Experimental Methods for Quantitative Analysis of Thermally
Driven Flows
Tomasz A. Kowalewski
Department of Mechanics and Physics of Fluids, IPPT PAN, Polish Academy of Sciences
Warsaw, Poland
Abstract. Properly designed validation experiments are necessary to establish a satisfac-
tory level of confidence in simulation algorithms. In this review recent achievements in
the measurement techniques used for monitoring macroscopic flow field features are pre-
sented. In particular, optical and electro-optical methods, for example thermography,
tomography or particle image velocimetry, are reviewed and their application to simple
solidification experiments demonstrated. Computer supported experimentation combined
with digital data recording and processing allows for the acquisition of a considerable
amount of information on flow structures. This data can be used to establish experimental
benchmarks for the validation of numerical models employed in solidification problems.
Three experimental benchmarks based on water freezing in small containers are proposed
to model flow configurations typically associated with crystal growth and mould-filling
processes.
1 Introduction
1.1 Why do we need to measure?
Modern computational fluid dynamics (CFD) began with the arrival of computers in the early
1950s. The field of computational modelling of flow with heat and mass transfer has subsequently
matured to the level it has. After half a century this is evidenced of a multitude of commercial
codes purporting to solve almost every problem imaginable and suggesting the époque of expen-
sive and complicated laboratory experimentation to have passed. Some foresights even profess
construction of universal Navier-Stokes solvers, which implemented in a black box will be used in
the predictable future almost the same way as pocket calculators now. Although all would wel-
come such a development, the validation of numerical results remains a concern tempering some
of this optimism (Roache 1997). Typical difficulties in obtaining credible predictions for industrial
problems lead to the often-encountered dilemma: Do we trust numerical simulations? This ques-
tion seems especially pertinent when modelling solid-liquid phase change problems (see Gobin &
Le Quere 2000), due to the complexity of the physical phenomena and the difficulties implied by
the multi-scale nature.
One of the basic problems encountered by any model attempting to simulate physics involved in
solidification is the broad diversity of length scales. The basic length scales fundamental to solidi-
fication processes arise from capillary forces, heat conduction, solutal diffusion, and convection.
Different mechanisms of convection, including forced, natural, and Marangoni convection can all