Inverse Thermal Analysis of a Coke Drum Using a Bayesian Probabilistic Approach

Abstract

There is much interest in predicting the optimal operating conditions of a coke drum in order to extend its life and optimize both maintenance and repair. Typically, only temperature measurements on the outer surface of the wall are available from monitoring. In order to predict damage due to thermal stresses and other mechanisms, the temperature distribution through the wall is required. This could be determined if the heat flux on the inner surface of the wall were known, but this is difficult to obtain directly. In this paper, the heat flux distribution on the inner wall is determined solely from thermocouple measurements taken on the outside of the wall by solving a stochastic inverse heat conduction problem (IHCP). A finite element analysis is used to solve the forward thermal problem, and a Bayesian inference approach is used to model the posterior probability distribution of the heat flux. A newly developed probabilistic sampling technique known as the Particle Raking Algorithm (PRA) is found to be quite effective at solving this inverse problem. Once determined, the heat flux distribution is then applied as a boundary condition for the finite element model to determine the through-wall temperature distribution.