Design Optimization of a Low Pressure Steam Turbine Radial Diffuser Using an Evolutionary Algorithm and 3D CFD

Abstract

The design of the radial exhaust hood of a low pressure (LP) steam turbine has a strong impact on the overall performance of the LP turbine. A higher pressure recovery of the diffuser will lead to a substantial higher power output of the turbine. One of the most critical aspects in the diffuser design is the steam guide, which guides the flow near the shroud from axial to radial direction and has a high impact on the pressure recovery. This paper presents a method for the design optimization of the steam guide of a steam turbine for industrial power generation and mechanical drive of centrifugal compressors. This development is in the frame of a continuous effort in GE Oil and Gas to develop more efficient steam turbines. An existing baseline exhaust and steam guide design is first analyzed together with the last LP turbine stage with a frozen rotor full 3D Computational Fluid Dynamics (CFD) calculation. The numerical prediction is compared to available steam test turbine data. The new exhaust box and a first attempt new steam guide design are then first analyzed by a CFD computation. The diffuser inlet boundary conditions are extracted from this simulation and used for improving the design of the steam guide. The maximization of the pressure recovery is achieved by means of a numerical optimization method that uses a metamodel assisted differential evolution algorithm in combination with a 3D CFD solver. The profile of the steam guide is parameterized by a Bezier curve. This allows for a wide variety of shapes, respecting the manufacturability constraints of the design. In the design phase it is mandatory to achieve accurate results in terms of performance differences in a reasonable time. The pressure recovery coefficient is therefore computed through the 3D CFD solver excluding the last stage, to reduce the computational burden. Steam tables are used for the accurate prediction of the steam properties. Finally, the optimized design is analyzed by a frozen rotor computation to validate the approach. Also off-design characteristics of the optimized diffuser are shown.