Spectral Analysis Using a Method of Sequential Transforms.

摘要

We are exploring the use of acoustic resonant techniques to identify defective components by monitoring the fundamental modes of vibration that arise following an impulse load. We employ the spectrum of the filtered data as a feature for use in automated statistical pattern recognition. The feature provides a measure of the severity of the defect using the acoustic signature of a defect free component as a baseline. Defects change the signature by shifting or splitting the frequencies and introducing beats. Identifying these changes in the spectrum relative to baseline data is difficult because of the large number of natural frequencies associated with complex geometries and the presence of noise in the data. In the case of real-time health monitoring, obtaining a representative spectrum is challenging since averaging cannot be employed without the use of multiple transducers. We have developed a new approach to obtain a smoother spectrum by analyzing sequential data using an approach similar to Rayleigh's quotient method for eigenvalue problems and super-resolution techniques. The resulting spectrum enhances the natural modes of vibration and provides a better feature for automated classification techniques. The numerical implementation of the algorithm requires modest computational resources. The algorithm for generating the Rayleigh quotient has been successfully tested on numerically generated waveforms and experimentally acquired data. In all cases, there was a significant reduction in the inherent noise and the numerically generated noise from digital filtering.