Application of artificial neural networks to gamma spectrometric measurements

Artificial neural networks (ANN), which is one of the most common subjects of recent years, has features such as learning ability, fast work, ease of identification, ability to work with less information, modeling nonlinear systems, processing information in parallel and adapting to new situations. In the field of nuclear spectroscopy, for example, gamma spectroscopy, alpha spectroscopy, peak-background ratio, energy resolution improvement of pulse type detectors, proton-induced X-ray emissions, X-ray fluorescence and Rutherford backscatter, and the like subjects. Considering the advantages of ANN and its use in nuclear spectroscopy, in this thesis, ANN was firstly applied for gamma spectral analysis using the HPGe detector found in the census laboratory of the Institute of Nuclear Sciences. The performance of the ANN model was tested by gamma spectral analysis of Çandarlı coastal sediment samples. With ANN, the presence of gamma emitting radionuclides in the samples has been successfully determined. In the second step of this study, ANN was used to find the minimum detectable activity (MDA) that the NaI (T1) Scintillation Gamma Spectrometer system could measure at different counting times. For this purpose, minimum detectable activity (MDA) values for 226Ra, 232Th and 40K radionuclides between 5000- 87000s were calculated based on Curie criteria, taking into account the counting times used in the Institute. When the MDA values obtained from ANN were compared with the calculated MDA values, it was seen that they overlap with each other. In this study, for the first time, ANN was applied for natural radionuclide analysis by NaI (T1) scintillation gamma spectroscopy. For this purpose, 126 different environmental materials; Radium, thorium and potassium concentrations in rock, soil, granite, marble and fertilizer samples were calculated. When the calculated activity values were compared with the values estimated from the ANN model, it was found that the ANN could accurately predict 226Ra, 232Th, and 40K activities in the samples.