Analyse statistique pour la caractérisation radiologique des déchets radioactifs au sein des accélérateurs de particules

par Biagio Zaffora

Projet de thèse en Sciences des matériaux

Sous la direction de Jean-Pierre Chevalier et de Catherine Luccioni.

Thèses en préparation à Paris, CNAM , dans le cadre de SMI - Sciences des Métiers de l'Ingénieur , en partenariat avec Pimm - Laboratoire Procédés et ingénierie en mécanique et matériaux (laboratoire) et de CoMet : Comportement et microstructure des Métaux (equipe de recherche) depuis le 02-06-2014 .


  • Résumé

    Le projet de thèse vise a développer une méthode globale pour la caractérisation radiologique des déchets radioactifs TFA au CERN. La méthode développée se base sur des calculs Monte Carlo et analytiques, sur l'échantillonnage et sur la mesure. Le processus de caractérisation commence par l'identification d'un déchet radioactif et par l'estimation d'un inventaire radiologique par calcul. Ensuite les activités sont évaluées soit par mesure directe soit en employant la méthode des facteurs de corrélation. Une description complète du projet de thèse est donnée dans le résumé en anglais.

  • Titre traduit

    Statistical analysis for the radiological characterization of radioactive waste in particle accelerators


  • Résumé

    1 General context Radioactive waste is produced at CERN as a result of maintenance, upgrade and dismantling of experimental areas. Prior to elimination towards the final repositories, the radioactive waste must be radiologically characterized. In particular, the radionuclide inventory (which is the list of the radionuclides with their activities) must be established by means of measurements and calculations for each item of waste. Such inventory is of outmost importance also for the classification of future waste and for an assessment of the risks related to temporary storage of radioactive waste. This thesis will focus on the development of new techniques for statistical analysis and sampling for the characterization of radioactive waste. Such method will provide a preliminary radionuclide inventory for radioactive waste which is still in the accelerator facilities, and validate the final inventory for waste which has already been characterized and which is temporarily stored in the CERN treatment centre. In particular, the preliminary inventory will be established by using prior information on the irradiation scenarios and measurement of waste samples. In addition, sampling and statistical analysis will be used to validate the final inventory, which is obtained via analytical calculations and radiological measurements. 2 The scaling factor method The scaling factor (SF) method is a technique used to quantify the activity of radionuclides which are difficult-to-measure (or DTM) following the definition in [1]. This technique relies on the foundation that a correlation between an easy-to-measure radionuclide (ETM) and a DTM exists. The SF technique is widely used for the radiological characterization of waste produced in nuclear power plant. This thesis will tests the applicability of the SF method to radioactive waste produced in particle accelerators. In particular, it will address the use of the method for the radiological characterization of historical and future waste, with a special focus on very-low-level (TFA) metallic waste activated in hadron accelerators. The tools needed for the calculation and the quantification of the SFs are introduced. In particular, the calculation code Actiwiz [2], the Monte Carlo code Fluka [3] [4] and the current techniques for the measure of gamma and beta emitters are described together with their applications. 3 Sampling techniques for radioactive waste Statistical techniques for a representative radiological and chemical sampling are presented. A selection of such techniques will be tested on the field for sampling TFA metallic waste produced at CERN. The approaches used in the frame of this study are based on data analysis and cover classical statistical models, Gy's sampling theory [5] and exploratory data analysis. After the collection of data using classical statistical methods such as probabilistic sampling (simple random sampling, systematic sampling, stratified and cluster sampling [6]) or authoritative sampling (convenience sampling, quota sampling, judgemental sampling and biased sampling [7]), data analysis approaches are used to recognize the most appropriate statistical model to represent data. In addition, Gy's sampling theory is introduced with the purpose of minimizing errors during the sampling process of materials. A rational procedure is introduced in order to choose the most appropriate sampling technique in the characterization process, by taking into account time and financial cost and impact on the uncertainty budget. This procedure will lead to traced optimization and justification of the operations carried out for the elimination of waste. 4 Radionuclide inventory A fundamental outcome of the work performed for the thesis is the list of radionuclides of interest for the radiological characterization of metallic waste. This list is called radionuclide inventory and contains the most important contributors to the so-called IRAS factor. The IRAS factor, which is associated with a waste package or a batch of packages, is a quantity that reflects the radiotoxicity of a waste and depends on the radionuclide inventory and levels of specific activity. This quantity must be calculated for the elimination of waste towards final repositories. The calculation of the radionuclide inventory depends on the material composition of the waste and on the irradiation scenarios. A comprehensive list of more than 50 materials commonly used at CERN is investigated. This list includes different grades of steel, iron, aluminium, copper and titanium. For each grade considered, the possible key-nuclides (KN) [1] [8] [9] are identified together with the complete list of DTMs. The contribution of each radionuclide to the IRAS factor is estimated. 5 Statistical learning techniques and calculation of scaling factors Statistical learning techniques can be used to estimate the value of scaling factors, to understand the relative importance of the input parameters in the activation process, to visualize sorting criteria for future waste, to estimate confidence interval for the scaling factors and to quantify the overall error on the calculation of the Hazard Factors. The thesis introduces common statistical learning techniques, such as decision trees, bagging, random forests, boosting and linear models [10], to study the behaviour of scaling factors in hadron accelerators. The calculation of SFs for historical and future metallic waste is performed based both on theoretical calculations and analysis of results from measurements. Extensive applications of the methods introduced will be described together with a detailed calculation of the errors in the radiological characterization process. 6 Conclusions The present work aims at establishing new techniques for the characterization of TFA metallic waste produced at CERN, based on statistical calculations and measurements methods. The research contributions expected are the radionuclide inventory for low-level metallic waste and the correlation study between DTMs and KNs for the radionuclides produced in hadron accelerators, the calculation of scaling factors for historical and future waste and a structured study of the errors in the radiological characterization process. 7 Bibliography [1] ISO 21238, "Nuclear energy - Nuclear fuel technology - Scaling factor method to determine the radioactivity of low and intermediate level radioactive waste packages generated at nuclear power plants," International Standard Organization, Geneva, 2007. [2] C. Theis and H. Vincke, "Actiwiz - Optimizing your nuclide inventory at proton accelerators with a computer code, Proceedings of the ICRS12 conference," Nara (Japan), 2012. [3] G. Battistoni, "The FLUKA code : Description and benchmarking," in Proceedings of the Hadronic Shower Simulation Workshop, Fermilab, 2006. [4] A. Ferrari, "FLUKA: a multi-particle transport code," CERN-2005-10, Geneva, 2005. [5] F. F. Pitard, Pierre Gy's sampling theory and sampling practice; heterogeneity, sampling correctness and statistical process control, Boca Raton, London, New York, Washington D.C: CRC Press, 1993. [6] R. O. Gilbert, Statistical methods for environmental pollution monitoring, New York: Van Nostrand Reinhold Company, 1987. [7] W. G. Cochran, Sampling techniques, New York, Chichester, Brisbane, Toronto, Singapore: John Wiley & Sons, 1977. [8] IAEA, "Strategy and methodology for radioactive waste characterization; IAEA-TECDOC-1537," IAEA, Vienna, 2007. [9] IAEA, "Determination and use of scaling factors for waste characterization in nuclear power plants; IAEA Nuclear Energy Series No. NW-T-1.18," IAEA, Vienna, 2009. [10] T. Hastie, R. Tibshirani and J. Friedman, The elements of statistical learning; data mining, inference and prediction, Stanford: Springer, 2008.