Gamma Spectrum Separation for Whole-Body Counter ANN Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing whole-body counters face challenges in accurately separating gamma spectra of multiple nuclides due to overlapping peaks, leading to increased RMSE values and reduced accuracy in radioactive contamination measurements, especially when interfering nuclides are present.
Innovation Solution
A gamma spectrum separation system for a whole-body counter that includes an efficiency calculation unit, a count rate contribution amount calculation unit, and a normalization unit to separate a single-nuclide spectrum from a mixed spectrum, using artificial neural networks for training data generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a NaI (Tl) detector is used for whole-body counting, then measurement time is reduced to 1-3 minutes with excellent gamma-ray detection efficiency, but the bad resolution causes the detector to fail to satisfy the RMSE criterion when nuclides are affected by interfering nuclides
Solution Approach 1:
The patent introduces an artificial neural network as an intermediary computational system that processes the gamma-ray spectrum data from the NaI (Tl) detector. The ANN learns to separate overlapping spectra of multiple nuclides by training on synthetic data generated through Monte Carlo simulations, enabling the system to achieve accurate measurement results despite the detector's inherent poor resolution.
Solution Approach 2:
The patent creates synthetic copies of gamma-ray spectra through Monte Carlo simulations of various nuclide combinations. These synthetic spectra are used to train the artificial neural network, allowing the system to learn separation techniques without requiring extensive actual measurement data. The synthetic training data replicates the statistical properties and overlapping patterns of real spectra.
2Measurement precision
If an artificial neural network is used to analyze gamma-ray spectra, then spectrum separation capability is improved, but it is difficult to obtain sufficient training spectra through actual measurements
Solution Approach 1:
The patent generates synthetic copies of gamma-ray spectra through Monte Carlo simulations that model the detection process, including energy deposition, scattering, and coincidence events. These simulated spectra serve as training data for the artificial neural network, providing sufficient quantity and variety of training examples without requiring extensive actual measurements.
Solution Approach 2:
The Monte Carlo simulation varies multiple parameters including nuclide types, activities, geometric configurations, and detection conditions to generate diverse synthetic spectra. This parameter variation creates a comprehensive training dataset that covers the full range of possible measurement scenarios, enabling the ANN to generalize well to real-world data.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enables accurate separation of single-nuclide spectra from mixed spectra, allowing for the generation of synthetic spectra for artificial intelligence training, thereby improving the accuracy of radioactive contamination measurements.
Implementation Method 1
A whole-body counter is an instrument for measuring the level of radioactive contamination by detecting gamma rays emitted from radionuclides
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Proposed is a gamma spectrum separation system for a whole-body counter for generating artificial neural network training data by separating a single-nuclide spectrum from a spectrum in which multiple nuclides are mixed. According to the present disclosure, the present disclosure includes an efficiency calculation unit configured to obtain an efficiency for each of channels A, B, and C of a whole-body counter, a count rate contribution amount calculation unit configured to calculate a count rate contribution amount of each channel by multiplying the efficiency of each channel by a value obtained by multiplying a mixed source for generating the artificial neural network training data by a gamma-ray emission rate of each nuclide, and a normalization unit configured to subtract the count rate contribution amount from the count rate of each channel measured by the nuclide and then perform normalization to obtain a normalized spectrum of the nuclide. Accordingly, a single-nuclide spectrum can be separated from a spectrum in which multiple nuclides are mixed, and the single-nuclide spectrum can be used to generate a synthetic spectrum for artificial intelligence training.