II 摘摘 要要 噪声污染是目前世界上主要的污染源之一。强烈的噪声不仅危害人体健康，干扰语言交流，妨碍生产任务的执行，甚至会影响机器产品的使用性能和寿命。本文的研究目的在于通过对噪声源识别技术方法的研究，识别复杂机器系统的主噪声源，为进一步实施降噪提供依据。 传统的噪声源识别方法有很多，各有不同的限制条件和应用场合。一个好的噪声源识别方法，不但要能准确地识别主噪声源，而且要能有效可靠地工作于复杂的生产现场。在实际的现场环境中， 经常面临多台机器同时发声或一台机器上同时存在多个噪声源的情形，而且存在着强烈的外来噪声干扰，此时仅仅使用传统的噪声源识别技术进行声源识别往往难以得到好的效果。本文联合应用独立分量分析（ICA） 、小波变换和高阶谱分析方法，进行噪声源的分析与识别。首先，利用盲源分离（BSS）方法，对传感器所测取的混合观测信号实施源分离，获得对系统本底源信号的估计；利用 ICA 方法，估计从源到传感器的频率响应特性，进而实现声源波达方向（DOA）的估计；然后，通过小波变换和高阶谱分析，对已分离的独立源信号进行特征分析与提取；最后，实现主噪声源的识别。 在对噪声源识别理论、方法与技术进行系统研究的基础上，搭建了用于新方法验证的数据采集系统，并独立开发了基于新方法的噪声源识别软件。仿真与实验结果表明：新的噪声源识别方法具有一定的可行性和有效性。但是为提高它的实际应用价值，方法本身还有待进一步的完善。 关键词关键词：独立分量分析;波达方向估计;主噪声源识别;小波变换;高阶谱分析 III Abstract Today, noise pollution has become one of the main pollution sources in the world. Strong noise not only hurts peoples health, but also disturbs oral communication and production procedure. Furthermore, it may deteriorate working performance of a machine, and shorten its working lifetime. The research aim of this paper is at recognizing the dominant acoustical sources in a complex machinery system by studying some techniques for acoustical sources recognition. Thus, practical evidences for further noise reduction are provided. There are many traditional methods for acoustical sources identification, whose applicable conditions and cases are different each other. A good method for acoustical sources identification not only can identify the dominant acoustical sources correctly, but also can reliably and effectively work in a complex worksite environment where exist such cases as multiple machines simultaneously emitting sounds, multiple parts in a machine simultaneously emitting sounds and/or strong noise interference. Under such condition, it may be difficult to correctly recognize acoustical sources only using these traditional methods. In this paper, we implement analysis and recognition of acoustical sources by the combined use of three techniques, i.e. independent component analysis (ICA), wavelet transformation (WT) and higher-order spectrum (HOS) analysis. Firstly, we use blind source separation (BSS) method to separate the underlying sources embedded in mixtures measured by sensors; we also use ICA to estimate frequency response from sources to sensors, thus direction-of-arrival (DOA) of multiple sources. Then, source features are analyzed and extracted by the use of WT and HOS. Finally, the dominant acoustical sources are identified. Based on systemic study of theories, methods and techniques for acoustical sources recognition, an experimental system for data acquisition is constructed, and software for acoustical sources identification is developed, both of them are based on the proposed method in this paper. Experiment results show that the proposed method is feasible and effective to some extent. Indeed, it is necessary to make further improvement on it, in order to enforce its applicability. Keyword: Independent component analysis; Direction of arrival (DOA); Dominant acoustical sources identification; Wavelet transformation (WT); Higher-order spectrum (HOS) IV 目 录 目 录 摘 要. Abstract.III 第一章 绪 论.1 1.1 论文研究的背景和意义.1 1.2 国内外研究现状和发展趋势.1 1.3 本文的研究内容.3 1.4 本章小结.5 第二章 噪声源识别理论与分析方法.5 2.1 噪声的性质与度量.5 2.1.1 噪声的性质.5 2.1.2 噪声的物理度量.5 2.1.3 噪声的主观评价.6 2.2 基于ICA的声源波达方向估计.9 2.2.1 传统的波达方向估计.9 2.2.2 独立分量分析理论.11 2.2.3 基于独立分量分析的波达方向估计.14 2.3 主噪声源