Prediction of Aerodynamic Noise for Centrifugal Fan of Air-Conditioner

C1 - Noise of Centrifugal Fans

Development of silent air-conditioners is one of the most important problems in recent changes to life-styles, such as demands of silent office and home. Aerodynamic noise from fans contributes to a large percentage of the overall noise from air-conditioners. Therefore, the development of silent fans would contribute to reducing the noise levels of air-conditioners. In this study, flow fields in the centrifugal fans for the indoor units of air-conditioners were calculated with finite element method-based large eddy simulation (LES) with the aim of predicting aerodynamic noise. The intended centrifugal fans were two types, old and new, and they had different blade shapes and fan performances. Calculations were implemented by using K computer, which is Japanese national super computer developed in order to high performance computing. The numerical simulation code employed throughout the LES was called FrontFlow/blue (FFB). The code was based on the finite element discretization of filtered incompressible continuity and Navier Stokes equations. The Dynamic Smagorinsky model was used as a sub-grid scale model. The computational model consists of three parts, i.e., the inlet, impeller, and outlet parts. The impeller part is in the rotating frame of reference. The inlet and the outlet parts are in the stationary frames. The grid is composed of hexahedral elements. Aerodynamic noise was calculated with Curle’s equation. The sound source was assumed to be acoustically compact. We compared 60 million grids (60M-grid) and 500 million grids (500M-grid) calculation results for investigation of influence on grid resolution. The 500-Mgrid was implemented by parallel computation with 1,024 nodes (8,192 cores) of K computer. Calculated distributions of velocities at blade exit agreed with experimental results. The accuracy of 500M-grid was improved compared to 60M-grid. Calculated sound pressure level by 500M-grid was also improved compared to 60M-grid calculation results. Number of captured streaks on the blade by 500M-grid increased as compared to those by 60M-grid. As a result, size of streak by 500M-grid became smaller than that by 60M-grid. The proper capturing of the streaks was contributed to improvement of calculation results. We confirmed that high performance computing by K computer was effective for improvement of calculation accuracy.