When simulating electric machines, the use of lowest order Finite Elements (FE) is the classical discretization approach due to the simplicity of meshing and local refinement, see Figure 1. However, triangular elements fail to represent circular sections exactly and derivative based optimization becomes almost impossible due to the necessity of remeshing. These issues can be overcome in an Isogeometric Analysis (IGA) setting, where the geometry mapping is predefined by Non-Uniform Rational B-Splines (NURBS), see Figure 2, allowing for exact geometry representation and analytic derivatives. Here, on the other hand, local refinement can be challenging. The task for this thesis is to combine the advantages of both methods and create an enhanced discretization scheme, where NURBS are used for the geometry mapping, but lowest order FE basis functions are used for the computation. This way, the mapping is exact and the FE space locally refined. As a consequence, the model does not need to be remeshed for geometry changes, making parameter studies or optimization very efficient.