Radial Compressor
The radial compressor is a pressure-producing machine. Like other turbomachines, energy is transferred by dynamic means from the impeller (rotating component) to the working fluid. During the last 50 years, the design of radial compressor has been improved using empirical, theoretical and experimental approaches. Today, CFD plays major role in the design process of radial compressors. The commercial CFD code, ANSYS CFX was used to analyze the rear radial stage portion of a multi-row axial plus Pratt & Whitney centrifugal configuration. This centrifugal stage consisted of a tandem-bladed impeller in combination with a fishtail-style pipe diffuser. The CFD analysis was performed in parallel with experimental testing of centrifugal stage at Pratt & Whitney Canada centrifugal compressor test rig.
Grid generation, for this centrifugal stage, was achieved by a combination of two different methods. While the impeller geometry is best suited to a structured mesh, the complicated diffuser vaneless space is best meshed in an unstructured manner. The mesh of the tandem impeller was generated using an in-house system known as ICAST and ANSYS ICEM CFD Hexa to produce a structured hexahedral H-type mesh having approximately 380k nodes. An impeller grid convergence study was performed previously on isolated impeller analyses to establish the required density of the impeller mesh. The unstructured diffuser mesh was generated using ANSYS ICEM CFD Tetra, and consisted of approximately 250k nodes. A diffuser grid convergence study was performed using the same impeller mesh while increasing the number of diffuser nodes. Figure 1 shows an example of both the impeller and diffuser meshes.
Figure 1: Over view of the impeller and diffuser meshes.
The stage was modeled in steady state using a mixing-plane interface and a complete characteristic was predicted at the design speed using both K-epsilon and SST models. The imposed inlet boundary conditions were the known profiles of total pressure, total temperature, and flow angles from the inlet traversing measurements. Between the trailing edge of the impeller and the diffuser entry, the diffuser backface bleed was extracted along a small strip in the impeller grid. A constant mass flux was imposed along this boundary to achieve a bleed flow of 1.6% for all points modeled. At the exit of the diffuser, a constant static pressure boundary condition was imposed near choke, while a mass flow boundary condition was imposed for points away from choke. In order to avoid an overshoot of the Mach number in the diffuser throat, both the rotational speed and the exit flow/pressure were ramped as a function of iteration count.
This procedure prevented situations in which the flow would accelerate to supersonic levels in the diverging diffuser passage, and enabled stage runs to be completed from no initial solution to full convergence in a single run. All runs were converged to achieve maximum residual levels of less than 10-3. This degree of convergence was typically achieved within 200 iterations. Each point was converged within approximately 9 hours using 4 SGI Origin 3000 CPU’s. Although significantly faster solution times could be achieved on newer machines, the important fact is that stage results could be produced overnight or faster. In addition, due to the use of ramping functions, several runs could be submitted simultaneously without one point depending on the convergence of another for initialization. This enabled several points to be run simultaneously. Providing that adequate CPU’s were available, a speedline having several points could easily be turned around overnight. Figures 2 to 3 show the overall stage pressure ratio and efficiency versus exit corrected flow.
Figure 2: Normalized Total_to_Static Stage Pressure ratio versus Exit corrected mass Flow
Figure 3: Normalized Total_to_Static Stage efficiency versus Exit corrected mass Flow
As shown in figures 2 and 3, the Overall results of the steady state CFD predictions agreed well with measured data. The SST model predictions matched the experimental data considerably better than those using the K-epsilon model.
Reference
Roberts, D. and Steed, R., “ A Comparison of Steady state Centrifugal stage CFD Analysis to Experimental Rig Data”, ANSYS/CFX USER CONFEREENCE, Canonsburg, June 2004.