Computationally efficient domain decomposition and near-bell coupling of rotary spray paint bells

Abstract

An industrial spray paint application is a multiscale, multiphysics and multiphase problem requiring large efforts for efficient and accurate modelling. Fraunhofer Chalmers Centre (FCC) has developed the computational tool “IBOFlow” for these purposes, the efficiency of which was the subject of this thesis. The multiphase aspects combine a strong shaping airflow with the injection of liquid paint or powder which is modelled using an Eulerian-Lagrangian framework with a two-way coupling between the phases. The multiphysics modelled are fluid dynamics, particle dynamics and electrostatic forces which are also coupled with each other. Lastly, and the focus of this thesis, is the modelling of the multiscale issue. The smallest spatial and temporal scales near the paint injection are currently modelled as compressible in a resource demanding process which is separate from the larger scales. The small scales are subsequently imported as a boundary condition on a coarser simulation where both the spatial and temporal scales extend to cover an entire paint job. Different desired finishing results require certain parameter optimizations where a multitude of compressible simulations are needed to model the different settings, referred to as brushes. This process quickly becomes both time and resource intensive mainly due to the large optimization spaces. This thesis has investigated the possibility of an incompressible simplification for the smaller temporal and spatial scales. The incompressible simplification was incrementally tested and validated for larger and larger fractions of the compressible boundary condition with the goal to reduce the number of required compressible simulations to one while also keeping sufficiently good accuracy. It was first tested on a pure airflow and subsequently on a fluid particle coupled flow. Lastly a cross validation was tested where the imported boundary condition was kept constant as different particle injection parameters varied. The final result showed that a pure airflow could be simulated incompressibly from 0.5 cm below the injection point and downwards with an error of 2.6% in the high velocity regions. The addition of particles and its two-way coupling on the fluid significantly reduced the errors in the re-circulation region and also resulted in a small improvement in the higher velocity regions. The cross validation showed that a single compressible boundary condition could be used for all injection cases with a maximum error of 2.8% in the high velocity regions and an average error of 2.15%. Larger errors, mostly between 3 and 4%, were obtained in a re-circulation region in the interior flow. Comparing the errors obtained from the airflow fields to corresponding errors in the paint thickness distributions showed that the magnitude of the errors from the airflow field could in several cases approximate how well the thickness distribution turned out. In other cases the results between the different validation methods diverged from each other.