Eventos Anais de eventos
COBEM 2019
25th International Congress of Mechanical Engineering
APPLICATION OF THE ADAPTATIVE HAMILTONIAN MONTE CARLO (AHMC) FOR FASTER CONVERGENCE IN THE ESTIMATION OF THERMAL PROPERTIES IN HOMOGENEOUS THIN PLATES
Submission Author:
Maicon de Paiva Torres , RJ
Co-Authors:
Maicon de Paiva Torres, Géssica Ramos da Silva, Diego Knupp, Leonardo Stutz, Antônio Silva Neto
Presenter: Maicon de Paiva Torres
doi://10.26678/ABCM.COBEM2019.COB2019-1918
Abstract
In this work, an adaptative version of the Markov Chain Monte Carlo method (MCMC) known as Hamiltonian Monte Carlo (HMC) is proposed to estimate the thermophysical properties of a thin plate made of a homogeneous material. The direct problem is based on the experimental setup for temperature nonintrusive measurements in which the plate stands vertically whilst one surface is partially heated by an electric resistance and the other surface loses energy to the vicinity by radiation and natural convection. With regard to parameter space exploration, HMC stands in front of conventional random walk MCMC, such as Metropolis-Hastings algorithm, due to its deterministic explorative nature based on the gradient of the posterior distribution probability density, being more consistent and efficient. In order to enhance the Markov chains convergency for the inverse problem solution, an Adaptative Hamiltonian Monte Carlo (AHMC) version was implemented. On account of this adaptation, it is possible to control the acceptance rate to a desirable level and achieve more stable solutions.
Keywords
heat conduction, Lumped formulation, Thermophysical properties estimation, Adaptative Hamiltonian Monte Carlo (AHMC)
