Fault diagnosis using wavelet packet and neural network in tilting control system of tilting train
-
摘要: 针对摆式列车倾摆控制系统故障的特点, 研究了神经网络结合小波包分析进行故障诊断的方法, 采用小波包分解和信号重构的方法, 将在摆式列车试验台上采集到的振动信号分解到不同的频带以提取有关的故障信息, 并将振动信号各频带内的能量特征作为训练样本输入前向神经网络, 用优于改进梯度下降法的Levenberg- Marquardt优化方法对网络进行训练, 对倾摆控制系统的常见故障进行识别和诊断。实践表明, 该方法对摆式列车倾摆系统故障的诊断是可靠的。Abstract: With the view of fault characteristics of tilting control system, this paper put forward a fault diagnosis approach employing a combination of wavelet packet and neural network.The method using wavelet packet decomposition and signal reconstruction was proposed to extract fault information from vibration signal obtained from testing jig of tilting train. By analyzing the energy of signal in full spectrum bands, the symptom that representes fault was inputted to a feed forward neural network trained by Levenberg-Marquardt optimization, which progress was very fast comparing with the improved gradient descent algorithm.The trained feed forward neural network can report the typical faults of tilting control system.Trial and research show that the method is practicable for fault diagnosis in tilting control system of reality tilting train.
-
Key words:
- tilting train /
- tilting control system /
- neural network /
- wavelet packet /
- fault diagnosis
-
表 1 常见故障和故障特征的相应关系
Table 1. Relationship between frequent faults and characteristics
样本 x1 x2 x3 x4 x5 x6 x7 x8 正常样本1 0.9796 0.1618 0.0528 0.0628 0.0399 0.0431 0.0461 0.0439 正常样本2 0.9773 0.1520 0.0645 0.0740 0.0530 0.0538 0.0561 0.0579 正常样本3 0.9717 0.1811 0.0648 0.0711 0.0559 0.0576 0.0612 0.0602 执行机构卡死故障1 0.6561 0.6991 0.1031 0.1919 0.0376 0.0608 0.1136 0.1143 执行机构卡死故障2 0.6928 0.6466 0.1672 0.2508 0.0358 0.0448 0.0578 0.0671 加速度传感器故障1 0.9222 0.3549 0.0729 0.0757 0.0514 0.0509 0.0648 0.0563 加速度传感器故障2 0.9207 0.3539 0.0826 0.1015 0.0416 0.0464 0.0581 0.0509 位移传感器故障1 0.9815 0.1791 0.0347 0.0381 0.0203 0.0200 0.0221 0.0240 位移传感器故障2 0.9834 0.1663 0.0353 0.0378 0.0247 0.0253 0.0257 0.0275 
下载: 导出CSV
表 2 网络训练样本的期望输出和实际输出
Table 2. The expected outputs and factual outputs of training samples
故障类型 期望输出 实际输出 节点1 节点2 节点3 节点4 节点1 节点2 节点3 节点4 正常 1 0 0 0 0.998 8 0.000 1 -0.000 2 0.001 8 正常 1 0 0 0 1.007 7 0.000 1 -0.016 3 0.008 4 正常 1 0 0 0 0.992 6 0.000 0 0.016 2 -0.008 9 执行机构卡死故障 0 1 0 0 0.001 0 0.999 4 -0.000 3 0.002 9 执行机构卡死故障 0 1 0 0 -0.000 1 0.999 9 0.000 9 -0.003 4 加速度传感器故障 0 0 1 0 0.001 7 -0.000 8 0.995 7 0.003 4 加速度传感器故障 0 0 1 0 0.000 7 0.000 8 1.003 2 -0.004 7 位移传感器故障 0 0 0 1 -0.002 5 0.000 1 -0.003 8 1.006 2 位移传感器故障 0 0 0 1 0.001 0 -0.000 0 0.004 2 0.994 8
下载: 导出CSV
表 3 待识别的样本
Table 3. The samples to be identified
待识别样本 x1 x2 x3 x4 x5 x6 x7 x8 样本1 0.9756 0.1667 0.0645 0.0711 0.0530 0.0542 0.0562 0.0581 样本2 0.6743 0.6734 0.1356 0.2216 0.0368 0.0549 0.0857 0.0919 样本3 0.9212 0.3541 0.0782 0.0893 0.0475 0.0498 0.0611 0.0532 样本4 0.9821 0.1734 0.0350 0.0381 0.0220 0.0234 0.0239 0.0260
下载: 导出CSV
表 4 待识别样本的实际输出和识别结论
Table 4. The factual outputs of the samples to be identified and the conclusions
故障类型 识别结论 实际输出 节点1 节点2 节点3 节点4 节点1 节点2 节点3 节点4 正常 1 0 0 0 1.0285 0.0001 -0.0037 -0.0249 执行机构卡死故障 0 1 0 0 -0.0165 0.9719 0.0374 0.0087 加速度传感器故障 0 0 1 0 0.0803 -0.0004 0.9687 -0.0486 位移传感器故障 0 0 0 1 -0.0046 -0.0000 0.0022 1.0024
下载: 导出CSV
-
[1] 崔锦泰, 程正兴. 小波分析导论[M]. 西安: 西安交通大学出版社, 1995. [2] Vemuri A T, Polycarpou M M. Neural-network-based robust fault diagnosis in robotic systems[J]. IEEE Transactions on Neural Network, 1997, 8(6): 1 410-1 420. [3] Naidu S R, Zafiriou E. Use of neural networks for sensor failure detection in control system[J]. IEEE Control System Magazine, 1990, 49(2): 225-231. [4] Srinivasan A, Batur C. Hopfield/ART -1 neural network based fault detection and isolation[J]. IEEE Transactions on Neural Network, 1994, 5(6): 890-899. [5] Bissessur Y, Martin E B, Morris A J. Fault detection in hot steel rolling using neural networks and multivariate statistics [J]. IEE Proceedings Control Theory and Application, 2000, 147(6): 633-640. doi: 10.1049/ip-cta:20000763 [6] 荆双喜. 小波包-神经网络在斜轴泵故障诊断中的应用研究[J]. 振动、测试与诊断, 2000, 20(2): 97—101. https://www.cnki.com.cn/Article/CJFDTOTAL-ZDCS200002004.htmJING Shuang-xi. Fault diagnosis of bend axis piston pump bywavelet packet and neural networks[J]. Journal ofVibration, Measurement & amp; amp; Diagnosis, 2000, 20(2): 97—101. (inChinese). https://www.cnki.com.cn/Article/CJFDTOTAL-ZDCS200002004.htm [7] 吴今培. 智能故障诊断与专家系统[M]. 北京: 科学出版社, 1997. [8] 戴小文. 小波分析与DSP在摆式列车倾摆控制监视系统中的应用研究[D]. 成都: 西南交通大学, 2001. -
点击查看大图
图(1) / 表(4)
计量
- 文章访问数: 193
- HTML全文浏览量: 43
- PDF下载量: 443
- 被引次数: 0
