Bayesian normal modes identification and estimation of elastic coefficients in resonant ultrasound spectroscopy. (14th May 2015)

• Record Type: Journal Article
• Authors: Bernard, Simon; Marrelec, Guillaume; Laugier, Pascal; Grimal, Quentin
• Is Part Of: Inverse problems. Volume 31:Number 6(2015:Jun.)
• Publication Date: 2015-05-14
• Subjects: RUS -- anisotropy -- stiffness tensor -- vibration modes -- damping -- frequency pairing -- Markov chain Monte Carlo; Inverse problems (Differential equations) -- Periodicals; 515.357
• DOI: 10.1088/0266-5611/31/6/065010
• Languages: English
• ISSNs: 0266-5611

Abstract

Abstract: Resonant ultrasound spectroscopy is an experimental technique for measuring the stiffness of anisotropic solid materials. The free vibration resonant frequencies of a specimen are measured and the stiffness coefficients of the material adjusted to minimize the difference between experimental and predicted frequencies. An issue of this inverse approach is that the measured frequencies are not easily paired with their predicted counterpart, leading to ambiguities in the definition of the objective function. In the past, this issue has been overcome through trial-and-error methods requiring the experimentalist to find the correct pairing, or through involved experimental methods measuring the shapes of the normal vibration modes in addition to their frequencies. The purpose of this work is to show, through a Bayesian formulation, that the inverse problem can be solved automatically and without requiring additions to the usual experimental setup. The pairing of measured and predicted frequencies is considered unknown, and the joint posterior probability distribution of pairing and stiffness is sampled using Markov chain Monte Carlo. The method is illustrated on two published data sets. The first set includes the exact pairing, allowing validation of the method. The second application deals with attenuative materials, for which many predicted modes cannot be observed, further complicating the inverse problem. In that case, introduction of prior information through Bayesian formulation reduces ambiguities.