Experimental and numerical investigation of a bend-diffusor configuration

Abstract

The current work has dealt with an axial to radial bend-diffuser configuration. The configuration is used behind axial compressors when axial space is limited. To the authors knowledge no work has been reported onthis type of configuration. Radial diffusers are more commen mounted behind centrifugal compressors, where the flow enters the diffuser radially,which result in a completely different flow situation. The current work contains detailed measurements of static pressure, as well as mean and turbulent velocities at several positions within the channel. The main findings are that the bend has an important influence on the flow in the diffuser. The flow in the diffuser following after the convex side if the bend, is considerably more vulnerable to separation than the case is for the flow following after the concave side of the bend. The main reason is found to be the effect that curved surfaces and adverse pressure gradients have on the turbulence. In and after the bend there is a difference in turbulence levels at two sides. This difference gives rise to a wall normal mean velocity,U y, directed from the convex side towards the concave side. This cross stream flow is kept alive throughout most of the diffuser until turbulent diffusion has equalized the cross sectional turbulence profiles. The U y component results in an improvement of the bloundary layer condition on one side, and a worsening at the other. The large difference in boundary layer shape along the sides in the diffuser should be taken into consideration when designing this type of diffusers. Avoiding to do so could very easily lead to poor diffuser performance. It is shown that traditional diffuser performance prediction methods bread down due to the impact if the bend. The Fluent computational software package has been used to simulate the flow employing several different turbulence models. Although pressure recovery is estimated failrly well with all turbulence models, there are large differences in the estimated mean velocity profiles. The realizable k-epsilon by Shi et. Al [82] clearly compared best with the data. Based on the current findings a new design is proposed, improving the boundary layer condition along the surface following the convex side of the bend. The k – epsilon was employed to analyze the new design and the results indicate an marked improvement diffuser performance.