Sorry, item "en_offcanvas-col1" does not exist.

Sorry, item "en_offcanvas-col2" does not exist.

Sorry, item "en_offcanvas-col3" does not exist.

Sorry, item "en_offcanvas-col4" does not exist.




This joint project is funded by the BMBF in the framework of the ErUM-Data action plan under the grant number 05D23XK1. The following public institutions and private companies are involved into this project:




The joint project RAPID is funded by the
of the Federal Ministry of Education and Research
under the funding code 05D23XK1.

Project duration: March 1st 2023 - February 28th 2026


There are two main problems with spectral data analysis:

  1. Duration of spectrum analysis
    Even if good measurements are available after a short time, spectra shows a complexity and a high dimensionality of the parameter space.
  2. Noise in the Spectrum
    Even high-quality measurement systems, which are sensitive to the measurement of very weak signals, exhibit noise in the spectrum, especially at short measurement times.

Both of these problems make it difficult to identify the elements in a sample. In addition, the large amounts of data and the variety of data represent a further difficulty in the analysis due to the wide energy range in the spectrum and the many measurements. In order to reduce the complexity of spectral data analysis, a generic machine learning approach will be developed and evaluated.


In this project, measured data will be analyzed using machine learning (ML) methods, a subset of artificial intelligence (AI), to optimize the evaluation of elementary characterization. During the process, the software/system learns the features of the reference data through training and then performs a variety of complex tasks.

From an industrial perspective, it is expected to use the spectral data for quick and reliable material characterization. Therefore, we use ML algorithms to develop a software that can

  • be adaptable to automated systems and enables efficient evaluation of spectral data with minimal manual intervention,
  • help in the interpretation of the spectra measured on a sample with an indication of the uncertainty,
  • be extended to other measurement systems and
  • provide the training data and training networks.

In this project, a software package for an efficient and scientific analysis of spectra is developed. This includes:

  • the integration of mathematical and physical knowledge (e.g. Physics-Informed Neural Networks),
  • the use of AI-based algorithms,
  • the calculation of realistic uncertainties,
  • the handling of noisy data during evaluation and
  • the determination of overall spectra as a combination of other spectra.

The measurement and evaluation method is collectively referred to as prompt gamma-neutron activation analysis (PGNAA) and differs from other approaches in which it can be carried out without sample preparation, resulting in an increase in mass throughput. In PGNAA, the samples to be analyzed are irradiated with neutrons and partially activated. As a result of the activation process, gamma radiation is emitted. Taking into account the measured energy-dependent gamma spectrum, the elements in the sample are determined and quantified.


The PGNAA is suitable for large-volume samples and is already used in the following areas:

  • Elemental analysis,
  • Oil exploration and
  • Quality assurance in the cement industry and coal mining.

The neutrons required for PGNAA are produced by a radionuclide source or a neutron generator. These two sources are suitable for industrial applications and allow the measurement of a sample in a short time. As a result, a high mass throughput can be achieved.

The AiNT develops and operates measurement systems, including PGNAA, which can be used to carry out elemental analysis of samples with a volume between 500 ml and 400 l. Furthermore, within the framework of the MetalClass project (BMBF, funding code FKZ 01IS20082A), AiNT has successfully developed and used ML methods for the classification of short-term measurements of scrap metal.


The PGNAA can be used for scientific and industrial applications. For example:

  • Phase analysis with X-ray and neutron diffraction,
  • Quasilastic neutron scattering and
  • Stellar spectroscopy, such as the study of exoplanets

This joint project is funded by the BMBF in the framework of the ErUM-Data action plan under the grant number 05D23XK1. The following public institutions and private companies are involved into this project:

  • Institute of Materials Physics, Helmholtz-Zentrum Hereon GmbH, Garching,
  • Aachen Institute for Nuclear Training (AiNT GmbH), Stolberg (Rhld.),
  • Research Neutron Source Heinz Maier-Leibnitz Zentrum, Technical University of Munich (TUM), Garching und
  • Institute of Computer Science in Mechanical Engineering, Helmut-Schmidt-Universität (HSU), Hamburg.

Associated partners are:

  • European Southern Observatory (ESO),
  • BASF SE,
  • European Spallation Source (ESS),
  • Forschungszentrum Jülich (Jülich),
  • University of Belgrade (new),
  • University of Novi Sad (new) and
  • FH Aachen (new).

The project participants as well as the thematic priorities are shown in the following figure:



For any questions please contact
Dr. Gözde Özden
+49 (0) 2402 10215-00
during our office hours

Monday to Friday
8:00 am - 4:00 pm

Go back