Transient heat loss of hot water tanks using CFD

Abstract

The product development process can be a time consuming process. A computational fluid dynamics (CFD) software can be used to shorten the design cycle, as well as lower the material waste used for experimental tests. CFD is the numerical method of predicting fluid behavior. It can solve a large variety of different complex problems, though, one of its drawbacks is the more numerical accurate solution the higher computational cost. This thesis focuses on developing a methodology that keeps the computational cost to a minimum while simultaneously provide results that correlate to experimental data of unsteady state simulations, also called transient CFD simulations.

The methodology was conducted through a literature analysis, analytical calculations and evaluated by being tried out on two simulations using the CFD software STAR-CCM+. The numerical results were compared to experimental data gathered from an experimental test called standing heat-loss test, performed on a thermal energy storage (TES) tank with and without insulated surfaces. The literature analysis suggested a numerical setup using the Boussinesq approximation with fully coupled modeling and an implicit time-stepping approach. The analytical calculations gave an initial value on the convection boundary condition (BC), they also indicated to use different heat transfer coefficients on different heights of the uninsulated tank, while a constant heat transfer coefficient can be used for the insulated tank.

A common approach to reduce the computation time is to decrease the domain size by using symmetry planes. Symmetry planes can however disrupt fluid flow and alter the numerical results. To remove any uncertainties of a reduced domain size it is advised to do a comparison between the results from the whole domain and the reduced size. In these simulations it could be observed that the symmetry planes caused instability of the solver in more turbulent flow regions but no other differences could be seen. A multi-time-stepping approach combined with a minimum convergence criterion of 10−6 gave the best reduction of computation time while providing results that correlated to experimental data. This approach however varies greatly from case to case and as larger time-steps approximates solutions this approach is not suggested if the goal is to have simulations with numerical independence.

With all time reducing strategies implemented, the simulations could be reduced from 700 cores, 48 hours to 4 cores and 1 hour.