A real-time architecture for the identification of faulty magnetic sensors in the JET tokamak

In a tokamak, the accurate estimation of the plasma boundary is essential to maximise the fusion performance and is also the first line of defence for the physical integrity of the device. In particular, the first wall components might get severely damaged if over-exposed to a high plasma thermal load. The most common approach to calculate the plasma geometry and related parameters is based in a large set of different types of magnetic sensors. Using this information, real-time plasma equilibrium codes infer a flux map and calculate the shape and geometry of the plasma boundary and its distance to a known reference (e.g. first wall). These are inputs to one or more controllers capable of acting on the shape and trajectory based in pre-defined requests. Depending on the device, the error of the estimated boundary distance must usually be less than 1 centimetre, which translates into very small errors on the magnetic measurement itself. Moreover, asymmetries in the plasma generated and surrounding magnetic fields can produce local shape deformations potentially leading to an unstable control of the plasma geometry. The JET tokamak was recently upgraded to a new and less thermally robust all-metal wall, also known as the ITER-like wall. Currently the shape controller system uses the output of a single reconstruction algorithm to drive the plasma geometry and the protection systems have no input from the plasma boundary reconstruction. These choices are historical and were due to architectural, hardware and processing power limitations. Taking advantage of new multi-core systems and of the already proved robustness of the JET real-time network, this paper proposes a distributed architecture for the real-time identification of faults in the magnetic measurements of the JET tokamak. Besides detecting simple faults, such as short-circuits and open-loops, the system compares the expected measurement at the coil location and the real measurement, producing a confidence valu- . Several magnetic reconstructions, using sensors from multiple toroidally distributed locations, can run in parallel, allowing for a voting or averaging scheme selection. Finally, any fault warnings can be directly fed to the real-time protection sequencer system, whose main function is to coordinate the protection of the JET's first wall.