Abstract:

Electrical discharges in vacuum/low pressure environments are one of the most critical restrictions in the efficiency/reliability of the three 1 MeV - 16 MW Neutral Beam Injectors (NBI) which will be installed in the International Thermonuclear Experimental Reactor (ITER). These NBIs are fundamental to attain the breakeven condition in the ITER thermonuclear Plasma. The ITER NBIs are complex systems, with very demanding ratings, much larger than any other NBI ever built so far (2 times the acceleration voltage, 10 times the negative ion current extracted, pulse duration 1 h). For this reason, the Neutral Beam Test Facility was built in Padova at the Consorzio RFX Laboratory. This facility consists of two experiments: (1) MITICA, aimed at testing the NBI system to demonstrate the feasibility to obtain 16 MW neutral beam @ 1 MeV, and (2) SPIDER aimed at optimizing the negative ion source designed to demonstrate all ITER source requirements. The goal of the project is to analyze the existing data of the Consorzio RFX experiments in order to develop data driven models for discharge prediction. The student will write algorithms for data exploration with unsupervised machine learning and discharge prediction with supervised machine learning. Additionally, the results will be validated with Explainable-Artificial Intelligence (Explainable AI). Initially, existing signals of the MITICA experiment, including voltage, current, pressure, residual gas, X Ray dose, and energy spectra will be analyzed. By implementing machine learning models, used for discharge prediction in radio frequent cavities at CERN, it should be investigated whether the existing probabilistic model [1] can be improved.

Tasks:

• Inspect and understand the existing signals with the underlying physics behind it.
• Implement and compare suitable algorithms for discharge prediction on the MITICA experiment
• Investigation of additional opportunities for applied machine learning at the SPIDER experiment

Your Profile:

• Motivation and interest in the topic
• Experience in python
• Background in machine learning or physics

Additional Information

• Throughout the project a collaboration with the machine learning groups of Consorzio RFX and CERN will be essential.

Contact

• Christoph Obermair (christoph.obermair@cern.ch)
• Franz Pernkopf (pernkopf@tugraz.at or 0316 873 4436)

[1] Pilan N., et al. “Study of high DC voltage breakdown between stainless steel electrodes separated by long vacuum gaps”, Nuclear Fusion, 2020