Bearing Fault Diagnosis Using Deep CNN and LSTM
-
摘要: 针对传统数据驱动故障诊断方法难以从轴承信号中自适应提取有效特征、没有充分利用故障数据的时序特点以及缺乏自适应处理动态信息能力的问题, 提出了一种深度卷积神经网络与长短期记忆网络相结合的智能故障诊断方法。本文方法构建的深度模型能够从轴承原始信号中自适应地提取鲁棒性特征, 然后利用长短期记忆网络学习特征中的时间依赖关系实现了高准确度的轴承故障诊断。该方法克服了传统特征提取方法依赖专家经验和信息利用不完全等问题, 实现了故障的智能、准确诊断。实验结果表明, 该方法可以提取更准确的特征而且由于利用了故障演变过程中的时序信息, 使得故障诊断更加智能、可靠。Abstract: There exists some problems in traditional data-driven fault diagnosis methods, for example, it is difficult to adaptively extract effective features from bearing signals, cannot make full use of the timing characteristics of fault data, and lacks of the ability to adaptively process dynamic information. An intelligent bearing fault diagnosis method combined deep convolutional neural network and long-term and short-term memory network is proposed. This method constructed a kind of deep networks and can adaptively extract the robust features from the original bearing signals, and then utilized the long short-term memory network to learn the time-dependent relationship in these features, and achieved high-accuracy bearing fault diagnosis. The proposed method overcame the problems existed in the traditional feature extraction methods, such as heavy dependence on expert experience and incomplete utilization for time series information, and realized intelligent and accurate diagnosis of faults. The experimental results show that the proposed method can extract more accurate features and make the fault diagnosis more intelligent and reliable by utilizing the timing information in the process of fault degradation
-
表 1 振动信号数据集
名称 负载 采样点数 A 0 2 400 B 1 2 400 C 2 2 400 D 3 2 400 表 2 模型的主要参数选择
网络参数 设置值 训练参数 CNN层数 5 学习率为8×10-4 池化移动步长 2 训练集比率为80% 全连接层 2 最大轮数为100 LSTM层数 3 Batch_size为32 时间步长 20 SGD 表 3 不同CNN的结构及其诊断结果
CNN1 CNN2 CNN3 CNN4 CNN5 L1 Conv-Maxpool Conv-Maxpool Conv-Maxpool L2 Conv-Maxpool Conv-Maxpool Conv-Maxpool L3 Conv-Maxpool Conv-Maxpool Conv-Maxpool L4 - Conv-Maxpool Conv-Maxpool L5 - - Conv-Maxpool FC1 1 024 1 024 1 024 1 024 1 024 FC2 1 024 1 024 1 024 1 024 512 FC3 - - 512 1024 - Acc/(%) 91.79 99.33 98.04 95.49 99.96 表 4 不同方法10次实验平均诊断结果
方法 平均准确率/% 准确率的标准差/% 本文提出方法 99.961 0.59 常用统计特征+SVM方法 72.692 1.17 BP方法 53.085 2.47 表 5 深度学习方法平均诊断结果
网络结构 准确率/% 标准方差/% DCLSTM 99.96 0.590 2 单一CNN 95.35 1.162 6 表 6 不同方法的F1分数
类别 本文提出方法 CNN方法 准确率/% 召回率/% F1分数 准确率/% 召回率/% F1分数 1 100 100 100 98 85 91 2 100 100 100 89 99 93 3 100 96 98 99 100 98 4 97 100 98 89 100 99 表 7 不同负载下不同方法的比较结果
方法 0 1 hp 2 hp 3 hp SVM 70.792 72.870 73.592 72.201 CNN 94.238 93.828 91.602 97.598 DCLSTM 99.883 99.805 99.219 99.688 -
[1] LI Y F, LIANG X H, ZUO M J. Diagonal slice spectrum assisted optimal scale morphological filter for rolling element bearing fault diagnosis[J]. Mechanical Systems and Signal Processing, 2017, 85: 146-161 doi: 10.1016/j.ymssp.2016.08.019 [2] ZHANG Z Y, WANG Y, WANG K S. Fault diagnosis and prognosis using wavelet packet decomposition, Fourier transform and artificial neural network[J]. Journal of Intelligent Manufacturing, 2013, 24(6): 1213-1227 doi: 10.1007/s10845-012-0657-2 [3] 马伦, 康建设, 孟妍, 等. 基于Morlet小波变换的滚动轴承早期故障特征提取研究[J]. 仪器仪表学报, 2013, 34(4): 920-926 doi: 10.3969/j.issn.0254-3087.2013.04.031MA L, KANG J S, MENG Y, et al. Research on feature extraction of rolling bearing incipient fault based on Morlet wavelet transform[J]. Chinese Journal of Scientific Instrument, 2013, 34(4): 920-926 (in Chinese) doi: 10.3969/j.issn.0254-3087.2013.04.031 [4] 陈彦龙, 张培林, 徐超, 等. 基于DCT和EMD的滚动轴承故障诊断[J]. 电子测量技术, 2012, 35(2): 121-125 doi: 10.3969/j.issn.1002-7300.2012.02.030CHEN Y L, ZHANG P L, XU C, et al. Fault diagnosis of rolling bearing based on DCT and EMD[J]. Electronic Measurement Technology, 2012, 35(2): 121-125 (in Chinese) doi: 10.3969/j.issn.1002-7300.2012.02.030 [5] 熊景鸣, 潘林, 朱昇, 等. DBN与PSO-SVM的滚动轴承故障诊断[J]. 机械科学与技术, 2019, 38(11): 1726-1731 doi: 10.13433/j.cnki.1003-8728.20190040XIONG J M, PAN L, ZHU S, et al. Bearing fault diagnosis based on deep belief networks and particle swarm optimization support vector machine[J]. Mechanical Science and Technology for Aerospace Engineering, 2019, 38(11): 1726-1731 (in Chinese) doi: 10.13433/j.cnki.1003-8728.20190040 [6] LEI Y G, JIA F, LIN J, et al. An intelligent fault diagnosis method using unsupervised feature learning towards mechanical big data[J]. IEEE Transactions on Industrial Electronics, 2016, 63(5): 3137-3147 doi: 10.1109/TIE.2016.2519325 [7] AHARON M, ELAD M, BRUCKSTEIN A. K-SVD: an algorithm for designing overcomplete dictionaries for sparse representation[J]. IEEE Transactions on Signal Processing, 2006, 54(11): 4311-4322 doi: 10.1109/TSP.2006.881199 [8] 蒋永华, 程光明, 阚君武, 等. 基于NGA优化SVM的滚动轴承故障诊断[J]. 仪器仪表学报, 2013, 34(12): 2684-2689 https://www.cnki.com.cn/Article/CJFDTOTAL-YQXB201312007.htmJIANG Y H, CHENG G M, KAN J W, et al. Rolling bearing fault diagnosis based on NGA optimized SVM[J]. Chinese Journal of Scientific Instrument, 2013, 34(12): 2684-2689 (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-YQXB201312007.htm [9] DENG S, JING B, SHENG S, et al. Impulse feature extraction method for machinery fault detection using fusion sparse coding and online dictionary learning[J]. Chinese Journal of Aeronautics, 2015, 28(2): 488-498 doi: 10.1016/j.cja.2015.01.002 [10] 雷亚国, 贾峰, 周昕, 等. 基于深度学习理论的机械装备大数据健康监测方法[J]. 机械工程学报, 2015, 51(21): 49-56 https://www.cnki.com.cn/Article/CJFDTOTAL-JXXB201521007.htmLEI Y G, JIA F, ZHOU X, et al. A deep learning-based method for machinery health monitoring with big data[J]. Journal of Mechanical Engineering, 2015, 51(21): 49-56 (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-JXXB201521007.htm [11] SHAO H D, JIANG H K, LIN Y, et al. A novel method for intelligent fault diagnosis of rolling bearings using ensemble deep auto-encoders[J]. Mechanical Systems and Signal Processing, 2018, 102: 278-297 doi: 10.1016/j.ymssp.2017.09.026 [12] 温江涛, 闫常弘, 孙洁娣, 等. 基于压缩采集与深度学习的轴承故障诊断方法[J]. 仪器仪表学报, 2018, 39(1): 171-179 https://www.cnki.com.cn/Article/CJFDTOTAL-YQXB201801021.htmWEN J T, YAN C H, SUN J D, et al. Bearing fault diagnosis method based on compressed acquisition and deep learning[J]. Chinese Journal of Scientific Instrument, 2018, 39(1): 171-179 (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-YQXB201801021.htm [13] INCE T, KIRANYAZ S, EREN L, et al. Real-time motor fault detection by 1-D convolutional neural networks[J]. IEEE Transactions on Industrial Electronics, 2016, 63(11): 7067-7075 doi: 10.1109/TIE.2016.2582729 [14] JING L Y, ZHAO M, LI P, et al. A convolutional neural network based feature learning and fault diagnosis method for the condition monitoring of gearbox[J]. Measurement, 2017, 111: 1-10 doi: 10.1016/j.measurement.2017.07.017 [15] 曲建岭, 余路, 袁涛, 等. 基于一维卷积神经网络的滚动轴承自适应故障诊断算法[J]. 仪器仪表学报, 2018, 39(7): 134-143QU J L, YU L, YUAN T, et al. Adaptive fault diagnosis algorithm for rolling bearings based on one-dimensional convolutional neural network[J]. Chinese Journal of Scientific Instrument, 2018, 39(7): 134-143 (in Chinese) [16] LU C, WANG Z Y, ZHOU B. Intelligent fault diagnosis of rolling bearing using hierarchical convolutional network based health state classification[J]. Advanced Engineering Informatics, 2017, 32: 139-151 doi: 10.1016/j.aei.2017.02.005 [17] YANG B L, SUN S L, LI J Y, et al. Traffic flow prediction using LSTM with feature enhancement[J]. Neurocomputing, 2019, 332: 320-327 doi: 10.1016/j.neucom.2018.12.016 [18] LU W N, LI Y P, CHENG Y, et al. Early fault detection approach with deep architectures[J]. IEEE Transactions on Instrumentation and Measurement, 2018, 67(7): 1679-1689 doi: 10.1109/TIM.2018.2800978 [19] LEI J H, LIU C, JIANG D X. Fault diagnosis of wind turbine based on Long Short-term memory networks[J]. Renewable Energy, 2019, 133: 422-432 doi: 10.1016/j.renene.2018.10.031 [20] LECUN Y, BENGIO Y, HINTON G. Deep learning[J]. Nature, 2015, 521(7553): 436-444 doi: 10.1038/nature14539 [21] BENGIO Y. Learning deep architectures for AI[M]. Hanover: The Essence of Knowledge, 2009: 1-127 [22] HOCHREITER S, SCHMIDHUBER J. Long short-term memory[J]. Neural Computation, 1997, 9(8): 1735-1780 [23] SMITH W A, RANDALL R B. Rolling element bearing diagnostics using the case western reserve university data: a benchmark study[J]. Mechanical Systems and Signal Processing, 2015, 64-65: 100-131 doi: 10.1016/j.ymssp.2015.04.021