【作者】 宋雷鸣；

【导师】 孙守光；

【作者基本信息】 北京交通大学 ， 车辆工程， 2010， 博士

【摘要】 波束形成是阵列信号处理的重要发展方向,是近年来发展起来的研究噪声源的一种重要方法,它基于传声器阵列的指向性原理,可以进行声源空间分布的测量,进行声源识别及空间声场特性的研究,实现空间声场重构,目前广泛用于航空航天、汽车工业、铁路列车、汽车、声呐、地震等领域。本文在波束形成技术研究的基础上,提出了一种可实现阵列小型化的基于波速形成的逆向声源识别方法,对该方法进行了理论与工程应用的研究,进行了试验验证,表明该方法可以有效提高测试系统的精度,特别是低频部分的测试精度。该测试技术与其他的高分辨率的阵列测试算法相比,其不需要先验知识,同时阵列系统的阵元数目,可以小于声源数目。逆向声源识别测试技术的实现是基于第一类算子方程的求解实现的,为由果求因的反问题,其通常是病态的：解不存在、解不唯一或不连续依赖数据,目前尚没有成熟统一的求解方法。为了抑制病态,数学上一般借助于正则化方法和优化控制理论。本文根据Tikhonov正则法,应用神经网络、L-曲线法两种方法,对基于波束形成算子内核的第一类积分方程进行了研究,在求解过程中发现上述两种算法或需要先验知识,或网络参数的确定需要优化计算,很难在本课题的工程应用中实现。在上述研究的基础上,以广义奇异值分解为工具,以正则矩阵的选取和正则参数的确定为主线,针对现有算法多集中于正则参数的确定,对于正则矩阵的选择,除利用先验知识外,缺少有效手段的情况,本文提出了一种选择正则矩阵的新方法,进一步发展了Tikhonov正则化方法,并成功地将算法应用到波速形成逆向声源识别线阵和面阵的工程化模型求解中,得到了良好的结果。最后,本文进行了工程应用的试验验证和汽车的噪声源识别,从测试结果可以看出,测试分析与实际情况相符,并优于现有的商业测试系统。更多还原

【Abstract】 Beam forming is an important direction of array signal processing research in recent years, developed into an important method of noise sources, based on the microphone array direction principle。It can measure the spatial distribution of sound source to carry out sound source identification and the study of spatial sound field characteristics and spatial sound field reconstruction, now widely used in aerospace technology, automotive, trains, automobiles, sonar, seismic and other fields. In this paper, an reverse sound source identification method is presented on the basis of beam-forming technology so that a smaller array can be realized. The reverse sound source identification method is researched on the theory and engineering application. Experimental verification shows that this method can effectively improve the accuracy of test system, especially the low-frequency part. Compared with other high resolution’s array test technology, this test technology does not need the apriori knowledge, and simultaneously the array system’s array element number may be smaller than the acoustic source number.The realization of this test technique is based on first kind of integral operator equations to achieve. The solution of first kind of operator equations usually does not exist, or is not unique or non-continuous data-dependent, yet a unified solution method is not mature. Ill-posed state can be suppressed on mathematics, generally by means of regularization methods and optimal control theory. Through the Tikhonov regularization method, neural network and L-curve methods are applicated in order that first-kind integral equations have been studied based on beamforming operator kernel. The priori knowledge or network parameters to optimize are need in order to solve first kind of operator equations through the above-mentioned two kinds of algorithms, but it is difficult in engineering application.In Tikhonov regularization method, the determination of the regularization parameter and regular matrix are need. In the paper, matrix parameters iteration is used for solving first kind of operator equations and a priori knowledge is not used, so this is the further development of the Tikhonov regularization method, and the algorithm is applied successfully into the reverse acoustic source recognition.Finally, the paper carried out experimental verification of engineering applications and automotive noise source identification. Compared, the testing results is agreed with the actual noise sources situation, and is better than existing commercial test systems.更多还原