In this article, a reduced-order model (ROM) is established by the Galerkin projection method based on a body-fitted coordinate (BFC) system. Differing from other proper orthogonal decomposition (POD) based ROMs in relevant literatures, the established model could accurately calculate the physical problems with geometric shapes different from the sample cases. Two typical numerical examples are given to validate this advantage. The results show that the average deviation between the results calculated by the proposed ROM and finite volume method (FVM) is less than 0.06 degrees C. In addition, the established ROM is proved efficient with the computation speed of the ROM being 128 times and 2749 times faster than that of FVM in the two examples respectively. (C) 2013 Elsevier Ltd. All rights reserved.