This work outlines an approach for localizing anomalies in nuclear reactor cores during their steady state operation, employing deep, one-dimensional, convolutional neural networks. Anomalies are characterized by the application of perturbation diagnostic techniques, based on the analysis of the so-called “neutron-noise” signals: that is, fluctuations of the neutron flux around the mean value observed in a steady-state power level. The proposed methodology is comprised of three steps: initially, certain reactor core perturbations scenarios are simulated in software, creating the respective perturbation datasets, which are specific to a given reactor geometry; then, the said datasets are used to train deep learning models that learn to identify and locate the given perturbations within the nuclear reactor core; lastly, the models are tested on actual plant measurements. The overall methodology is validated on hexagonal, pre-Konvoi, pressurized water, and VVER-1000 type nuclear reactors. The simulated data are generated by the FEMFFUSION code, which is extended in order to deal with the hexagonal geometry in the time and frequency domains. The examined perturbations are absorbers of variable strength, and the trained models are tested on actual plant data acquired by the in-core detectors of the Temelín VVER-1000 Power Plant in the Czech Republic. The whole approach is realized in the framework of Euratom’s CORTEX project.

A strongly energetic spectral component has been observed in the operating noise of a research reactor before the beginning of reactivity initiated accidents (RIA). We identify the source of this component by means of analytical derivations of torsional wave propagation in the test device. Numerical simulations, performed with a spectral-element method, confirm that this component is a resonant frequency of the device, and allow to evaluate the impact of wave travelpath in the test device on the acoustic emission signals recorded during RIA experiments. The transfer function of the test device is strongly dependent on both the source and the receiver locations, which precludes signal processing by deconvolution when the location of the AE source mechanism is unknown. Moreover, the geometry and the configuration of the test device may not influence signals in the low-frequency range, and hence the signals generated by source mechanisms located on the fuel clad such as clad failures. Finally, replacing the heat transfer fluid (pressurized water) by sodium leads to an increase in the recorded signal amplitude and a small shift of the frequency content towards the high frequencies. The results obtained in this paper may be useful to improve design of future RIA experiments. They are also useful to understand wave propagation in the core of the reactor and to choose appropriate signal processing tools for recovering the source of acoustic emission signals generated during RIA experiments.

CABRI is an experimental pulse reactor, funded by the French Nuclear Safety and Radioprotection Institute (IRSN) and operated by CEA at the Cadarache research center. It is designed to study fuel behavior under RIA (Reactivity Initiated Accident) conditions. In order to produce the power transients, reactivity is injected by depressurization of a neutron absorber (3He) situated in the so-called “transient rods” inside the reactor core. The CABRI reactivity injection system allows us to generate structured transients based on specific sequences of depressurization. For such transients, the time difference between the openings of two valves of the reactivity injection system has an important impact on the power pulse shape. A kinetic point code, SPARTE, was developed in order to replace the older DULCINEE code dedicated to the modeling and prediction of CABRI power transients. The SPARTE code includes new models of 3He depressurization based on CFD calculations, variable Doppler coefficient based on Monte Carlo calculations and variable axial neutron flux profile. The density and Doppler models have a large impact on power transient prediction. For low initial pressure transients, the major uncertainty comes from the reactivity injected by the 3He depressurization. For high initial pressure transients, the 3He heating during the power pulse (“TOP effect”) is responsible of an additional injection of reactivity that needs to be modeled precisely.

Fuel rod cladding failure in a nuclear reactor produces different phenomena related to vibrations and fluid–structure interaction. The most significant aspect of those phenomena is the creation of a pressure wave at the failure position and its propagation in the coolant fluid flowing around the fuel rod. An accurate understanding of the propagation of the pressure wave around the fuel rod can help us design a method to detect a failure, determine its position, and estimate some of its characteristics with a single and simple sensor, such as a pressure sensor or a piezoelectric acoustic sensor, that can be mounted relatively far from the failure. Such a method can be useful for the monitoring of nuclear fuel rods, where instrumentation possibilities are restricted (because of neutron flux, radiation, high temperature, and available space) as well as for any kind of application involving annular ducts and limited instrumentation possibilities. The current paper is related to the specific application of nuclear fuel rod monitoring. It deals with preliminary numerical simulations that are necessary to know the evolution of a fluid pressure profile along the system containing the rod. They are carried out by finite element methods, using the EUROPLEXUS code. They provide the necessary information about the propagation of pressure waves around the rod to design measurement and signal processing methods as well as properly interpret experimental results from tests in industrial reactors, research reactors, or experimental mock-ups. They also provide some information that could not be experimentally obtained because of the constraints in a nuclear environment. Despite the specific application we show in this article, similar calculation methods, theoretical observations, and results interpretations can be easily adapted to the other mentioned applications.

The CABRI facility is an experimental pulse nuclear reactor funded by the French Nuclear Safety and Radioprotection Institute and operated by the French Atomic Energy Commission. It is designed to study the behavior of fuel rods at high burnup under reactivity-initiated accident (RIA) conditions, such as a control rod ejection. The distinctive feature of this reactor is its reactivity injection system. The fast depressurization into a discharge tank of ³He (strong neutron absorber) previously introduced inside 96 tubes (so-called transient rods) located among the fuel rods allows us to create a power burst from 100 kW to 20 GW with a full-width at half-maximum of 10 to 80 ms. The total energy deposit in the tested rod is adjusted by dropping the control and safety rods after the power transient. The neutron flux is measured online by compensated boron chambers located outside the reactor and operated in the current mode. These neutron detectors are calibrated during a commissioning phase thanks to standards given by a conventional heat balance. To assess the energy released into the test rod, we had to integrate the driver core power signal measured online. Thus, the beginning of the transient, called transient overpower (TOP) onset, has to be estimated. The TOP onset of a transient test is defined as the instant of the beginning of the test. It is determined by experimentalists during the processing phase. It corresponds to the beginning of the increase of the neutron detector signal, measured by the compensated boron chamber devices sufficiently sensitive at low current levels. So far, the choice of this instant has been realized by a visual choice zooming in the zone of interest, which may induce some shift according to experimentalists. In an attempt to overcome this issue, we put forward in this paper a theoretical method of determination to calculate the TOP onset. The main asset of the method is to formalize the TOP onset determination. Furthermore, it provides the possibility of associating an uncertainty, which is impossible by the manual process. The methodology relies on the fact that at the beginning of the RIA transient, the neutron flux at any point of the reactor core undergoes an exponential evolution as a function of the time. Then, a logarithmic transform allows us to show that the search for the TOP onset is equivalent to solving a nonlinear regression. The methodology has been validated in the last 14 experiments. Moreover, the reactor restarted in October 2015 and now gives us the opportunity to apply this methodology on signals recently acquired and pertaining to the power commissioning phase with a view to preparing the experiment foreseen next year.