Sparse Adaptive Method Of Order Determination And Feature Extraction Of Vibration Signals

The feature extraction of vibration signals is an important foundation for vibration monitoring. Faster, more accurate and intelligent methods are intensely needed for enormous amount of building project on line monitoring. So putting forward new modal parameter extraction methods and refining the technology are essential.Compressive sensing technology is a hot area of research. This paper improves the parameter identification method by combining the time domain method and compressive sensing technology. Firstly, a sparse basic matrix of free response signal is constructed adaptively to its character. Which is done based on AR model derived from rational fraction of free response. After that, many different equations with same solution are deduced through many Gaussian random measurement matrixes to measure the signal. Solving these equations with sparse optimization method for sparse solutions and correcting them is based on the statistics of solutions. The final sparse solution is the basis to determine whether the modal is true or false. This idea can avoid the order estimation and ensure a strong performance of anti-noise. The experiment of bridge model demonstrates the efficiency of improved method.Stabilization diagram method shows the relation of modal parameters and order. The false modal may affect the estimation of order when the order is much larger than it is which may yield many false modal. When sparse coefficients of modal shape are regarded as constraints in the calculation of modal frequency, the number of false modal may reduce substantially. It can reduce the interruption from false modal to improve the accuracy. The stabilization diagram method must calculate the modal parameters to estimate the order. While, it is wasteful if the order just is the purpose. This paper proposes a method to estimate order depending on the sparse coefficients of modal shape. Some measures are used to improve the accuracy of sparse solution. Then the order is found from final sparse solution. The experiments show that estimation on the order basis on the coefficients of modal shape is an effective way. It avoids the calculation of mode parameters and saves much time to identify the order.The improved method in this paper is based on the theory of compressive sensing. This theory ensures that many unstable sparse solutions can be optimized as a final accurate sparse solution. So it can work well in strong noise environment. On the other hand, the sparse basic matrixes in the paper are structured adaptively to make the most of signal information. It is different with fixed basic matrixes such as DCT matrix, wavelet matrix. So the identification method improved is more dependable and intelligent.