-
摘要: 针对稀疏自动编码器(Sparse auto encoder, SAE)采用sigmoid激活函数容易造成梯度消失的问题, 用一种新的Tan函数替代原有的sigmoid函数; 针对SAE采用Kullback-Leibler (KL) 散度进行稀疏性约束在回归预测方面的局限性, 以dropout机制替代KL散度实现网络的稀疏性. 利用改进SAE对滚动轴承振动信号进行无监督深层特征自适应提取, 无需人工设计标签进行有监督微调. 同时, 考虑到滚动轴承剩余使用寿命(Remaining useful life, RUL)预测方法一般仅考虑过去信息而忽略未来信息, 引入双向长短时记忆网络(Bi-directional long short-term memory, Bi-LSTM)构建滚动轴承RUL的预测模型. 在2个轴承数据集上的实验结果均表明, 所提基于改进SAE和Bi-LSTM的滚动轴承RUL预测方法不仅可以提高模型的收敛速度而且具有较低的预测误差.Abstract: Since the sigmoid activation function of sparse auto-encoder (SAE) is easy to cause the gradient to disappear, a new Tan function is used to replace the original sigmoid function. In SAE, for the limitations in regression prediction when Kullback-Leibler (KL) divergence is used for sparseness constraints, KL divergence is replaced with the dropout mechanism to achieve network sparsity. And the improved SAE is used to perform unsupervised adaptive deep feature extraction for the vibration signals of rolling bearings, without designing labels manually for supervised fine adjustment. Meanwhile, for the remaining useful life (RUL) prediction method of rolling bearing, generally only the past information is considered and the future information is ignored, the bi-directional long short-term memory (Bi-LSTM) is introduced to construct an RUL prediction model of the rolling bearing. Using two bearing data sets, experimental results both show that the proposed RUL prediction method of a rolling bearing based on improved sparse auto encoder and Bi-LSTM can improve the convergence speed of the model and has lower prediction error.
-
表 1 实验数据(PHM2012轴承数据集)
Table 1 Experimental data (PHM2012 bearing datasets)
数据集划分 不同轴承 非全寿数据 (组) 全寿数据 (组) 训练集 1_1 — 2803 1_2 — 871 2_1 — 911 2_2 — 797 3_1 — 515 3_2 — 1637 测试集 1_3 1802 2375 1_4 1139 1428 1_5 2302 2463 1_6 2302 2448 1_7 1502 2259 2_3 1202 1955 2_4 612 751 2_5 2002 2311 2_6 572 701 2_7 172 230 3_3 352 434 表 2 3种优化算法的训练误差
Table 2 Training error of three optimization algorithms
不同优化算法 Adam RMSProp SGDM MSE 0.0008 0.0011 0.0009 MAE 0.0450 0.0794 0.0485 MAPE 0.2292 0.3302 0.3394 MSPE 0.0183 0.0213 0.0406 RMSE 0.0624 0.0985 0.0651 误差之和 0.3557 0.5304 0.4945 表 3 本文预测方法与其他3种方案的构成
Table 3 The composition of the proposed prediction method and other three schemes
预测方法 特征提取模型 预测模型 本文方法 改进 SAE Bi-LSTM 方案 1 SAE Bi-LSTM 方案 2 改进 SAE LSTM 方案 3 SAE LSTM 表 4 不同轴承RUL预测误差结果对比(PHM2012轴承数据集) (%)
Table 4 Comparison of RUL prediction results of different bearings (PHM2012 bearing datasets) (%)
不同轴承 本文方法 方案 1 方案 2 方案 3 文献[20] 文献[21] 1_3 8.03 0.52 0.70 −6.98 43.28 −31.76 1_4 −8.30 0.70 −3.81 2.42 67.55 62.76 1_5 −44.72 −26.09 −45.34 −110.56 −22.98 −136.03 1_6 −2.74 −23.29 21.92 −13.70 21.23 −32.88 1_7 −3.04 7.13 −9.51 −33.03 17.83 −11.09 2_3 −4.12 −20.85 −15.94 −1.73 37.84 44.22 2_4 0.72 −3.60 −0.72 −27.30 −19.42 −55.40 2_5 −6.15 16.83 −38.51 12.62 54.37 68.61 2_6 3.10 −37.21 −13.95 −6.20 −13.95 −51.94 2_7 1.72 −1.72 5.17 −1.72 −55.17 −68.97 3_3 −15.85 2.44 2.44 17.07 3.66 −21.96 平均误差 −6.49 −7.74 −9.81 −15.37 32.48 53.24 平均得分 0.576 0.522 0.477 0.425 0.263 0.065 表 5 实验数据(XJTU-SY轴承数据集)
Table 5 Experimental data (XJTU-SY bearing datasets)
数据集划分 不同轴承 非全寿数据 (组) 全寿数据 (组) 训练集 1_1 — 123 1_2 — 161 2_1 — 491 2_2 — 161 3_1 — 2538 3_2 — 2496 测试集 1_3 126 158 1_4 98 122 1_5 42 52 2_3 426 533 2_4 34 42 2_5 271 339 3_3 297 371 3_4 1212 1515 3_5 91 114 表 6 不同轴承 RUL 预测误差结果对比(XJTU-SY 轴承数据集) (%)
Table 6 Comparison of RUL prediction results of different bearings (XJTU-SY bearing datasets) (%)
不同轴承 本文方法 方案 1 方案 2 方案 3 1_3 15.63 21.88 12.50 18.75 1_4 8.33 −8.33 −4.17 20.83 1_5 −10.00 −30.00 20.00 −10.00 2_3 −24.30 −21.78 15.89 −23.36 2_4 −12.50 −25.00 −25.00 −12.50 2_5 10.29 27.94 14.71 22.06 3_3 31.08 23.78 17.57 22.97 3_4 −18.25 5.83 −50.83 −36.96 3_5 4.35 −26.09 −8.70 −39.13 平均误差 0.51 −3.53 −0.89 −4.15 平均得分 0.419 0.282 0.418 0.267 -
[1] 刘建昌, 权贺, 于霞, 何侃, 李镇华. 基于参数优化VMD和样本熵的滚动轴承故障诊断. 自动化学报, 2022, 48(3): 808−819 doi: 10.16383/j.aas.190345Liu Jian-Chang, Quan He, Yu Xia, He Kan, Li Zhen-Hua. Rolling bearing fault diagnosis based on parameter optimization VMD and sample entropy. Acta Automatica Sinica, 2022, 48(3): 808−81 doi: 10.16383/j.aas.190345 [2] 张正新, 胡昌华, 司小胜, 张伟. 双时间尺度下的设备随机退化建模与剩余寿命预测方法. 自动化学报, 2017, 43(10): 1789-1798Zhang Zheng-Xin, Hu Chang-Hua, Si Xiao-Sheng, Zhang Wei. Degradation modeling and remaining useful life prediction with bivariate time scale. Acta Automatica Sinica, 2017, 43(10): 1789-1798 [3] 赵光权, 刘小勇, 姜泽东, 胡聪. 基于深度学习的轴承健康因子无监督构建方法. 仪器仪表学报, 2018, 39(6): 82-88Zhao Guang-Quan, Liu Xiao-Yong, Jiang Ze-Dong, Hu Cong. Unsupervised health indicator of bearing based on deep learning. Chinese Journal of Scientific Instrument, 2018, 39(6): 82-88 [4] 杨宇, 张娜, 程军圣. 全参数动态学习深度信念网络在滚动轴承寿命预测中的应用. 振动与冲击, 2019, 38(10): 199-205+249Yang Yu, Zhang Na, Cheng Jun-Sheng. Global parameters dynamic learning deep belief networks and its application in rolling bearing life prediction. Journal of Vibration and Shock, 2019, 38(10): 199-205+249 [5] Hinchi A Z, Tkiouat M. Rolling element bearing remaining useful life estimation based on a convolutional long-short-term memory network. Procedia Computer Science, 2018, 127: 123-132 doi: 10.1016/j.procs.2018.01.106 [6] Li X, Zhang W, Ding Q. Deep learning-based remaining useful life estimation of bearings using multi-scale feature extraction. Reliability Engineering and System Safety, 2019, 182: 208-218 doi: 10.1016/j.ress.2018.11.011 [7] Cheriyadat A M. Unsupervised feature learning for aerial scene classification. IEEE Transactions on Geoscience and Remote Sensing, 2014, 52(1): 439-451 doi: 10.1109/TGRS.2013.2241444 [8] 孙文珺, 邵思羽, 严如强. 基于稀疏自动编码深度神经网络的感应电动机故障诊断. 机械工程学报, 2016, 52(9): 65-71 doi: 10.3901/JME.2016.09.065Sun Weng-Jun, Shao Si-Yu, Yan Ru-Qiang. Induction motor fault diagnosis based on deep neural network of sparse auto-encoder. Journal of Mechanical Engineering, 2016, 52(9): 65-71 doi: 10.3901/JME.2016.09.065 [9] 陈宇, 温欣玲, 刘兆瑜, 马鹏阁. 稀疏自动编码器视觉特征融合的多弹分类算法研究. 红外与激光工程, 2018, 47(8): 386-393Chen Yu, Wen Xin-Ling, Liu Zhao-Yu, Ma Peng-Ge. Research of multi-missile classification algorithm based on sparse auto-encoder visual feature fusion. Infrared and Laser Engineering, 2018, 47(8): 386-393 [10] 朱宵珣, 周沛, 苑一鸣, 徐博超, 韩中合. 基于KL-HVD的转子振动故障诊断方法研究. 振动与冲击, 2018, 37(16): 250-255Zhu Xiao-Xun, Zhou Pei, Yuan Yi-Ming, Xu Bo-Chao, Han Zhong-He. A study on the method of rotor vibration fault diagnosis based on KL-HVD. Journal of Vibration and Shock, 2018, 37(16): 250-255 [11] Wang S X, Wang X, Wang S M, Wang D. Bi-directional long short-term memory method based on attention mechanism and rolling update for short-term load forecasting. Electrical Power and Energy Systems, 2019, 109: 470-479 doi: 10.1016/j.ijepes.2019.02.022 [12] Bin Y, Yang Y, Shen F M, Xie N, Shen H T, Li X L. Describing video with attention-based bidirectional LSTM. IEEE Transactions on Cybernetics, 2019, 49(7): 2631-2641 doi: 10.1109/TCYB.2018.2831447 [13] 刘国梁, 余建波. 知识堆叠降噪自编码器. 自动化学报, 2022, 48(3): 774−786 doi: 10.16383/j.aas.c190375Liu Guo-Liang, Yu Jian-Bo. Knowledge-based stacked denoising Autoencoder. Acta Automatica Sinica, 2022, 48(3): 774−786 doi: 10.16383/j.aas.c190375 [14] Srivastava N, Hinton G, Krizhevsky A, Sutskever I, Salakhutdinov R. Dropout: a simple way to prevent neural networks from overfitting. Journal of Machine Learning Research, 2014, 15: 1929-1958 [15] Wu S, Gebraeel N, Lawley M A. A neural network integrated decision support system for condition-based optimal predictive maintenance policy. IEEE Transactions on Systems, Man, and Cybernetics-Part A: Systems and Humans, 2007, 37(2): 226-236. doi: 10.1109/TSMCA.2006.886368 [16] 史加荣, 王丹, 尚凡华, 张鹤于. 随机梯度下降算法研究进展. 自动化学报, 2020: https://doi.org/10.16383/j.aas.c190260 doi: 10.16383/j.aas.c190260Shi Jia-Rong, Wang Dan, Shang Fan-Hua, Zhang He-Yu. Research advances on stochastic gradient descent algorithms. Acta Automatica Sinica, 2020: https://doi.org/10.16383/j.aas.c190260 doi: 10.16383/j.aas.c190260 [17] Nectoux P, Gouriveau R, Medjaher K, Ramasso E, Morello B, Zerhouni N, et al. PRONOSTIA: An experimental platform for bearings accelerated degradation tests. In: Proceedings of the Conference on Prognostics and Health Management. Denver, Colorado, USA: IEEE, 2012. 1−8 [18] 石晓辉, 阳新华, 张向奎, 李文礼. 改进的形态差值滤波器在滚动轴承故障诊断中的应用. 重庆理工大学学报(自然科学), 2018, 32(1): 1-6Shi Xiao-Hui, Yang Xin-Hua, Zhang Xiang-Kui, Li Weng-Li. Application of improved morphological difference filter in fault diagnosis of rolling bearings. Journal of Chongqing University of Technology(Natural Science), 2018, 32(1): 1-6 [19] Singleton R K, Strangas E G, Aviyente S. Extended kalman filtering for remaining-useful-life estimation of bearings. IEEE Transactions on Industrial Electronics, 2015, 62(3): 1781-1790 doi: 10.1109/TIE.2014.2336616 [20] Guo L, Li N P, Jia F, Lei Y G, Lin J. A recurrent neural network based health indicator for remaining useful life prediction of bearings. Neurocomputing, 2017, 240: 98-109 doi: 10.1016/j.neucom.2017.02.045 [21] Sheng H, Zheng Z, Enrico Z, Kan H. Condition assessment for the performance degradation of bearing based on a combinatorial feature extraction method. Digital Signal Processing, 2014, 27: 159-166 doi: 10.1016/j.dsp.2013.12.010 [22] 雷亚国, 韩天宇, 王彪, 李乃鹏, 闫涛, 杨军. XJTU-SY滚动轴承加速寿命试验数据集解读. 机械工程学报, 2019, 55(16): 1-6 doi: 10.3901/JME.2019.16.001Lei Yang-Guo, Han Tian-Yu, Wang Biao, Li Nai-Peng, Yan Tao, Yang Jun. XJTU-SY rolling element bearing accelerated life test datasets: a tutorial. Journal of Mechanical Engineering, 2019, 55(16): 1-6 doi: 10.3901/JME.2019.16.001