Copper losses dissipated in the windings of electric machines are the sum of classical ohmic DC losses and additional AC eddy current losses. In fact, the level of eddy current losses is strongly correlated to the manner of disposition of coil conductors in machine slots. Then, to improve the efficiency in electric machines, the selection of an optimal winding configuration becomes substantial. Since eddy current losses derive from the strong electromagnetic coupling between current density and time-dependent magnetic field which cannot be solved easily, numerical analyses such as particularly the one using the finite element method are often used. As for the finite element modeling, it can employ moving band technique to perform the rotor motion and Newton-Raphson iterations to deal with the nonlinear behavior of magnetic circuits. It leads then to a substantial computational time that hinders any process of conception or optimization of winding geometries. To overcome this issue, a 2D finite element model reduction based on the perturbation method is proposed. It starts from one approximate finite element solution of a simplified complete machine modeling to find fast but accurate solutions in slots subdomains when any variation of geometrical or physical data occurs. It allows adapting nonconforming meshes and provides clear advantages in repetitive analyses when we search the optimized winding configuration for a given number of turns.