Research on Voiceprint Model of Pipeline Blockage Recognition Under Variable Working Conditions
-
摘要: 针对排水管道堵塞检测实际应用中训练样本所包含的工况类别受限导致新工况识别率低下的问题,提出一种基于精细复合多尺度散布熵(Refined composite multi-scale dispersion entropy,RCMDE)和高斯混合隐马尔可夫模型(Gaussian-mixed-model hidden Markov model,GMM-HMM)的管道堵塞声纹识别模型。首先,采用基于子带谱熵的单参数双门限端点检测算法对单一和复杂工况下降噪后整体声压信号进行端点检测和信号分割,得到对应管道内堵塞物、三通件和管道尾端的个体声压信号。然后,提取精细复合多尺度散布熵作为特征向量。最后,将单一工况下不同类别的声压信号的特征向量用于模型训练,训练好的模型用于复杂工况下的堵塞物、三通件以及管道尾端的识别。结果表明:所提出的声纹识别模型在训练样本工况类别受限的条件下能有效识别变工况下排水管道中的堵塞物,三通件以及管道尾端,综合识别率为93.75%。验证在不同工况下堵塞物对声波的影响具有共性,与三通件、管道尾端具有差异性,具有一定的工程应用价值。Abstract: In order to solve the problem that the recognition rate of new working conditions is low due to the limited working conditions category contained in the training samples in the practical application of drainage pipe blockage detection, a new voice print recognition model based on Refined Composite Multi-scale Dispersion Entropy and Gaussian Mixture Hidden Markov Model is proposed. Firstly, this method uses single-parameter dual-threshold endpoint detection algorithm based on sub-band spectral entropy to perform endpoint detection and signal segmentation for the sound pressure signal after noise reduction in single and complicated working conditions, and obtain sound pressure signals corresponding to blockage, lateral connection and pipe end in the pipeline. Then the refined composite multi-scale dispersion entropy features are extracted. Finally, the feature vectors of different types of individual sound pressure signals in a single working condition are used for model training, and the trained parameter model used for complex working conditions recognition of blockages, lateral connection and pipe end. The experimental results have shown that the proposed method can effectively identify blockages, pipe fittings such as lateral connection and pipe end under the condition of limited training sample categories, the comprehensive recognition rate is 93.75%, which verified that the voiceprint of the blockage has commonality under different working conditions, and it is different from lateral connection and pipe end and has certain engineering application value.
-
表 1 模型识别正确率对比
信号类型 测试样本数量 MSE-GMM-HMM/% MDE-GMM-HMM/% RCMDE-GMM-HMM 55 mm 堵塞物 20 75 80 90 40 mm堵塞物 20 80 85 90 三通件 20 80 80 95 管道尾端 20 100 100 100 平均识别率 20 83.75 86.25 93.75 -
[1] HAO T, ROGERS C D F, METJE N, et al. Condition assessment of the buried utility service infrastructure[J]. Tunnelling and Underground Space Technology, 2012, 28: 331-344. doi: 10.1016/j.tust.2011.10.011 [2] DATTA S, SARKAR S. A review on different pipeline fault detection methods[J]. Journal of Loss Prevention in the Process Industries, 2016, 41: 97-106. doi: 10.1016/j.jlp.2016.03.010 [3] 伍林峰, 冯早, 黄国勇, 等. 小波包增强稀疏表征分类的管道堵塞故障识别[J]. 电子测量与仪器学报, 2019, 33(3): 35-43. doi: 10.13382/j.jemi.B1801870WU L F, FENG Z, HUANG G Y, et al. Pipeline jam fault identification based on wavelet packet enhanced sparse representation classification[J]. Journal of Electronic Measurement and Instrumentation, 2019, 33(3): 35-43. (in Chinese) doi: 10.13382/j.jemi.B1801870 [4] 李洋, 冯早, 黄国勇, 等. 基于广义Fisher-互信息的管道堵塞故障特征选择方法[J]. 电子测量与仪器学报, 2018, 32(11): 1-8. doi: 10.13382/j.jemi.2018.11.001LI Y, FENG Z, HUANG G Y, et al. Feature selection method for pipe blockage based on generalized Fisher-mutual information[J]. Journal of Electronic Measurement and Instrumentation, 2018, 32(11): 1-8. (in Chinese) doi: 10.13382/j.jemi.2018.11.001 [5] 黎煊, 赵建, 高云, 等. 基于连续语音识别技术的猪连续咳嗽声识别[J]. 农业工程学报, 2019, 35(6): 174-180. doi: 10.11975/j.issn.1002-6819.2019.06.021LI X, ZHAO J, GAO Y, et al. Pig continuous cough sound recognition based on continuous speech recognition technology[J]. Transactions of the Chinese Society of Agricultural Engineering, 2019, 35(6): 174-180. (in Chinese) doi: 10.11975/j.issn.1002-6819.2019.06.021 [6] 王丰华, 王邵菁, 陈颂, 等. 基于改进MFCC和VQ的变压器声纹识别模型[J]. 中国电机工程学报, 2017, 37(5): 1535-1542. doi: 10.13334/j.0258-8013.pcsee.152581WANG F H, WANG S J, CHEN S, et al. Voiceprint recognition model of power transformers based on improved MFCC and VQ[J]. Proceedings of the CSEE, 2017, 37(5): 1535-1542. (in Chinese) doi: 10.13334/j.0258-8013.pcsee.152581 [7] 王斐, 房立清, 齐子元. 基于多尺度样本熵和VPMCD的自动机故障诊断[J]. 振动、测试与诊断, 2018, 38(3): 564-569. doi: 10.16450/j.cnki.issn.1004-6801.2018.03.020WANG F, FANG L Q, QI Z Y. Automaton fault diagnosis based on multiscale sample entropy and VPMCD[J]. Journal of Vibration, Measurement & Diagnosis, 2018, 38(3): 564-569. (in Chinese) doi: 10.16450/j.cnki.issn.1004-6801.2018.03.020 [8] 程珩, 励文艳, 权龙, 等. 基于VMD-MDE和ELM的柱塞泵微弱故障诊断[J]. 振动、测试与诊断, 2020, 40(4): 635-642. doi: 10.16450/j.cnki.issn.1004-6801.2020.04.001CHEN Y, LI W Y, QUAN L, et al. Weak fault diagnosis of axial piston pump based on VMD-MDE and ELM[J]. Journal of Vibration, Measurement & Diagnosis, 2020, 40(4): 635-642. (in Chinese) doi: 10.16450/j.cnki.issn.1004-6801.2020.04.001 [9] AZAMI H, ROSTAGHI M, ABÁSOLO D, et al. Refined composite multiscale dispersion entropy and its application to biomedical signals[J]. IEEE Transactions on Biomedical Engineering, 2017, 64(12): 2872-2879. doi: 10.1109/TBME.2017.2679136 [10] 尚永爽, 许爱强, 吴忠德. 基于SOFM神经网络和HMM的动调陀螺仪故障预测方法研究[J]. 机械科学与技术, 2012, 31(10): 1711-1715. doi: 10.13433/j.cnki.1003-8728.2012.10.026SHANG Y S, XU A Q, WU Z D. A method for predicting faults of dynamically tuned gyroscope using SOFM neural network and hidden Markov model[J]. Mechanical Science and Technology for Aerospace Engineering, 2012, 31(10): 1711-1715. (in Chinese) doi: 10.13433/j.cnki.1003-8728.2012.10.026 [11] 肖熙, 徐晨. 基于声学状态似然值得分模型及监督状态模型的语音识别特征融合算法[J]. 清华大学学报(自然科学版), 2019, 59(6): 476-481. doi: 10.16511/j.cnki.qhdxxb.2019.21.011XIAO X, XU C. Speech feature fusion algorithm based on acoustic state likelihood and supervised state modelling[J]. Journal of Tsinghua University (Science and Technology), 2019, 59(6): 476-481. (in Chinese) doi: 10.16511/j.cnki.qhdxxb.2019.21.011 [12] 刘华平, 李昕, 郑宇, 等. 一种改进的自适应子带谱熵语音端点检测方法[J]. 系统仿真学报, 2008, 20(5): 1366-1371. doi: 10.16182/j.cnki.joss.2008.05.027LIU H P, LI X, ZHENG Y, et al. Speech endpoint detection based on improved adaptive band-partitioning spectral entropy[J]. Journal of System Simulation, 2008, 20(5): 1366-1371. (in Chinese) doi: 10.16182/j.cnki.joss.2008.05.027 [13] 李晔, 张仁智, 崔慧娟, 等. 低信噪比下基于谱熵的语音端点检测算法[J]. 清华大学学报(自然科学版), 2005, 45(10): 1397-1440. doi: 10.3321/j.issn:1000-0054.2005.10.027LI Y, ZHANG R Z, CUI H J, et al. Voice activity detection algorithm with low signal-to-noise ratios based on the spectrum entropy[J]. Journal of Tsinghua University (Science & Technology), 2005, 45(10): 1397-1440. (in Chinese) doi: 10.3321/j.issn:1000-0054.2005.10.027 [14] WU B F, WANG K C. Robust endpoint detection algorithm based on the adaptive band-partitioning spectral entropy in adverse environments[J]. IEEE Transactions on Speech and Audio Processing, 2005, 13(5): 762-775. doi: 10.1109/TSA.2005.851909 [15] ROSTAGHI M, AZAMI H. Dispersion Entropy: A measure for time-series analysis[J]. IEEE Signal Processing Letters, 2016, 23(5): 610-614. doi: 10.1109/LSP.2016.2542881 [16] 王永鑫, 贾珈, 张雨辰, 等. 基于HMM语音合成的语调控制[J]. 清华大学学报(自然科学版), 2013, 53(6): 781-786. doi: 10.16511/j.cnki.qhdxxb.2013.06.010WANG Y X, JIA J, ZHANG Y C, et al. Control of intonation in HMM based text-to-speech systems[J]. Journal of Tsinghua University (Science and Technology), 2013, 53(6): 781-786. (in Chinese) doi: 10.16511/j.cnki.qhdxxb.2013.06.010 [17] 刘嘉敏, 彭玲, 刘军委, 等. 遗传算法VMD参数优化与小波阈值轴承振动信号去噪分析[J]. 机械科学与技术, 2017, 36(11): 1695-1700. doi: 10.13433/j.cnki.1003-8728.2017.1110LIU J M, PENG L, LIU J W, et al. Denoising analysis of bearing vibration signal based on genetic algorithm and wavelet threshold VMD[J]. Mechanical Science and Technology for Aerospace Engineering, 2017, 36(11): 1695-1700. (in Chinese) doi: 10.13433/j.cnki.1003-8728.2017.1110 [18] 李从志, 郑近德, 潘海洋, 等. 基于精细复合多尺度散布熵与支持向量机的滚动轴承故障诊断方法[J]. 中国机械工程, 2019, 30(14): 1713-1719.LI C Z, ZHENG J D, PAN H Y, et al. Fault diagnosis method of rolling bearings based on refined composite multiscale dispersion entropy and support vector machine[J]. China Mechanical Engineering, 2019, 30(14): 1713-1719. (in Chinese)