Contents lists available at ScienceDirect Progress in Nuclear Energy journal homepage: www.elsevier.com/locate/pnucene Adaptive simulation for failure identification in the Advanced Lead Fast Reactor European Demonstrator Pietro Turati a , Antonio Cammi b , Stefano Lorenzi b , Nicola Pedroni c , Enrico Zio a,b,* a Chaire Systems Science and the Energy Challenge, Fondation Electricite’ de France (EDF), Laboratoire Genie Industriel, CentraleSupélec, Université Paris-Saclay, Grande voie des Vignes, 92290 Chatenay-Malabry, France b Energy Department, Politecnico di Milano, Via La Masa 34, Milano, 20156, Italy c Energy Department, Politecnico di Torino, Corso Duca degli Abruzzi 24, Torino, 10129, Italy ARTICLE INFO Keywords: Advanced lead fast reactor european demonstrator (ALFRED) Accident analysis Critical regions exploration Adaptive simulation Polynomial chaos expansion Kriging Design of experiments ABSTRACT The identification undesired or abnormal states of a nuclear power plant is of primary importance for defining accident prevention and mitigation actions. To this aim, computational models and simulators are frequently employed, as they allow to study the system response to different operational conditions. For complex systems like the nuclear power plants, this is in general challenging because the simulation tools are i) high-dimensional; ii) black-box; iii) dynamic and iv) computationally demanding. In this paper, an adaptive simulation framework recently proposed by some of the authors is tailored for the analysis of accident scenarios involving the control system of the Advanced Lead-cooled Fast Reactor European Demonstrator (ALFRED). The results confirm that the adaptive simulation framework proposed is effective in identifying critical re- gions of operation with a limited number of calls to the computationally expensive model. The time of occur- rence and magnitude of the failures of the components of the control system are identified as key factors to characterize the critical regions. In particular, it is shown that the order of occurrence of the components’ failures strongly affects the evolution of the accident scenarios. 1. Introduction The Lead-cooled Fast Reactor (LFR) has been selected by the Generation IV International Forum as one of the candidates for the next generation of Nuclear Power Plants (NPPs). This innovative nuclear system can offer a great potential for plant simplification and reach higher operating efficiency compared to nuclear concepts employing other coolants. On the other hand, it introduces new and different safety concerns and design challenges. To address these, computational models are used particularly for identifying undesired or abnormal states (Turati et al., 2015, 2017a; Zio, 2016). Indeed, modeling and simulation allows investigating the response of the system in different scenarios and transients, under uncertain conditions, including possibly hazardous ones. Design-Of-Experiment (DOE) approaches have been proposed to analyze the system response with respect to specified performance criteria, e.g. of safety, reliability, resilience, business continuity, etc. (Santner et al., 2003; Simpson et al., 2001; Zeng and Zio, 2017). The interest lies in the identification of the factors, parameters and variables values that lead the system to undesired conditions or deviations from operational limits ( Bier et al., 1999; Zio, 2016). In this paper, we focus on system responses represented by mathe- matical models of the form = Y fX ( ). Within this setting, we are in- terested in identifying the Critical Region (CR) formed by the set of input configurations X that lead the output safety-significant para- meters Y to cross a given safety threshold, i.e., = ∈ > CR x D st fx Y { . . () } X thres , where Y thres represents the physical threshold beyond which the system fails in an undesired state. For ex- ample, for the safe operation of a steam generator it is necessary that the pressure does not exceed an upper design limit value. Indeed, a possible strategy to discover the CRs is to resort to a large number of model simulations and a posteriori retrieve the information of interest. Several types of DOE have been proposed to span as uni- formly as possible the input space, in order to have a global exploration of the I/O relation. Latin Hypercube Sampling (LHS) (Iman, 2008; McKay et al., 1979) and Quasi Monte Carlo (QMC) sampling such as Sobol’ sequences (Sobol, 1976), are among the best known (Chen et al., 2006). However, although they have been designed for efficiently https://doi.org/10.1016/j.pnucene.2017.11.013 Received 26 April 2017; Received in revised form 9 November 2017; Accepted 21 November 2017 * Corresponding author. Chaire Systems Science and the Energy Challenge, Fondation Electricite’ de France (EDF), Laboratoire Genie Industriel, CentraleSupélec, Université Paris- Saclay, Grande voie des Vignes, 92290 Chatenay-Malabry, France. E-mail addresses: enrico.zio@polimi.it, enrico.zio@centralesupelec.fr (E. Zio). Progress in Nuclear Energy 103 (2018) 176–190 0149-1970/ © 2017 Elsevier Ltd. All rights reserved. T