PGNAA Spectral Classification of Aluminium and Copper Alloys with Machine Learning

TL;DR

This work tackles real-time differentiation of copper and aluminium alloys using Prompt Gamma Neutron Activation Analysis (PGNAA) spectra. It evaluates data-generation (Categorical Sampling and CVAE), preprocessing (scaling, subsetting, DAE, EP/DEP, and Unique Peaks), and a broad range of classifiers, identifying Maximum Likelihood Classifier (MLC) with CVAE data as the strongest combination. The study reveals detector-dependent performance: CeBr$_{3}$ is advantageous for very short measurement times, while HPGe offers superior accuracy at longer times, guiding detector choice by application needs. The findings inform practical deployment in metal recycling, illustrating how detector selection and ML pipeline interact with measurement time budgets to enable fast, non-destructive alloy classification.

Abstract

In this paper, we explore the optimization of metal recycling with a focus on real-time differentiation between alloys of copper and aluminium. Spectral data, obtained through Prompt Gamma Neutron Activation Analysis (PGNAA), is utilized for classification. The study compares data from two detectors, cerium bromide (CeBr$_{3}$) and high purity germanium (HPGe), considering their energy resolution and sensitivity. We test various data generation, preprocessing, and classification methods, with Maximum Likelihood Classifier (MLC) and Conditional Variational Autoencoder (CVAE) yielding the best results. The study also highlights the impact of different detector types on classification accuracy, with CeBr$_{3}$ excelling in short measurement times and HPGe performing better in longer durations. The findings suggest the importance of selecting the appropriate detector and methodology based on specific application requirements.

Figures (5)

• Figure 1: Example prompt gamma spectra of the gamma quants of an aluminium alloy measured with a HPGe detector as basis for classification. The long-term measurement of 1h in (\ref{['fig:plot_a']}) can easily be classified into a specific aluminium alloy by considering the clearly recognizable characteristic peaks. The short-term measurement of 1s in (\ref{['fig:plot_b']}) is very noisy and statistical methods cannot recognize the characteristic peaks. Classifying alloys by such spectra is the goal of this paper.
• Figure 2: Accuracy of the best models on aluminium measurements (Table \ref{['tab:all_methods']}).
• Figure 3: Accuracy of the best models on copper measurements (Table \ref{['tab:all_methods']}).
• Figure 4: Comparison of the accuracy over time of the two detectors with aluminium chips data using CVAE and MLC.
• Figure 5: Section of an example spectrum of an aluminium alloy generated with different methods.