Compressed sensing based on dictionary learning for reconstructing blade tip timing signals
Compressed sensing based on dictionary learning for reconstructing blade tip timing signals
复制标题
DOI：
10.1109/phm.2017.8079253
复制
发表时间：
2017-07
期刊：
影响因子：
--
通讯作者：
Minghao Pan;F. Guan;Haifeng Hu;Yongmin Yang;Hailong Xu
中科院分区：
文献类型：
--
作者：
Minghao Pan;F. Guan;Haifeng Hu;Yongmin Yang;Hailong Xu
叶尖定时（BTT）是一种很有前途的叶片振动测量技术。由于欠采样是BTT测量的固有缺陷，压缩感知理论能够恢复稀疏信号，为BTT信号的重构提供了一种新的方法。考虑到实际工况下激振方法和叶片材料的可变性，不能将叶片的瞬时位移仅仅建模为多分量正弦波。因此，需要合适的字典来表示稀疏的叶尖振动信号。本文提出了一种非规则叶尖挠度信号的重构方法，该方法包括四个关键步骤。首先，基于光学探头的位置，确定压缩感知的测量矩阵。其次，提出了一种基于稀疏表示的BTT信号数学模型。第三，结合K-SVD字典学习方法，利用非线性情况下实测叶片位移的实验结果实现了字典的训练。最后利用基追踪算法恢复叶尖定时信号。通过振动实验中应变片信号的数值仿真实例验证了上述方法的有效性。该方法的主要优点是能很好地适应强非线性情况下叶片振动的特征识别。
Blade tip timing (BTT) is a promising technique for the blade vibration measurement. Since under-sampling is an inherent drawback of BTT measurement, compressed sensing theory which enables the recovery of sparse signals provides a novel method to the reconstruction of BTT signals. Considering the variability of the excitation method and the material of blades under actual working condition, the instantaneous displacement of each blade could not be merely modeled as multiple component sine waves. Therefore, proper dictionary is needed for the representation of sparse blade tip vibration signals. We proposed a method for reconstructing irregular blade tip deflection signals in this paper which consists of four key steps. First, based on the optical probe placements, measurement matrix for compressed sensing is determined. Secondly, a mathematical model of BTT signals is proposed based on sparse representation. Third, combining the K-SVD dictionary learning method, training dictionary is realized through experimental results of measured blade displacements under nonlinear case. Finally, blade tip-timing signals are recovered by Basis Pursuit Algorithm. Numerical simulation examples of strain gage signals from vibration experiments are performed to prove the validation of mentioned methods. The main advantage of this method is the well adapting to feature identification of blades vibrations in strong nonlinear case.