficient[EB/OL]. [2021-12-11].
[1] 张冲,胡云峰,宫洵,孙耀. 燃料电池阴极流量无模型自适应滑模控制器设计[J]. 吉林大学学报(工学版), 2022, 52(9): 2085-2095.
[2] 张佩,王志伟,杜常清,颜伏伍,卢炽华. 车用质子交换膜燃料电池空气系统过氧比控制方法[J]. 吉林大学学报(工学版), 2022, 52(9): 1996-2003.
[3] 池训逞,侯中军,魏伟,夏增刚,庄琳琳,郭荣. 基于模型的质子交换膜燃料电池系统阳极气体浓度估计技术综述[J]. 吉林大学学报(工学版), 2022, 52(9): 1957-1970.
[4] 裴尧旺,陈凤祥,胡哲,翟双,裴冯来,张卫东,焦杰然. 基于自适应LQR控制的质子交换膜燃料电池热管理系统温度控制[J]. 吉林大学学报(工学版), 2022, 52(9): 2014-2024.
[5] 李永明,裴小轩,伊曙东. 混合动力汽车动力电池自适应神经网络优化控制[J]. 吉林大学学报(工学版), 2022, 52(9): 2063-2068.
[6] 陈凤祥,伍琪,李元松,莫天德,李煜,黄李平,苏建红,张卫东. 2.5吨燃料电池混合动力叉车匹配、仿真及优化[J]. 吉林大学学报(工学版), 2022, 52(9): 2044-2054.
[7] 武小花,余忠伟,朱张玲,高新梅. 燃料电池公交车模糊能量管理策略[J]. 吉林大学学报(工学版), 2022, 52(9): 2077-2084.
[8] 高青,王浩东,刘玉彬,金石,陈宇. 动力电池应急冷却喷射模式实验分析[J]. 吉林大学学报(工学版), 2022, 52(8): 1733-1740.
[9] 王奎洋,何仁. 基于支持向量机的制动意图识别方法[J]. 吉林大学学报(工学版), 2022, 52(8): 1770-1776.
[10] 王骏骋,吕林峰,李剑敏,任洁雨. 分布驱动电动汽车电液复合制动最优滑模ABS控制[J]. 吉林大学学报(工学版), 2022, 52(8): 1751-1758.
[11] 刘汉武,雷雨龙,阴晓峰,付尧,李兴忠. 增程式电动汽车增程器多点控制策略优化[J]. 吉林大学学报(工学版), 2022, 52(8): 1741-1750.
[12] 杨红波,史文库,陈志勇,郭年程,赵燕燕. 基于某二级减速齿轮系统的齿面修形优化[J]. 吉林大学学报(工学版), 2022, 52(7): 1541-1551.
[13] 聂光明,谢波,田彦涛. 基于Frenet框架的协同自适应巡航控制算法设计[J]. 吉林大学学报(工学版), 2022, 52(7): 1687-1695.
[14] 郝帅,程川泰,王军年,张君媛,俞有. 运动型SUV驾驶室布置人机优化设计与测试评价[J]. 吉林大学学报(工学版), 2022, 52(7): 1477-1488.
[15] 张家旭,郭崇,王晨,赵健,王欣志. 基于半实物仿真平台的自动泊车系统性能评价[J]. 吉林大学学报(工学版), 2022, 52(7): 1552-1560.
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
No Suggested Reading articles found!