基于奇异值分解的磁记忆信号特征提取方法

发布时间：2018-09-11 13:27

【摘要】：针对金属磁记忆信号容易受到环境噪声影响,使得缺陷信号可检测性降低的情况,首先,利用传统的奇异值分解方法对场桥主梁磁记忆信号进行分解和重构,发现尽管可以取得较为理想的降噪效果,但如何自适应确定重构时的奇异值个数仍存在困难;然后,将磁记忆信号按照二进递推方法构造矩阵,重复进行奇异值分解可以获得具有不同分辨率的近似信号和细节信号,从而形成多分辨奇异值分解,其中细节信号对应磁记忆中的噪声成分,近似信号为去除噪声之后的有效磁记忆信号,从而实现了磁记忆信号的降噪。将该方法用于某场桥主梁磁记忆信号的处理,有效地提高了重构信号的信噪比,准确地判断出了该主梁的应力集中区域,为评估其应力状态和早期故障诊断奠定了基础。
[Abstract]:In view of the fact that metal magnetic memory signal is susceptible to ambient noise and the detectability of defect signal is reduced, firstly, the traditional singular value decomposition method is used to decompose and reconstruct the magnetic memory signal of the main girder of the field bridge. It is found that although the ideal noise reduction effect can be achieved, how to adaptively determine the singular value of the reconstructed signal is obtained. The number is still difficult; then, the magnetic memory signal matrix is constructed by binary recursive method, and the approximate signal and detail signal with different resolution can be obtained by repeated singular value decomposition, thus forming a multi-resolution singular value decomposition, in which the detail signal corresponds to the noise component in the magnetic memory, and the approximate signal is after the noise is removed. The method is applied to the magnetic memory signal processing of the main girder of a field bridge. The signal-to-noise ratio of the reconstructed signal is effectively improved, and the stress concentration area of the main girder is accurately determined, which lays a foundation for evaluating its stress state and early fault diagnosis.
【作者单位】：北京工业大学机电学院;
【基金】：国家自然科学基金资助项目(51075009) 北京市优秀人才培养计划资助项目(2011D005015000006)
【分类号】：TN911.7

【参考文献】