汽车制动异响频域盲提取模型及关键算法研究

Abnormal Noise source feature identification is a very important part in vehicle brake noise control, however complex acoustic environment and multi-source interference in practical road test will affect the application of related signal processing algorithms seriously. This project, which bases on the previous research results, attempt to establish and improve the frequency domain blind extraction model aiming at brake abnormal noise sources. Thus, key theories and algorithms are further studied. The overall research ideas are: 1. Multi-scale morphological filter which built based on dynamic particle swarm are applied to weaken the background noises; the source contribution rate are studied based on artificial fish swarm optimization algorithm to make the key sound features more obviously, then to decrease the affect brought by large number of sound sources and complex acoustic field. 2. Then, the equivalence relation between brake noise source extract and blind deconvolution is established, and the complex blind source separation algorithm based on reference signal and nonlinear function is studied; after that similarity measure between frequency bands' separation matrix and complex signal is used to solve the permutation problem, the row vector norm normalization and other methods are applied to solve the amplitude uncertainty; which in order to locate and study the contribution rate of each noise source. 3. Finally, take vehicle brakes for research object, study brake abnormal noise blind extraction under all kinds of experimental conditions, then separate and evaluate the features and magnitude order of main noise sources, provide scientific basis for vehicle vibration and noise reduction and foundation brake performance improvements.

异响噪声源特征识别是汽车制动噪声治理中极为重要的环节，但实际路试声场环境复杂、多源干扰等特性严重制约相关信号处理算法的应用。本项目基于前期研究成果，建立完善针对制动异响噪声的频域盲提取模型，进而深入研究其关键理论及算法。其总体研究思路是：①基于动态粒子群算法构建适用的多尺度形态学滤波器，最大程度削弱背景噪声干扰；基于人工鱼群寻优算法研究声源贡献率，优化声源数目，降低复杂声场及大数目声源对算法分离精度的影响。②建立制动噪声源提取与盲解卷积之间的等价关系，研究基于参考信号及非线性函数的改进复数盲分离算法；基于频段间分离矩阵及复数信号的相似测度解决次序不确定性问题，基于行向量范数归一等方法解决幅值不确定性问题，以期准确量化各噪声源贡献率。③以汽车制动器为研究对象，在各种实验条件下进行制动异响盲提取研究，分离并评估主要汽车制动噪声源特性和量级，为汽车减振降噪及基础制动器性能改进提供相应的科学依据。

异响噪声源特征识别是汽车制动噪声治理中极为重要的环节，但实际路试声场环境复杂、多源干扰等特性严重制约相关信号处理算法的应用。本项目主要建立和完善针对制动异响噪声的频域盲提取模型，进而深入研究其关键理论及算法：①构建基于人工鱼群寻优的改进多尺度形态滤波器，以最大程度削弱背景噪声干扰；同时辅以级联结构元素级为不同尺寸的广义形态滤波器，利用归一化峭度估计源信号数目，优化声源数目，降低复杂声场及大数目声源对算法分离精度的影响。②基于模拟退火算法改进模糊C-均值聚类，建立压缩感知与欠定盲源分离等价关系，进而使用压缩感知重构算法的正交匹配追踪算法重构源信号; 采用包络谱余弦测度作为独立分量间距离测度，结合冲击信号峭度指标优选独立分量，进而通过频域稀疏分量分析对估计信号做进一步的分离，形成结合改进块模型算法及频域稀疏分量分析的复合声信号特征盲提取方法。③通过匹配关键零部件结构参数，基于先验知识构造出理论上的参考信号。以此参考信号为约束，通过将观测信号与构造参考信号进行相似性度量，进而提取有限数量的估计信号；基于频段间分离矩阵改进KL距离解决次序不确定性问题，基于行向量范数归一等方法解决幅值不确定性问题，最终形成适用于制动噪声特征的频域盲提取算法，以准确量化各噪声源贡献率。④在各种实验条件下进行制动异响盲提取研究，分离并评估主要汽车制动噪声源特性和量级。所研发的测试分析系统分别在万向钱潮上海和深圳比亚迪汽车的路试场地进行不同工况下的制动试验，分近、远场进行噪声数据拾取，随后对声信号进行分析，为汽车减振降噪及基础制动器性能改进提供相应的科学依据，基本实现了理论研究和实际应用的结合。通过三年的研究，依托本项目共计发表学术论文11篇、授权专利6项、登记软著5项,已按时按量完成项目既定目标。