基于势函数与压缩感知的欠定盲源分离及应用
发布时间：2018-12-11 13:25
【摘要】：盲源信号分离因其应用广泛，自提出就备受关注。正定及超定盲源分离研究已较为成熟，而测量信号少于源信号的欠定盲源分离问题在理论和算法方面却存在着一定的技术瓶颈有待进一步探索。 信号的稀疏性是欠定盲源分离算法的前提，但是在实际应用中，只有极少数的信号在时域具备稀疏性，所以对于多数信号，在进行欠定盲源分离之前必须进行稀疏分解，使之具备稀疏性。传统基于过完备原子库的匹配追踪稀疏分解算法存在运算速度较慢、运算结果精度不高等不足，鉴于此，本文将基于梯度信息改进的粒子群优化算法用于稀疏分解过程中寻找最佳原子，大幅增加了算法的收敛速度与分解速度，并在一定程度上也提高了算法精度。在同等精度要求的前提下，本文算法所用时间仅为经典稀疏分解算法的五分之一。将此改进算法用于信号去噪，对信噪比为5dB的信号进行去噪，本文算法的去噪效果比小波去噪高7.5519dB，效果优于小波去噪，充分体现了算法优势。 欠定盲源分离算法分为估计混叠矩阵与重构源信号两个步骤。传统的基于K均值聚类算法及最小路径法的欠定盲源分离两步法存在K值难以确定，对初始值敏感，噪声和奇异点难以排除以及相对缺乏理论依据等诸多不足，针对以上问题，本文提出了基于势函数及压缩感知理论的新型两步算法。该算法首先利用多峰值粒子群寻优算法改进的势函数法来估计混合矩阵，然后利用估计矩阵来构建传感矩阵，并将基于正交匹配追踪的压缩感知算法引入欠定盲源分离过程中，最终实现源信号的重构。本文分别对混合正弦信号与混合声音信号进行欠定盲源分离实验，仿真实验结果表明，混合矩阵最高估计精度达到99.13%，重构信号干扰比均高于10dB，很好的满足了重构精度的要求，验证了本文算法的有效性。所提算法对一维混合信号的欠定盲源分离具有良好的普适性和较高的准确率。 与实际工程应用相结合，将本文算法应用于风机齿轮箱振动信号的分析处理与故障诊断。首先，在保留故障信息的前提下，将采集到的风机齿轮箱振动信号利用本文改进匹配追踪算法进行稀疏分解及去噪处理。其次，利用本文算法对去噪后的风机齿轮箱振动信号进行欠定盲源分离，对比正常运行状态下的信号分离结果，可以初步断定故障点位置。最后，对分离后风机齿轮箱振动信号进行频谱分析，并且结合故障诊断相关知识，可以最终断定故障原因。通过风机拆机检修，发现风机故障部位与故障原因均与推断结果相符，在实践中验证了算法的可行性。 全文最后，，总结了算法的优点与不足，同时结合无线传感网络，勾勒出风机状态无线实时监测的美好蓝图，展望了本文算法的应用前景。
[Abstract]:Blind source signal separation has attracted much attention because of its wide application. The study of positive definite and overdetermined blind source separation is mature, but the problem of under-determined blind source separation with less measured signal than the source signal has some technical bottlenecks in theory and algorithm to be further explored. The sparsity of the signal is the premise of the underdetermined blind source separation algorithm, but in practical applications, only a few signals have sparsity in the time domain, so for most signals, the sparse decomposition must be carried out before the underdetermined blind source separation. Make it sparse. The traditional matching tracing sparse decomposition algorithm based on over-complete atomic library has some shortcomings, such as slow operation speed and low precision. In this paper, an improved particle swarm optimization algorithm based on gradient information is used to find the best atoms in the sparse decomposition process, which greatly increases the convergence speed and decomposition speed of the algorithm, and improves the accuracy of the algorithm to a certain extent. On the premise of the same precision requirement, the time of this algorithm is only 1/5 of that of the classical sparse decomposition algorithm. The improved algorithm is applied to signal denoising and the signal to noise ratio (5dB) is de-noised. The denoising effect of this algorithm is 7.5519dBhigher than that of wavelet denoising, which fully reflects the superiority of the algorithm. The underdetermined blind source separation algorithm is divided into two steps: estimating the aliasing matrix and reconstructing the source signal. The traditional two-step undetermined blind source separation method based on K-means clustering algorithm and minimum path method is difficult to determine K value, sensitive to initial value, difficult to eliminate noise and singularity, and relatively lack of theoretical basis. In this paper, a new two-step algorithm based on potential function and compression sensing theory is proposed. Firstly, the hybrid matrix is estimated by the improved potential function method of multi-peak particle swarm optimization algorithm, and then the sensing matrix is constructed by using the estimation matrix, and the compression sensing algorithm based on orthogonal matching tracking is introduced into the process of under-determined blind source separation. Finally, the source signal is reconstructed. In this paper, the underdetermined blind source separation experiments of mixed sinusoidal signal and mixed sound signal are carried out respectively. The simulation results show that the maximum estimation accuracy of mixed matrix is 99.13 and the interference ratio of reconstructed signal is higher than 10 dB. It meets the requirements of reconstruction accuracy and verifies the effectiveness of this algorithm. The proposed algorithm has good universality and high accuracy for undetermined blind source separation of one-dimensional mixed signals. Combined with practical engineering application, this paper applies the algorithm to the vibration signal analysis and fault diagnosis of fan gearbox. Firstly, on the premise of retaining fault information, the vibration signals of fan gearbox are processed by sparse decomposition and denoising using the improved matching tracking algorithm in this paper. Secondly, by using the algorithm in this paper, the vibration signals of the de-noised fan gearbox are separated by under-determined blind source, and the location of the fault point can be preliminarily determined by comparing the results of the signal separation under normal operation. Finally, the vibration signal of the separated fan gearbox is analyzed by frequency spectrum analysis and combined with the knowledge of fault diagnosis, the cause of the fault can be determined finally. By examining and repairing the fan disassembly machine, it is found that the fault location and cause of the fan are consistent with the inferred results, and the feasibility of the algorithm is verified in practice. At the end of the paper, the advantages and disadvantages of the algorithm are summarized. At the same time, combining with the wireless sensor network, the good blueprint of wireless real-time monitoring of fan status is drawn, and the application prospect of this algorithm is prospected.
【学位授予单位】：辽宁大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN911.7
本文编号：2372629
[Abstract]:Blind source signal separation has attracted much attention because of its wide application. The study of positive definite and overdetermined blind source separation is mature, but the problem of under-determined blind source separation with less measured signal than the source signal has some technical bottlenecks in theory and algorithm to be further explored. The sparsity of the signal is the premise of the underdetermined blind source separation algorithm, but in practical applications, only a few signals have sparsity in the time domain, so for most signals, the sparse decomposition must be carried out before the underdetermined blind source separation. Make it sparse. The traditional matching tracing sparse decomposition algorithm based on over-complete atomic library has some shortcomings, such as slow operation speed and low precision. In this paper, an improved particle swarm optimization algorithm based on gradient information is used to find the best atoms in the sparse decomposition process, which greatly increases the convergence speed and decomposition speed of the algorithm, and improves the accuracy of the algorithm to a certain extent. On the premise of the same precision requirement, the time of this algorithm is only 1/5 of that of the classical sparse decomposition algorithm. The improved algorithm is applied to signal denoising and the signal to noise ratio (5dB) is de-noised. The denoising effect of this algorithm is 7.5519dBhigher than that of wavelet denoising, which fully reflects the superiority of the algorithm. The underdetermined blind source separation algorithm is divided into two steps: estimating the aliasing matrix and reconstructing the source signal. The traditional two-step undetermined blind source separation method based on K-means clustering algorithm and minimum path method is difficult to determine K value, sensitive to initial value, difficult to eliminate noise and singularity, and relatively lack of theoretical basis. In this paper, a new two-step algorithm based on potential function and compression sensing theory is proposed. Firstly, the hybrid matrix is estimated by the improved potential function method of multi-peak particle swarm optimization algorithm, and then the sensing matrix is constructed by using the estimation matrix, and the compression sensing algorithm based on orthogonal matching tracking is introduced into the process of under-determined blind source separation. Finally, the source signal is reconstructed. In this paper, the underdetermined blind source separation experiments of mixed sinusoidal signal and mixed sound signal are carried out respectively. The simulation results show that the maximum estimation accuracy of mixed matrix is 99.13 and the interference ratio of reconstructed signal is higher than 10 dB. It meets the requirements of reconstruction accuracy and verifies the effectiveness of this algorithm. The proposed algorithm has good universality and high accuracy for undetermined blind source separation of one-dimensional mixed signals. Combined with practical engineering application, this paper applies the algorithm to the vibration signal analysis and fault diagnosis of fan gearbox. Firstly, on the premise of retaining fault information, the vibration signals of fan gearbox are processed by sparse decomposition and denoising using the improved matching tracking algorithm in this paper. Secondly, by using the algorithm in this paper, the vibration signals of the de-noised fan gearbox are separated by under-determined blind source, and the location of the fault point can be preliminarily determined by comparing the results of the signal separation under normal operation. Finally, the vibration signal of the separated fan gearbox is analyzed by frequency spectrum analysis and combined with the knowledge of fault diagnosis, the cause of the fault can be determined finally. By examining and repairing the fan disassembly machine, it is found that the fault location and cause of the fan are consistent with the inferred results, and the feasibility of the algorithm is verified in practice. At the end of the paper, the advantages and disadvantages of the algorithm are summarized. At the same time, combining with the wireless sensor network, the good blueprint of wireless real-time monitoring of fan status is drawn, and the application prospect of this algorithm is prospected.
【学位授予单位】：辽宁大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN911.7
【参考文献】
相关期刊论文 前3条
1 尹忠科;邵君;Pierre Vandergheynst;;利用FFT实现基于MP的信号稀疏分解[J];电子与信息学报;2006年04期
2 刘亚新;赵瑞珍;胡绍海;姜春晖;;用于压缩感知信号重建的正则化自适应匹配追踪算法[J];电子与信息学报;2010年11期
3 ;A Method for Gear Fault Diagnosis Based on the Empirical Mode Decomposition[J];International Journal of Plant Engineering and Management;2004年04期
本文编号：2372629
本文链接：https://www.wllwen.com/kejilunwen/wltx/2372629.html
