盲源分离在机械设备声学信号特征提取中的应用

【作者】 王宇；

【导师】 迟毅林；

【作者基本信息】 昆明理工大学 ， 机械制造及其自动化， 2007， 硕士

【摘要】 在工程信号处理中，由于测量噪声以及其它信号源的干扰，传感器接收到的信号往往是多个信号源的混合信号，从接收到的混合信号中将原始信号分离出来的过程称为盲源分离(BSS)。盲源分离技术已成功地应用到生物医学、地震数据处理、通信等多个领域。本文将盲源分离技术应用到机械设备状态监测与故障诊断中，以线性瞬时盲源分离模型为基础，利用信号自身的不同属性，对设备故障诊断中的声学信号特征提取进行了探讨，研究了多种情况下信号的自适应盲分离算法——快速独立分量分析、稳健的二阶盲辨识、双盲分离方法，分析比较了三种算法的基本原理、适用范围及分离精度性能等，并将得到的理论结果应用到仿真信号处理中，进行大量的仿真试验验证。论文针对盲源分离过程中，信号源数目一般预先无法得知且可能动态变化的问题，研究了相应的盲信号源数目估计算法，在主分量分析方法的基础上，提出了基于盲提取和相关分析的改进算法，并通过计算机仿真实验进行了验证。在理论研究和仿真实验的基础上，在半消声室里分别对两个扬声器产生的混合声信号、一个小型钻机与一台电风扇产生的混合声信号进行盲源分离实验，实验结果表明论文所提出方法的可行性、有效性和正确性。更多还原

【Abstract】 In signal processing of engineering, the signals received by sensors are frequently mixed signals which come from many signal sources due to the signals from other sources together with additive noise. The Process of recovering original signals from mixed signals is called blind source separation(BSS) which has been successfully used in the fields of acoustic signal processing, biomedicine, earthquake date processing and communications.In this paper, the technology of blind source separation is applied to state monitoring and fault diagnosis of equipments. Based on the model of linear instantaneous blind source separation, the feature extraction of acoustic signals of equipments using the own varied property of signals is discussed. Three algorithms including Fast Independent Component Analysis(FastICA), Robust Second Blind Identification and Double Blind Separation are investigated and compared in various conditions, the results of theoretical analysis is applied to the simulation of signal processing, a large number of simulation experiments are carried out to verify the legitimacy of algorithms that discussed above.In the process of blind source separation, the number of signal source usually is unknown and it also may be variable dynamically, so the related estimation algorithms of the number of blind signal source are investigated. Based on traditional method of principal component analysis, a new algorithm on the basis of blind source extraction and correlation analysis is proposed, the validity of the algorithm is proved by lots of computer simulation experiments.Based on theoretical analysis and simulation experiments, the experiments of blind source separation were conducted in a half-anechoic room, the mixed noise was respectively generated by two loudhailers, a small drill and a fanner. Results of experiments show the validity of the methods proposed in the paper.更多还原