The 'DeMaND' project is conducted under the supervision of the Institute of Atomic Physics and is financed through the 5^th Programme - Research in domains of strategic interest, Subprogramme 5.2 - Participation to international organisms and programs in atomic and subatomic field - FAIR-RO Module

Activity reports

   The results obtained in this project have been reported to the Contracting Authority in Annual Summary Document at the end of each stage.

Results obtained in Stage I - November - December 2020 a. for the Comparative analysis of experimental data - simulations using existing reconstruction algorithms activity, we were able to: identify which are the major advantages and drawbacks of the standard algorithms currently used in reconstruction of tracks inside neutron detectors; asses the performances of algorithms for reconstructing neutrons from NeuLAND - the SVA, STA, PNT and R3BRootNT algorithms; comparatively analyze the SAMURAI19 experimental data and simulations using the most efficient existing algorithm. b. for the Identification of reconstruction algorithms based on machine learning activity, we have: identified the machine learning algorithms used in high energy physics; pointed out the major advantages or disadvantages of most-effective algorithms currently used for particle tracks reconstruction; selected the most reliable machine learning algorithm to help fulfill the goals of this project.

These results are detailed in Annual Summary Document for Stage I

Results obtained in Stage II - January - December 2021 identify class managers of different stages in R3BRoot and their usage for simulations, digitization, clustering and reconstruction; evaluate several scenarios of simulated data using R3BRoot with various detector planes configuration, different physics list activated or different features of beam characteristics; evaluate neutron reaction probabilities as function of the number of double planes used for an energy range from 200 MeV to 1 GeV; identify the most common reactions and the most important reaction products for incoming neutrons with various energies using different physics list implementations; determine light output influence on energy deposition in simulated data for various physics list implementations; evaluate the total energy deposition and number of clusters for different double planes configurations and at different energies; determination of properties of machine learning algorithms and their main required inputs parameters; generate large amounts of simulated data for NeuLAnd using R3BRoot in order to train the machine learning algorithms with various scenarios of detector configurations, energies, physics lists or primary particles generators; check reconstruction methods used in TDR and the new methods based on clustering scoring which are implemented in R3BRoot; identify the method used to integrate neural network into R3BRoot infrastructure for event reconstruction in NeuLand; evaluate Keras, TensorFlow and SciKit-learn as reconstruction tools.

These results are detailed in Annual Summary Document for Stage II

Results obtained in Stage III - January - December 2022 develop a standalone Geant4 application to perform simulations for various scenario with NeuLand and Root analysis macro to evaluate the results; identify class managers of different stages in R3BRoot and their usage for simulations, digitization, clustering and reconstruction; evaluate several scenarios of simulated data using R3BRoot and Geant4 standalone application with various detector planes configuration, different physics list activated or different features of beam characteristics; evaluate neutron reaction probabilities as function of the number of double planes used for an energy range from 200 MeV to 1 GeV; identify the most common reactions and the most important reaction products for incoming neutrons with various energies using different physics list implementations; evaluate the total energy deposition and number of clusters for different double planes configurations and at different energies; generate large amounts of simulated data for NeuLand using R3BRoot in order to train the machine learning algorithms with various scenarios of detector configurations, energies, physics lists or primary particles generators; check reconstruction methods used in TDR and the new methods based on clustering scoring which are implemented in R3BRoot; identify the method used by machine learning algorithms for classification problems under supervised learning; evaluate the neutron multiplicity determination of Neural Network and Convolutional Neural Network methods; asses the performance of NN and CNN algorithms on neutron multiplicity; compare the performance of the NN and CNN methods with NeuLand’s TDR method and ‘best performance’ of Geant4 on neutron multiplicity and 4-neutron invariant mass spectra.

These results are detailed in Annual Summary Document for Stage III