La quantification et la propagation des incertitudes dans
le plasma de fusion nucléaire (thèse 2023 - 2026)
Activités
Incertitude,
Fusion nucléaire,
Trajectoires des particules énergétiques.
Publications scientifiques au M2P2
2027
Ioannis Mavrogiannis, D. Zarzoso, Mitra Fouladirad. Global sensitivity analysis and uncertainty quantification comparison via surrogate models for auxiliary heating of nuclear fusion plasmas. Reliability Engineering and System Safety, 2027, 278 (part 2), pp.113535. ⟨10.1016/j.ress.2026.113535⟩. ⟨hal-05790337⟩ Plus de détails...
Surrogate models are often used to approximate the statistical behavior of complex, non-linear numerical codes, under the influence of input uncertainty. Knowledge of the code’s performance under uncertainty increases the confidence of the code results in realistic situations where parameters do not have a fixed value. Surrogates furthermore facilitate the accurate estimation of global, variance-based sensitivity indices, with a smaller number of samples that would otherwise be required. In this work, we investigate the uncertainty of surrogate model performance, for surrogates trained under different sampling methods and sample sizes. For each surrogate, an uncertainty range is obtained for its accuracy, estimated statistical moments and Sobol’ indices. Computationally fast physical models are employed, making this comparison possible. The physical system chosen concerns the heating of a nuclear fusion plasma, with neutral beam injection (NBI). Such auxiliary heating systems are necessary for reaching the extreme plasma temperatures at which fusion reactions occur. Uncertainties in the thermal plasma equilibrium and ionization cross-sections are propagated through the TAPaS code (Toroidal Accelerated PArticle Simulator), obtaining quantities of interest that describe the beam attenuation in the plasma, the heating power density transferred to the plasma ions and electrons, and the NBI shine-through losses. Such a GSA-UQ analysis setting therefore increases the safety during the plasma pulse. It allows reliable prediction of the shine-through losses, which can damage the machine walls, and can help avoid potential instabilities. It can help maximize the plasma pulse duration, achieving steady-state production of energy.
Ioannis Mavrogiannis, D. Zarzoso, Mitra Fouladirad. Global sensitivity analysis and uncertainty quantification comparison via surrogate models for auxiliary heating of nuclear fusion plasmas. Reliability Engineering and System Safety, 2027, 278 (part 2), pp.113535. ⟨10.1016/j.ress.2026.113535⟩. ⟨hal-05790337⟩
Journal: Reliability Engineering and System Safety
Alex Karagrigoriou, Ioannis Mavrogiannis, Georgia Papasotiriou, Ilia Vonta. An Exponentiality Test of Fit Based on a Tail Characterization against Heavy and Light-Tailed Alternatives. Risks, 2023, 11 (10), pp.169. ⟨10.3390/risks11100169⟩. ⟨hal-04543899⟩ Plus de détails...
Log-concavity and log-convexity play a key role in various scientific fields, especially in those where the distinction between exponential and non-exponential distributions is necessary for inferential purposes. In the present study, we introduce a testing procedure for the tail part of a distribution which can be used for the distinction between exponential and non-exponential distributions. The conspiracy and catastrophe principles are initially used to establish a characterization of (the tail part of) the exponential distribution, which is one of the main contributions of the present work, leading the way for the construction of the new test of fit. The proposed test and its implementation are thoroughly discussed, and an extended simulation study has been undertaken to clarify issues related to its implementation and explore the extent of its capabilities. A real data case is also investigated.
Alex Karagrigoriou, Ioannis Mavrogiannis, Georgia Papasotiriou, Ilia Vonta. An Exponentiality Test of Fit Based on a Tail Characterization against Heavy and Light-Tailed Alternatives. Risks, 2023, 11 (10), pp.169. ⟨10.3390/risks11100169⟩. ⟨hal-04543899⟩
