""The computation of magnetic fields produced by coils with complex 3D geometrical shapes may represent a quite demanding computational task when highly accurate fi\\feld maps are required in a large number of points, an example being the magnetic computation in fusion devices.. As a matter of fact, approximate numerical methods such as \\fnite elements or boundary elements can hardly provide suitable approximations of the coils shapes with discretization levels compatible to large scale volumes; therefore they can only provide an approximate response.. On the other hand, if superposition can be applied (linear behavior of ferromagnetic materials are. admissible, or just the magnetic \\feld due to current sources needs to be computed), the coils can be. modelled by an arbitrarily re\\ffined superposition of elementary sources (e.g. massive blocks with uniform current density or current sticks, if conductor-level discretization of the coil winding is required).. Such elementary sources do possess simple analytical or semi-analytical expressions for magnetic fi\\feld,. and are then well suited for a high accuracy \\ffield computation.. Unfortunately, when the discretization level becomes very refined, the computational times gets high,. and a massively parallel architectures can provide the necessary speed up to achieve results in reasonable times. In particular, the GPU*s are well suited for this task since they can be used as a vector mathematical coprocessor where each processing element computes the contribution of a dierent. elementary source.. In the paper the implementation on GPU architectures of a magnetic \\feld computation code based. on current sticks, called MISTIC, will be presented and some application examples on ITER TFC and PFC accurate field computation will be critically discussed.""
Fast magnetic field computation in fusion technology using GPU technology
CHIARIELLO, Andrea Gaetano;FORMISANO, Alessandro;MARTONE, Raffaele
2013
Abstract
""The computation of magnetic fields produced by coils with complex 3D geometrical shapes may represent a quite demanding computational task when highly accurate fi\\feld maps are required in a large number of points, an example being the magnetic computation in fusion devices.. As a matter of fact, approximate numerical methods such as \\fnite elements or boundary elements can hardly provide suitable approximations of the coils shapes with discretization levels compatible to large scale volumes; therefore they can only provide an approximate response.. On the other hand, if superposition can be applied (linear behavior of ferromagnetic materials are. admissible, or just the magnetic \\feld due to current sources needs to be computed), the coils can be. modelled by an arbitrarily re\\ffined superposition of elementary sources (e.g. massive blocks with uniform current density or current sticks, if conductor-level discretization of the coil winding is required).. Such elementary sources do possess simple analytical or semi-analytical expressions for magnetic fi\\feld,. and are then well suited for a high accuracy \\ffield computation.. Unfortunately, when the discretization level becomes very refined, the computational times gets high,. and a massively parallel architectures can provide the necessary speed up to achieve results in reasonable times. In particular, the GPU*s are well suited for this task since they can be used as a vector mathematical coprocessor where each processing element computes the contribution of a dierent. elementary source.. In the paper the implementation on GPU architectures of a magnetic \\feld computation code based. on current sticks, called MISTIC, will be presented and some application examples on ITER TFC and PFC accurate field computation will be critically discussed.""I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.