In the framework of the development of new passive safety systems for the second and third generations of nuclear reactors, the numerical simulations, involving complex turbulent two-phase flows around thin or massive inflow obstacles, are privileged tools to model, optimize and assess new design shapes. In order to match industrial demands, computational fluid dynamics tools must be the fastest, most accurate and most robust possible. To face this issue, we have chosen to solve the Navier-Stokes equations using a projection scheme for a mixture fluid coupled with an Immersed Boundary (IB) approach: the penalized direct forcing method - a technique whose characteristics inherit from both penalty and immersed boundary methods - adapted to infinitely thin obstacles and to a Finite Element (FE) formulation. Various IB conditions (slip, no-slip or Neumann) for the velocity on the IB can be managed by imposing Dirichlet values in the vicinity of the thin obstacles. To deal with these imposed Dirichlet velocities, we investigated two variants: one in which we use the obstacle velocity and another one in which we use linear interpolations based on discrete geometrical properties of the IB (barycenters and normal vectors) and the FE basis functions. This last variant is motivated by an increase of the accuracy/computation time ratio for coarse meshes. As a first step, concerning academic test cases for one-phase dilatable-fluid laminar flows, the results obtained via those two variants are in good agreement with analytical and experimental data. Moreover, when compared to each other, the linear interpolation variant increases the spatial order of convergence as expected. An industrial test case illustrates the advantages and drawbacks of this approach. In a shortcoming second step, to face two-phase turbulent fluid simulations, some methodology modifications will be considered such as adapting the projection scheme to low-compressible fluid and immersed wall-law boundary conditions.