基于三维声强探头阵列和波束形成的远场声源识别方法研究

Noise is an important representation of the operating status of mechanical equipment. However, in the real environment where multiple sound sources are mixed, the source intensity, frequency spectrum, and waveform of the sound source are difficult to identify or separate accurately, which hinders the application of noise monitoring. The research project aims to solve this issue by: 1) Sound source intensity and position solving model based on the three-dimensional intensity probe array: designing sound intensity probe arrays, and constructing a set of nonlinear equations for solving sound source parameters based on the vector synthesis principle of sound intensity. Direct recognition of sound source intensity and position by numerical solution can be achieved. 2) Optimization of source parameter solving model based on stereo microphone array beamforming: Sound intensity array is reorganized into a stereo microphone array, and beamforming algorithms are introduced to obtain initial estimates of sound source position and sound field conditions. The results are introduced into the source solving model to improve recognition accuracy and stability. 3) Source waveform recovery and separation methods for multiple sound sources: constructing array weighting function, recovering source signal from correlated sound sources based on delay and accumulation, and employing blind deconvolution algorithm to separates irrelevant source signals. 4) Research on moving sound source identification methods: Estimating motion trajectory and velocity of the sound source, establishing dynamic models for source solving, and performing spectral and amplitude corrections on the reconstructed signals. The project carries out experimental research on articulated industrial robots, and aims to lay the foundation for non-contact acoustic condition monitoring.

噪声是机械设备运行状态的重要表征，但实际环境中多声源声场混合，声源的强度、频谱、波形难以准确识别或分离，阻碍了噪声监测的应用。课题针对此问题开展研究，内容包括：1）基于三维声强探头阵列的声源强度和位置求解模型：设计声强探头阵列，根据声强矢量合成原理构建求解声源参数的非线性方程组，通过数值解法直接识别声源强度和位置；2）基于立体传声器阵列波束形成的声源求解算法优化：声强阵列重组为立体传声器阵列，引入波束形成算法获取声源位置和声场条件的初始估计，提高声源求解模型识别准确性和稳定性；3）混合声场中声源时域波形恢复和分离方法：构造阵列加权函数，基于延迟累加恢复相关声源时域信号，研究通过盲解卷积算法分离不相关声源信号；4）运动声源识别方法研究：估计运动轨迹和速度，建立声源求解的动力学模型，并对识别信号进行频谱和幅度校正。项目以关节型工业机器人为对象开展实验研究，为非接触式声学状态监测奠定基础。

课题结合三维声强测量和波束形成方法，开展远场声源位置和强度识别研究，为非接触式声学状态监测提供新方法。取得的主要研究成果包括：.在基于三维声强探头阵列的声源强度和位置求解模型研究方面：（1）建立了声源信息和空间三维声强的非线性方程组，实现从三维声强探头阵列测量的矢量声强识别多声源位置、强度。（2）推导分析了三维声强矢量测量误差的来源和影响程度，包括相位失配、阵列几何误差等。（3）提出了基于多个已知位置声源的传声器阵列相位失配标定方法，提高三维声强测量精度。.在基于立体传声器阵列波束形成的声源求解算法优化方面：（1）通过压缩感知波束形成正交匹配追踪来估计源的角度，引入到多声源识别模型中，显著降低模型维度。（2）在波束形成算法中引入奇异值分解，提出一种基于声聚焦和奇异值分解波束形成的自适应声源深度识别技术。（3）提出基于波束形成和声源深度信息的声强缩放方法，用于准确估计移动声源声功率。.在混合声场中声源恢复和分离方法研究方面：（1）基于三维声强声源识别模型，结合粒子群优化方法，建立了多声源声功率频谱重建方法。（2）提出基于广义旁瓣抵消器算法的波束形成信号重构方法，提高声源信号重构信噪比。（3）利用声波的传播和叠加特性，提出了适用于复杂环境的声功率级去噪方法，提高频谱重构的准确性。.在工业机器人等低速运动声源识别方法研究方面：（1）引入立体视觉技术获取目标的三维表面，缩小重建声场和声源识别的搜索范围。（2）提出基于平均广义逆波束形成的低速大型轴承故障声学定位方法，抑制环境反射影响，提高空间分辨率。（3）基于立体阵列和波束形成方法，提出了声源三维坐标快速定位方法。.课题组在《Applied Acoustics》等SCI检索国际期刊上发表和录用论文10篇，在《声学学报》、《振动与冲击》等国内EI期刊上发表或录用学术论文5篇，获授权发明专利8项，受理发明专利申请3项。培养博士2人，硕士10人。达到了立项时的成果要求。