Large-scale des analysis of unsteady flow field in a multi-stage axial flow compreßor at off-design condition using k computer

The paper presents the results of large-scale numerical simulations which were conducted for better understanding of unsteady flow phenomena in a multi-stage axial flow compreßor at off-design condition. The compreßor is a test rig compreßor which was used for development of the industrial gas turbine, Kawasaki L30A. The compreßor consists of 14 stages, the front two stages and the front half stages of which were investigated in the present study. The final goal of this study is to elucidate the flow mechanism of the rotating stall inception in the multi-stage axial compreßor for actual gas turbines, and according to the test data it is considered that the 2nd stage and the 5th or 6th stage are suspected of leading to the stall. In order to capture precise flow physics in the compreßor, a computational mesh for the simulation was generated to have at least several million cells per paßage, which amounted to 650 million cells for the front 2-stage simulation and two billion cells for the front 7-stage simulation (about three hundred million cells for each stage). Since these were still not enough for the largeeddy simulation (LES), the detached-eddy simulation (DES) was employed, which can calculate flow fields except near-wall region by LES. The required computational resources were quite large for such simulations, so the computations were conducted on the K computer (RIKEN AICS in Japan). The simulations were well validated, showing good agreement with the measurement results obtained in the test. In the validation, the effect of the boundary condition for the casing wall was also investigated by comparing the results between the adiabatic boundary condition and the isothermal boundary condition. As for the unsteady effect, the wake/blade interaction was investigated in detail. In addition, unsteady flow phenomena in the present compreßor at off-design condition were analyzed by using data mining techniques such as vortex identification and limiting streamline drawing with the LIC (line integral convolution) method. The simulation showed that they could be caused by the corner separation on the hub side.