3D modelling of the R&W TF coil design for EU-DEMO including the winding pack and frictional effects

The SPC team has proposed a design for the Toroidal Field (TF) coil of the DEMO tokamak based on the use of react & wind (R&W) Nb3Sn superconductor. Thanks to the grading of both the superconductor and the steel in the jackets the size of the TF coil was considerably reduced and hence the overall cost. The mechanical analysis in the recent works was focused on using 2D models and consideration of the equatorial plane only. In this work a full 3D model was built accounting for the casing, detailed geometry of the winding pack (WP) as well as the forces from the PF (poloidal field) coils and the vertical force from the central solenoid (CS). The WP was meshed including the superconductor, the jackets and the electrical insulation modeled with orthotropic material and the ground insulation. The cable details were neglected and modeled with an artificially soft isotropic material E=0.1 GPa. Two non-linear contacts were included in the model, between the external surface of the WP and the casing, and between the casing and the insulation between neighboring TF coils in the toroidal direction. Thanks to the symmetry, only one TF coil was analyzed assuming periodic boundary conditions. The Lorentz forces were computed based on a global electromagnetic (EM) model of the DEMO tokamak and mapped on the mechanical mesh. Two load conditions were considered in the study: cool-down and EOF (end-of-flat top), being the worst case electromagnetic condition. Due to the uncertainty of the friction coefficient between the winding pack and the inner surface of the steel casing a parametric study was performed to understand the redistribution of the electromagnetic load between the winding pack and the casing. The friction coefficient µWP was changed from 0 (frictionless contact), through frictional contact with µWP={0.1, 0.2}, up to a fully bonded interface µWP=∞. A probabilistic approach was also proposed to compute an expected value of the stress state and its standard deviation in the whole domain as a results of a given probability density of the friction coefficient. The computational cost of the model was analyzed with the Ansys Mechanical APDL 2024R1 solver, reveling that the iterative solver is 3x faster than the direct solver (DS). Furthermore the change of the convergence tolerance of the iterative solver from 1e-8 to 1e-4 revealed an additional speed-up of 2.6x without the loss of the results accuracy. The total simulation time of the developed detailed 3D model was below 2h on a 8 CPU workstation.