Page 12 - 应用声学2019年第2期

158 2019 年 3 月

形产生的弹性波，其频带为 100 kHz∼300 kHz。因 [8] Li X. A brief review: acoustic emission method for tool
此，接下来对采集到的 AE 信号进行 db5 小波软阈 wear monitoring during turning[J]. International Journal
of Machine Tools & Manufacture, 2002, 42(2): 157–165.
值降噪。对于经过降噪的 AE 信号进行分帧处理并
[9] Yao Y, Li X, Yuan Z. Tool wear detection with fuzzy clas-
提取每帧信号的 8 个特征，组成 8 维 AE 数据集。最 sification and wavelet fuzzy neural network[J]. Interna-
后使用 AE 数据集作为训练数据集，并通过弱极限 tional Journal of Machine Tools & Manufacture, 1999,
39(10): 1525–1538.
近似采样算法学习HDP-HSMM。其学习结果表明，
[10] Li X, Dong S, Yuan Z. Discrete wavelet transform for tool
该方法能够有效识别砂轮在加工过程中的状态变 breakage monitoring[J]. International Journal of Machine
化并能对砂轮的钝化程度进行分级。其在测试数据 Tools & Manufacture, 1999, 39(12): 1935–1944.
[11] Williams R V. Acoustic emission[J]. Journal of Mechanical
集上具有93.7%的精度，具有很高的工业应用价值。
Design, 1982: 25–27.
HDP-HSMM结合了显式时间分布的半马尔可 [12] 唐英, 顾崇衔, 孙荣平, 等. 金属切削过程声发射机理 [J]. 北
夫模型和贝叶斯非参数技术，不仅可以对有监督和 京科技大学学报, 1995, 17(5): 439–443.
无监督的非马尔可夫时序数据学习，也可以对更大 Tang Ying, Gu Chongxian, Sun Rongping, et al. Acous-
tic emission mechanism in cutting process[J]. Journal of
层次的模型进行推理。通过弱极限近似采样算法可 University of Science and Technology Beijing, 1995, 17(5):
以快速准确地对该模型进行训练。本文已经证明了 439–443.
[13] 刘希强, 杨京, 程建春, 等. 磨削加工过程的声发射机理研
该方法在磨削砂轮不同钝化状态识别中的有效性，
究 [C]. 中国声学学会青年学术会议会议, 2015.
这些方法也将可以为更多连续时间序列问题提供 [14] 张淑清, 上官寒露, 袁计委, 等. 基于内禀模态能量比呼
解决方案。 吸信号特征参数提取方法 [J]. 仪器仪表学报, 2010, 31(8):
1706–1711.
Zhang Shuqing, Shangguan Hanlu, Yuan Jiwei, et al.
参 考 文 献 Study on the extraction method of characteristic parame-
ters of respiration signals based on intrinsic mode energy
[1] Kurada S, Bradley C. A review of machine vision sensors ratio[J]. Chinese Journal of Scientific Instrument, 2010,
for tool condition monitoring[J]. Computers in Industry, 31(8): 1706–1711.
1997, 34(1): 55–72. [15] Wu H, Yu Z, Yan W. Real-time FDM machine condi-
[2] Mokbel A A, Maksoud T M A. Monitoring of the condi- tion monitoring and diagnosis based on acoustic emission
tion of diamond grinding wheels using acoustic emission and hidden semi-Markov model[J]. International Journal
technique[J]. Journal of Materials Processing Technology, of Advanced Manufacturing Technology, 2017, 90(5–8):
2000, 101(1): 292–297. 2027–2036.
[3] Hwang T W, Whitenton E P, Hsu N N, et al. Acous- [16] Xu Z, Xuan J, Shi T, et al. A novel fault diagnosis method
tic emission monitoring of high speed grinding of silicon of bearing based on improved fuzzy ARTMAP and mod-
nitride[J]. Ultrasonics, 2000, 38(1): 614–619. ified distance discriminant technique[J]. Expert Systems
[4] Lezanski P. An intelligent system for grinding wheel with Applications, 2009, 36(9): 11801–11807.
condition monitoring[J]. Journal of Materials Processing [17] Johnson M J, Willsky A S. Bayesian nonparametric hid-
Technology, 2001, 109(3): 258–263. den semi-Markov models[J]. Journal of Machine Learning
[5] Hosokawa A, Mashimo K, Yamada K, et al. Evaluation Research, 2013, 14(1): 673–701.
of grinding wheel surface by means of grinding sound dis- [18] Yu S Z. Hidden semi-Markov models[J]. Artificial Intelli-
crimination[J]. Jsme International Journal, 2004, 47(1): gence, 2010, 174(2): 215–243.
52–58. [19] Teh Y W, Jordan M I, Beal M J, et al. Hierarchical dirich-
[6] Kwak J S, Ha M K. Detection of dressing time using the let processes[J]. Journal of the American Statistical Asso-
grinding force signal based on the discrete wavelet decom- ciation, 2006, 101(476): 1566–1581.
position[J]. International Journal of Advanced Manufac- [20] Fox E B. Bayesian nonparametric learning of complex dy-
turing Technology, 2004, 23(1–2): 87–92. namical phenomena[D]. Cambridge, MA: MIT, 2009.
[7] Liao T W, Ting C F, Qu J, et al. A wavelet-based method- [21] Fox E B, Sudderth E B, Jordan M I, et al. An HDP-HMM
ology for grinding wheel condition monitoring[J]. Interna- for systems with state persistence[C]. 25th International
tional Journal of Machine Tools & Manufacture, 2007, Conference on Machine Learning (ICML) Helsinki, Fabi-
47(3): 580–592. aninkatu, July, 2008: 312–319.