Simulation of Tube Bundle Vibrations: A Numerical 3D-Method

The shell-side cross-flow in tubular heat exchangers may cause vibrations leading to failure within hours or in long term. Design is still based on half-empirical correlation, based on the equation of Connors (1978). Overdesign (Kassera 1996) and singular cases of damage (Fischer and Strohmeier, 2002) are the result. Therefore a structural model for the tube motions has been developed further and coupled to the commercial flow simulation code ANSYS CFX. The predictive capability of such coupled methods is limited by the flow simulation. Still simplifications or modeling are needed, especially for turbulence. The paper starts with an overview of modeling assumptions used so far. In addition to simulations of flow around rigid circular cylinders (Reichel and Strohmeier, 2008) LDA-measurements of flow through rigid glass-bundles have been compared to flow simulations. The sample of results presented below demonstrates that, besides level of fluctuations is predicted far too low, the overall velocity distribution on the shell side is predicted well by the SST turbulence model (Menter 1994), making URANS models like SST worth a try, if mainly flow forces are needed. To capture the tube dynamics an Euler-Bernoulli beam model of Fischer (2001), discretized by central differences in space and Newmark’s method (1959) in time, has been extended and implemented into ANSYS CFX. Calculations will be presented, showing that simulations of initially deflected tubes almost perfectly match analytic predictions. To adapt the numerical grid for the flow calculations to the tube displacements, the code inherent standard methods at large displacements resulted in negative volumes and solver failure. Therefore the standard methods have been replaced by own routines for grid deformation. Even for grids with fine near wall resolution, this method is able to cope with large displacements. Finally, coupled simulations are conducted of a single cylinder and of a cantilevered tube bundle in cross flow. For the single cylinder amplitudes are extremely overpredicted as long as 2D-modeling is used. 3D-modeling shows a phase shift of vortex shedding along the cylinder, which results in noticeably lower tube deflections. But, using the SST-model, amplitudes are still higher than measured. A model extension for laminar to turbulent transition leads to further improvement. The same holds for the tube bundle. Onset velocity of instability is predicted too low, amplitudes are too high. Modeling transition and large scale 3D-effects moves results closer to experimental observations. Further improvements are expected taking small scale 3D-effects into account by introducing more grid layers along the tubes.