Data-driven optimization of pixelated CdZnTe spectrometers for uranium enrichment assay

TL;DR

This work targets uranium NDA with pixelated CZT detectors by integrating a data-driven spectroscopic optimization framework (spectre-ml) with the GEM-based uranium enrichment code (pyGEM). It uses voxel-level performance grouping via non-negative matrix factorization and clustering, with enrichment relative uncertainty as the optimization objective. Across six H3D M400 detectors, the approach yields a mean ~20% reduction in U-235 enrichment relative uncertainty for 30-minute measurements, translating into significant time savings without compromising accuracy. The modular API enables plugging in other end-user analyses and suggests extensions to Pu spectroscopy, strengthening practical safeguards capabilities.

Abstract

In recent work [Vavrek et al. (2025)], we developed the performance optimization framework spectre-ml for gamma spectrometers with variable performance across many readout channels. The framework uses non-negative matrix factorization (NMF) and clustering to learn groups of similarly-performing channels and sweep through various learned channel combinations to optimize the performance tradeoff of including worse-performing channels for better total efficiency. In this work, we integrate the pyGEM uranium enrichment assay code with our spectre-ml framework, and show that the U-235 enrichment relative uncertainty can be directly used as an optimization target. We find that this optimization reduces relative uncertainties after a 30-minute measurement by an average of 20%, as tested on six different H3D M400 CdZnTe spectrometers, which can significantly improve uranium non-destructive assay measurement times in nuclear safeguards contexts. Additionally, this work demonstrates that the spectre-ml optimization framework can accommodate arbitrary end-user spectroscopic analysis code and performance metrics, enabling future optimizations for complex Pu spectra.

Figures (6)

• Figure 1: Uranium gamma spectra as a function of depth in the PNNL M400 detector. Top: heatmap of all depth spectra. The white dotted lines indicate the depth bin selections used in the bottom panel. Bottom: bulk spectrum and spectra at three depth bin selections, each normalized to their intensity at $186$ keV (black dotted line) to better show their shapes. Depth bins near the anode degrade in resolution and can exhibit gain drifts, while depth bins near the cathode have increased backgrounds.
• Figure 2: Overview of the spectre-ml + pyGEM pipeline, adapted from Fig. 2 of Ref. vavrek2025data.
• Figure 3: Uranium enrichment relative uncertainty optimization example with the PNNL M400 detector. Top left: best cluster labels. Top right: best cluster mask. Center left: top $5$ spectra ranked by enrichment relative uncertainty. Center right: best spectrum from each class of clustering algorithm. Bottom left: pyGEM fit to the bulk spectrum. Bottom right: pyGEM fit to the best spectrum.
• Figure 4: Best cluster mask from each detector optimization.
• Figure 5: Trends in pyGEM analysis metrics vs dwell time $t$, using the optimum long-dwell-computed mask from the INL (top row), SNL (middle row), and LANL (bottom row) detectors. Left column: U-235 enrichment (wt%). The optimized (orange) points are slightly offset along the $x$-axis from their corresponding bulk (blue) points for visual clarity. The gray band shows the SRM 969 certification of $1.9420 \pm 0.0014$ wt% srm969. Right column: U-235 relative uncertainty (%).