Peak Finding within Spectral Data from an Aircraft T-Tail using a Black-Box Variational Bayesian Approach

摘要

The use of fully automated systems for identifying the modal characteristics of a structure is a key step in achieving real-time vibration-based structural health monitoring. Such analyses usually involve identifying the modal characteristics of a particular structure, which are in turn characterized via the peaks of the resulting spectral curves. It is the aim of this paper to present a novel approach to the classic, peak finding problem for spectral curves. Our proposed solution to this problem is based upon a Variational Bayesian Gaussian Mixture Model (VB-GMM), and it performs peak finding in an autonomous manner. This is achieved by observing that the peak finding problem can be approached from a probabilistic perspective, which therefore opens up access to new innovations in the machine learning field. This idea will be demonstrated on experimental spectral data which was obtained from an in-house aircraft T-Tail test bed.