Newsletter n°3
Subscribe to the CORTEX newsletter and get fresh news about the project!
We'd love to inform you about the latest achievements and results generated in the CORTEX project, and share with you our latest reports, scientific publications and upcoming events. If you haven't subscribed yet and would like to continue hearing from us please click the button below to opt in. It only takes one minute!
Word from the coordinator - Christophe Demazière
Time flies and the CORTEX project is now entering its fourth and last year. The project is on track, despite the outbreak of the COVID-19 pandemic, with all major code and method developments completed or close to be completed.
The third year was marked by very tight collaborations between, on one hand, the experimentalists and code developers (Work package [WP] 1 and 2) for code validation, and, on the other hand, the code developers, and signal processing and artificial intelligence experts (WP1 and WP3) for the application of the methods to actual plant data (WP4).
For code validation, new experiments at both the CROCUS and the AKR-2 facilities were performed. The analysis of those experiments allowed to gain a deeper understanding of the physics, itself benefiting the modelling, and new research questions for future work and new modelling approaches were identified. I would also like to express my gratitude to the EPFL and TUD teams, who managed to perform the AKR-2 new measurements during the pandemic. The dedication of those teams is highly appreciated.
For the application of the methods to actual plant data, the techniques developed during the two first years of the project were merged and new sets of simulation data specific to the selected reactors at chosen conditions were created. The completeness of the simulated scenarios corresponding to possible anomalies also resulted in extremely large amount of data, which represented a major challenge for the machine learning architectures earlier developed in the project. Early results of the unfolding of anomalies from the induced neutron noise measured at the plants confirmed the capabilities of the developed techniques for anomaly detection, classification, and localization. Further tests are being carried out to fine tune the methods, and new measurement data sets will be soon processed.
The major developments of the CORTEX project will be presented during a final workshop, to be held on 21 and 22 June 2021 in Sweden (or remotely, if the sanitary situation requires it). The workshop is open to the entire community. I hope to see as many of you at this workshop, either IRL or remotely!
Christophe Demazière
CORTEX Coordinator
CORTEX had its fourth annual meeting from 5 to 9 October 2020. The meeting took place virtually due to the Covid-19 situation. 
During the whole week, consortium members presented the results and achievements of the project showing that the CORTEX project is progressing very well and on schedule.
Read more
WP1 & WP2 - Paolo Vinai & Mathieu Hursin
WP1 Comparison of neutron noise solvers based on numerical benchmarks
In WP1 of the CORTEX project, numerical solvers have been developed for the analysis of neutron noise problems. They are based on Monte Carlo and deterministic (higher-order transport and diffusion) methods.
For the study of their validity and limitations, an extensive verification and validation work has been implemented and includes the simulation of numerical exercises and experiments. An effort has recently started to compare the developed solvers over neutron noise benchmarks defined in a 2-D simplified UOX fuel assembly (see Figure 1, picture on the left). The noise sources used for the exercises include an overall oscillation of the properties of a fuel pin, a superposition of different oscillations of various cross sections in a fuel pin, fluctuations of cross sections describing the mechanical vibration of one fuel pin or clusters of fuel pins. A preliminary result is shown in Figure 1 (plot on the right), for the case of a superposition of oscillations of various cross sections in a fuel pin.
The stochastic solver embedded in the Monte Carlo code TRIPOLI-4® (CEA), the Monte Carlo solver developed by Kyoto University, the IDT solver embedded in APOLLO3® (CEA), and the discrete ordinates solver developed by Chalmers are in good agreement. The calculations obtained from the diffusion-based solvers CORE SIM+ (Chalmers) and FEMFFUSION (UPV) reproduce the main features of the induced noise, although quantitative discrepancies are found. These discrepancies can be significant close to the neutron noise source, i.e. where the diffusion approximation is expected to be less reliable.
Figure 1 - Simplified UOX fuel assembly where the black circle identifies a perturbed fuel pin (left) and relative neutron noise calculated along the diagonal crossing the perturbed fuel pin (right)
WP2 Generation of high quality neutron noise experimental data
WP2 targets the generation of high quality neutron noise experimental data for the subsequent validation of computers methods and models developed in WP1.

Four experimental campaigns (out of the six planned) have been carried out so far, two at each facility, CROCUS and AKR-2. A view of the experimental hall with the various data acquisition systems used during the latest campaign at AKR-2 in July 2020 can be found in Figure 2. Even though this campaign happened during the COVID-19 epidemic, it was carried out successfully and on schedule by the TUD team with the support of the EPFL experimentalists. During this campaign, and for the first time since the beginning of the project, two noise sources were imposed simultaneously: a vibrating absorber in the central channel and an absorber of variable strength in a tangential channel.
Figure 2 - AKR-2 experimental hall during the second CORTEX campaign
On the hardware side, the latest campaign at AKR-2 allowed testing new miniature scintillators in development at EPFL: the small size of such detector proved very useful during the campaign as it allowed measuring the neutron noise in the close vicinity of the AKR-2 core. Their extensive use is planned for the third campaign in CROCUS.

In the past year, large efforts were dedicated to the processing and analysis of the experimental data generated during the first campaigns. It allowed the determination of reliable experimental values with reduced uncertainty estimates, needed to perform a meaningful validation exercise. Moreover, those reduced uncertainties allowed demonstrating that a deviation from the point kinetics behaviour of the noise was observed in CROCUS. Due to the small size of the reactor, such deviation was not expected to be large.

The validation of the noise simulators developed in WP1 continued. Recently, uncertainty estimate for the considered quantity of interests became available for CORESIM+, completing the list of required ingredients for a meaningful validation exercise. Currently, the difference between simulations and measurements is not covered by the computational and experimental uncertainties but continuous improvements of the models allows for optimism that these deviations will be explained and hopefully resolved during the fourth year of the project.

The frequent exchanges between modellers and experimentalists are key in this respect as they have allowed so far to make progress towards assessing the accuracy of the research facilities models and understanding some of the discrepancies between simulations and experiments, especially when considering the higher harmonics of a perturbation at a given frequency.
WP3 - Detecting anomalies in nuclear reactor core - Stefanos Kollias
WP3 focuses on developing techniques that allow detecting anomalies in nuclear reactor cores, such as abnormal vibrations of fuel assemblies and core internals, flow blockage, coolant inlet perturbations. The techniques are based on monitoring of the neutron noise (fluctuations in neutron flux recorded by in-core and ex-core neutron instrumentation) in a non-intrusive way. WP3 developments have been completed in February 2020 and efforts have been thereafter focused on applying the generated techniques to real plant data, within WP4 implementation.

WP3 generated a variety of signal processing and analysis methods, in the frequency and time domains, in order to find coherence and phase correlation relationships between the different noise signals, using signal transformations, based on the Hilbert-Huang and the Discrete Wavelet Transform. It also included the generation of a rich machine and deep learning framework for the unfolding of reactor transfer functions to enable the classification and localization of multiple and simultaneous reactor perturbations.

The machine learning approach involved semantic segmentation, i.e., classification of each voxel in the input data to a semantic (class) perturbation label. The basis of the approach, which is used in WP4 to adapt from simulated to real reactor readings, is a 3-D fully-convolutional encoder-decoder segmentation network, generating a prediction mask of perturbation classes, with each voxel representing the source at which that perturbation occurred. The ability to simultaneously detect and localize multiple reactor core perturbations for, e.g., a 32x32x34 voxel space, given only 48 in-core and 8 ex-core detectors, is a non-trivial task, which has been successfully tackled in WP3 and evaluated on a large variety of simulated input data cases.
Figure 3 - The Figure illustrates a combined network of voxel-wise semantic segmentation and a 3-D Convolutional Neural Network (CNN). The output features of the pre-trained 3-D CNN are concatenated with the final layer of the encoder network. The network produces four volumetric predictions: perturbation classification, k-theta vibration parameters and relative perturbation amplitude.
WP4 - Analyzing power reactor plant data - Joachim Herb
In WP4 actual plant data is used to demonstrate the applicability as well as usefulness of the tools newly developed within WP1 and WP3 of CORTEX. The tools are used to analyse data recorded at different nuclear power plants representing different reactor types from different European countries (a 3-loop and a 4-loop pre-Konvoi PWR and a VVER440 and a VVER1000).

In the past year of the CORTEX project the partners have worked together intensively in WP4. Input to WP4 were the newly developed tools and reactor models of WP1 for simulating the behaviour of the neutron flux and the methods for signal processing and machine learning developed in WP3.

GRS and TU Dresden applied the enhanced tools for simulating the interactions between the coolant and the structures in the reactor core to simulate the resulting movements of the core structures and the resulting fluctuations of the neutron flux.

Enhanced neutron physics simulations were performed by Chalmers, PSI, UPV to generate massive amounts of data as training input for the machine learning tools. These simulations were based on different hypotheses of perturbations, which could cause fluctuations of the neutron flux. For the VVER reactors simulations were performed using the tool FEMFFUSION while for the pre-Konvoi reactors the tools CORE SIM+ and SIMULATE-3K were used.

Measurements from the different reactors mentioned above were distributed to the organisations responsible for applying signal analyses methods developed within WP3 (UJV, MTA EK, UPM, CEA). They used different enhanced and newly developed methods to analyse those data. Besides the traditional Fourier analyses, more sophisticated methods like Singular Value Decomposition and Singular Spectrum Analysis were used. These methods were also used to pre-process the measurement data for the application of the machine learning tools.

The project partners UoL and ICCS-NTUA applied the newly developed machine learning methods to data measured at the power reactors. They used the measurements of the nuclear power plants, the pre-processed data from the partners who had developed the pre-processing methods and the training data created by simulations. Different machine learning approaches were adopted: one was applying various clustering techniques to group signals together, as well as performing anomaly detection. Another was using different neural network methods for classification and localization of perturbations. The results were the classification and subsequent localization of multiple, simultaneously occurring perturbations. Emphasis has been placed on robust prediction of perturbation type, location, and parameters ensuring effective adaptation of learnt knowledge to real plant data.
Figure 4 shows examples for the kind of identifications of perturbations a neutral network makes when applied to the measurements of a nuclear power reactor. It had been trained with simulated data for different hypotheses about possible perturbations. The figure shows what kind of the predicted perturbations and where they were located within the reactor core. In the next step these predictions will be used to simulate the same kind of perturbations and compare the simulation results with the actual measurements to assess the quality of the predictions.
Figure 4 - The Figure illustrates examples for the kind of identifications of perturbations a neutral network makes when applied to the measurements of a nuclear power reactor 
Despite the fact, that there is still one year to go in the project, the tools and methods developed within CORTEX have already shown their applicability to real plant data. This was documented in the publicly available deliverable 4.4. In the remaining year of the project, more applications of the tools to real plant data are planned. This will lead to a deeper understanding of the different causes for neutron flux fluctuations in nuclear power plants.
WP5 - Knowledge dissemination and education - Christophe Demazière
Two short courses and an internal workshop arranged by CORTEX
During the third year of the project, several short courses and workshops were organized by the consortium.
On 4-6 December 2019, a short course was arranged by the Technical University of Munich, Germany, during which the principles of uncertainty and sensitivity analysis were presented, as well as their specific applications to the case of neutron noise calculations, taking real examples from the project.
On 10 and 11 September 2020, another short course on computer simulations of neutron noise problems was given by the Paul Scherrer Institut, Switzerland. Because of the COVID-19 pandemic, the course was entirely and successfully given online, with many attendees.
The pandemic was also behind the decision to transform an internal workshop between experimentalists and modellers, originally planned on 12 and 13 March 2020 in Munich, into a two-day fully online event. Despite the online nature of the workshop, having the possibility to present and discuss the advancement of the modelling of the various experiments was extremely beneficial. New research questions for future work were also identified.
In the remaining year of the project, three additional events open to the entire community will be organized: another workshop on the experiments carried out in the research reactors and on the validation of the neutronic tools, a hands-on training session on the CROCUS reactor at l’Ecole Polytechnique Fédérale de Lausanne, Switzerland, and the final CORTEX workshop, during which the methods for reactor noise analysis will be demonstrated to the community.
In case you want to get further news on those events, follow us on LinkedIn and on our website.
Final CORTEX workshop on the demonstration of the methods for reactor noise analysis against plant data 
The final CORTEX workshop on the demonstration of the methods for reactor noise analysis against plant data will be organised on 21 & 22 June 2021 in Stenungsbaden, Stenungsund, Sweden.
Please note that the CORTEX consortium is closely monitoring the COVID-19 situation and might decide to hold this event online.
This workshop will
  • present the findings and lessons learnt in the CORTEX project
  • produce recommendations on techniques and instrumentations for core monitoring and surveillance to make nuclear units safer and more reliable
  • give an overview of the future in this field

The event will be open to all stakeholders, beyond the consortium partners: academia, research institutes, technical and scientific support organisations, and the nuclear industry as a whole.

Further information and registration will be available on the CORTEX website in the upcoming months.
Communication & dissemination related to the CORTEX project
The third year of the project resulted in many publications and deliverables submitted to the European Commission. All journal articles, conference papers, posters, presentations and public deliverables can be found on the CORTEX website HERE.
Our mailing address:
LGI Consulting
6, Cité de l'Ameublement
75011 Paris
CORTEX received funding from the Euratom Research and Training Programme 2014-2018 under grant agreement No 754316.
This email was sent to .
You received this email because you are a partner or have expressed interest in the CORTEX project.