Logo of the Physikalisch-Technische Bundesanstalt

Digital "mixing" of gamma-spectra

23.12.2020

As part of the digitalization of the metrology sector, efforts are underway to apply artificial intelligence methods in the field of spectrometry. Convolutional neural networks (CNNs), for example, are being tested for use in nuclide identification and, in combination with Monte Carlo simulation, in the deconvolution of spectra.

Within this context, a technique has been developed to enable a more direct characterization of mixed radionuclide solutions. Since the primary methods in general do not allow nuclide‑specific activity determination, the activity of mixed solution standards is determined exclusively by means of γ‑spectrometry. In order to obtain the small uncertainties associated with primary methods for mixed solutions as well, the applied technique must be as direct as possible and contribute no further uncertainty.

To this end, a new Bayesian procedure has been developed at PTB in which the spectra of the parent solution and of the mixed solution being determined are measured using precisely identical geometries.

Activity transfer factors are determined through the simultaneous rebinning of the spectra as well as multivariate Gaussian optimization (minimization) of the deviation of the spectra. In future, the uncertainty distributions of these activities are to be determined using Markov chain Monte Carlo methods.

Figure 1 shows the histogram of the γ‑spectrum of the mixed solution (black dots) as a function of the energy of the γ‑quanta. The individual nuclide spectra (Cd‑109, Eu‑154 (Eu‑155), Co‑60, Cs‑134) that make up the mixed solution are shown here as pale histograms. The histogram of the optimized, digital combination of these spectra is shown for comparison in bright red.

Fig.1: Histogram of the γ‑spectrum of the mixed solution (black dots) as a function of the energy of the γ‑quanta. The individual nuclide spectra (Cd‑109, Eu‑154 (Eu‑155), Co‑60, Cs‑134) that make up the mixed solution are shown here as pale histograms. The histogram of the optimized, digital combination of these spectra is shown for comparison in bright red.

Figures 2 and 3 present detail views from this spectrum for the particularly interesting energy range around the main line for Cd‑109 at Eγ = 88.03 keV.

Fig. 2: Detail view from this spectrum for the particularly interesting energy range around the main line for Cd‑109 at Eγ = 88.03 keV.

Fig. 3: Detail view from this spectrum for the particularly interesting energy range around the main line for Cd‑109 at Eγ = 88.03 keV.

We can see that both the non‑constant background and the Eu‑155 line with its very low count rate (Eγ = 86.55 keV, pγ 0.307) result in a perfect fit, even though Eu‑155 is here merely an impurity of the Eu‑154 single nuclide solution.

Contacts:

Opens local program for sending emailF. Mertes, Department 6.1, Working Group 6.13

Opens local program for sending emailA. Honig, Department 6.1, Working Group 6.13

Opens local program for sending emailS. Röttger, Department 6.1, Working Group 6.13