Bayesian-based Resonance Ultrasound Spectroscopy with Particle Swarm Optimization

Abstract

Resonance ultrasound spectroscopy (RUS), an experimental method to identify the elasticity of solids from its mechanical resonant frequencies, remains the challenge of sensitive dependence on the initial elasticity guess. Some researchers have shown that Bayesian inference with prior knowledge could mitigate sensitivity to the initial values. A popular gradient-based optimizer with local rapid convergence like the quasi-Newton method is no longer appropriate due to the nonlinearity and non-convexity of the Bayesian objective function. A Markov Chain Monte Carlo (MCMC) sampling solver can be used, but at the expense of massive computation time. To accelerate the calculation, the global optimizer particle swarm optimization (PSO) was introduced here. The detailed performance comparison between MCMC and PSO was studied in RUS for a series of measured cortical bone specimens.