Simplified modeling of the heat transfer at the magnets - cryoplant interface in a superconducting tokamak

Tokamak nuclear fusion reactors based on the magnetic confinement of the plasma, such as the ITER, under construction in France, are characterized by very large temperature differences (from the 108 K of the plasma core to the few K of the superconducting magnets) over a few meters. Heat transfer is therefore a critical issue, especially at the interface between different reactor subsystems, for instance the magnet – cryoplant interface, at which pulsed heat loads coming from the superconducting coils during tokamak operation are released to the cryoplant. The smoothing of these variable heat loads is a critical issue in order to avoid the oversizing of the refrigerators as well as to ensure a stable operation with power loads as constant as possible. Experimental devices, like the HELIOS loop at CEA Grenoble (France), as well as detailed computational tools, such as the 4C code, have been used to study possible control strategies to smooth the pulsed heat loads to the cryoplant. These tools can provide a detailed overview of the magnets dynamics, up to the conductors cooling channels, requiring a large computational effort to solve sets of partial non-linear differential equations. However, from the cryoplant side, simplified models of the magnetic system response to the different operating scenarios, predicting the evolution of the heat load, are sufficient to carefully design the control scenarios and strategies. Simplified models based on Artificial Neural Networks (ANNs) have been developed and tested on cryogenic loops, and have proved to be fast and accurate in predicting the power released from the magnets to the cryoplant during nominal operation, i.e. without any regulation acting on the system. The present work proposes a new approach for a simplified modeling of the heat transfer to the cryoplant, based again on ANNs, easy to handle and able to cope with control and regulation of the plant. The feasibility of the new approach is demonstrated and validated on different HELIOS test cases, adopting the state-of-the-art 4C code as source of data for the ANN training. Next, an optimized ANN-based model is developed for the ITER CS and TF magnets and its predictive capability demonstrated against 4C data. Thanks to their high speed of computation, the ANN-based models are used to address different control strategies for the smoothing of the pulse heat loads released to the cryoplant and perform fast parametric studies for the optimization of the control and regulation acting on the system. Once the best control strategy has been identified, a detailed 4C simulation is used as a benchmark to prove the accuracy of the ANN-based model.