Leandro Barbosa da Silva

The limited availability of experimental gamma-ray spectrometry data remains a major obstacle to the development of reliable machine learning models for predicting radiological hazard indices in building materials. Although oversampling techniques have been widely adopted to alleviate data scarcity, their relative effectiveness for radiological prediction tasks has not been systematically established. This study addresses this gap by benchmarking three oversampling strategies, Random Oversampling (ROS), Synthetic Minority Oversampling with Gaussian Noise (SMOGN), and MixUp, using a dataset of 248 HPGe gamma-ray spectra acquired from 23 categories of Brazilian building materials. Each strategy was evaluated using a Dense Rectifier Neural Network (DRNN), Random Forest (RF), and Extreme Gradient Boosting (XGBoost). The DRNN significantly outperformed XGBoost for D𝛾 and I𝛾 (p < 0.05), while all three models exhibited comparable performance for IBra. Under LeaveOne-Group-Out (LOGO) validation, the DRNN maintained RMSE values below 0.04 for previously unseen materials categories, demonstrating strong generalization capability. SHAP analyses confirmed the physical consistency of the learned relationships, identifying 232Th as the dominant contributor across D𝛾, I𝛾, and IBra, followed by 40K and 226Ra. Monte Carlo uncertainty propagation revealed substantially wider confidence intervals for low-activity samples than for samples with higher activity concentrations. MixUp consistently outperformed SMOGN across all models and targets, reducing RMSE by up to 63% for I𝛾 with the DRNN. Despite its simplicity, ROS matched MixUp performance for IBra and outperformed both MixUp and SMOGN for D𝛾 and I𝛾 across all evaluated models, emerging as the most effective overall oversampling strategy. These findings demonstrate that the interaction between oversampling strategy and learning architecture is a key determinant of predictive performance under limited-data conditions, while providing a validated, physically interpretable, and uncertainty-aware framework for machine learning-assisted radiological assessment of building materials.

Resumo This study evaluates the extent to which input normalization influences the performance of a Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP) trained on gamma-spectrometric measurements of 226 Ra, 232 Th, and 40 K activity concentrations, together with six UNSCEAR-derived radiological hazard indices, for commercial building materials from Rio de Janeiro, Brazil. Six candidate preprocessing pipelines were initially defined: linear MinMax scaling, logarithmic transformation followed by MinMax scaling, empirical quantile mapping to approximately Gaussian marginals followed by MinMax scaling, median-IQR robust scaling followed by MinMax scaling, a composite logarithmic-standarization-MinMax pipeline, and a skewness-guided adaptive procedure. Prior to any training, algebraic analysis established that two of these pipelines reduce to simpler ones under the adopted implementation, leaving four geometrically distinct preprocessing strategies. Forty training runs (four conditions, ten paired random seeds each) supported a randomized complete block design and non-parametric inference across six evaluation axes: adversarial stability, marginal and multivariate fidelity, manifold quality, structural preservation, physical consistency, and memorization risk. The results suggest that normalization may affect multivariate similarity more consistently than marginal reconstruction. Nonlinear transformations were generally associated with lower Maximum Mean Discrepancy (MMD) relative to linear scaling, whereas no pairwise difference in marginal Earth Mover's Distance (EMD) survived multiplicity correction. Quantile normalization was associated with a lower adversarial-collapse rate but also with reduced manifold recall and coverage under the fixed training budget, a pattern compatible with a stability-coverage trade-off. Uniform logarithmic transformation substantially degraded the 40 K marginal, a variable with an approximately symmetric distribution, supporting the use of variable-specific preprocessing when radiometric features exhibit heterogeneous distributional profiles. Violations of the deterministic relationships defining derived radiological indices persisted across all conditions and all six indices, suggesting that normalization alone may be insufficient to ensure physical consistency, and that constraint-aware generative architectures may be required when radiological compliance is a primary objective. No evidence of training-data memorization was identified under the adopted diagnostic framework.

Beach sands may contain elevated concentrations of naturally occurring radionuclides because of sediment provenance, hydraulic sorting and local accumulation of heavy minerals. This study establishes an HPGe-based radiological baseline for beach sands from Rio de Janeiro State, Brazil, and evaluates external gamma exposure through an uncertainty-propagated screening approach. Forty-seven surface sand samples were collected from recreational beach sectors and analysed by high-purity germanium gamma-ray spectrometry. Activity concentrations of ²²⁶ Ra, ²²⁸ Ra, ²²⁸ Th and ⁴⁰ K were determined, and their reported analytical uncertainties were propagated to derived exposure quantities. To avoid double counting when applying a conventional whole-series thorium dose coefficient, a central ²³² Th-series screening equivalent was calculated as the arithmetic mean of the ²²⁸ Ra and ²²⁸ Th indicators. Sensitivity calculations using ²²⁸ Ra alone and ²²⁸ Th alone were also performed. Activity concentrations ranged from 1.10 to 1408.10 Bq kg ⁻¹ for ²²⁶ Ra, 0.03–5449.93 Bq kg ⁻¹ for ²²⁸ Ra, 0.42–6957.21 Bq kg ⁻¹ for ²²⁸ Th and 2.07–1008.36 Bq kg ⁻¹ for quantified ⁴⁰ K results. Canto was the dominant point-specific anomaly, with a central calculated absorbed dose-rate estimate of 4420 ± 169 nGy h ⁻¹ . Excluding Canto, the mean absorbed dose rate was 38.6 nGy h ⁻¹ , whereas the complete-dataset median was 16.8 nGy h ⁻¹ . Under all three thorium-series assumptions, the same eight sampling points exceeded the comparative outdoor value of 55 nGy h ⁻¹ and only Canto exceeded 1 mSv y ⁻¹ under the frequent local-use scenario. All short-term tourist-use estimates remained below 0.3 mSv y ⁻¹ . The 55 nGy h ⁻¹ value is used only for comparison and not as a regulatory or action level. Most sampling points therefore represent lower-priority external gamma exposure conditions, whereas a limited number of local anomalies require confirmatory sampling, in situ dose-rate mapping and mineralogical characterisation before any beach-wide radiological protection conclusion is drawn.

This study assessed natural radioactivity and radiological risk in 48 cladding and finishing materials marketed in Rio de Janeiro State, Brazil. Activity concentrations of 226Ra, 232Th and 40K were measured by high-resolution gamma-ray spectrometry using a 30% relative-efficiency HPGe detector, and radiological indices and annual effective dose were calculated. Mean activities were 23.2 ± 25.0, 31.2 ± 52.8 and 163.3 ± 309.0 Bq kg-1 for 226Ra, 232Th and 40K, respectively. Mean radium equivalent activity was 80.4 ± 112.3 Bq kg-1 and mean annual effective dose was 0.34 ± 0.46 mSv y-1. Non-parametric comparisons showed a clear activity gradient among functional groups (G1 > G2 > G3; p < 0.001), with mean doses of 0.679, 0.118 and 0.039 mSv y-1, respectively. Principal component analysis showed that one component explained 89.1% of the variance, while hierarchical clustering reproduced the functional classification for 42 of 48 samples (adjusted Rand index = 0.688), supporting a common underlying radioactivity gradient. Fifteen samples exceeded the conservative dose screening level of 0.48 mSv y-1, all in G1, and two mortars exceeded 1 mSv y-1. However, under the Radiation Protection 112 activity-index criterion for superficial materials, no sample exceeded the limit associated with 1 mSv y-1. Radiological concern is therefore concentrated in a limited subset of G1 products, particularly mortars and phyllite clays, supporting targeted raw-material selection and quality control.

Building materials are a primary source of indoor exposure to naturally occurring radioactive materials (NORMs), yet existing regulatory frameworks lack practical, cost-based tools to support the implementation of the ALARA principle in the construction sector. This study introduces the Ionizing Radiation Index (Γioniz), a novel approach that converts annual effective dose into a monetary value (US$), enabling transparent radiological risk classification of building materials. Activity concentrations of 226Ra, 232Th, and 40K were measured in 266 samples of 50 building material types, organized into eight groups (G1–G8) and collected from construction sites and stores in Rio de Janeiro, Brazil, between July 2020 and August 2024. Measurements were carried out using high-purity germanium (HPGe) gamma spectrometry after secular equilibrium between 238U, 232Th, and their decay products was confirmed. A log-normal distribution provided an adequate fit for all radionuclides (Kolmogorov–Smirnov test, p > 0.05 for all variables). The overall mean annual effective dose was 0.37 ± 0.02 mSv y-1. Group G1 — comprising crushed stone, sand, cement, and brick — recorded the highest Γioniz (US$ 90.39, classified as high radiological cost), followed by G3 (US$ 62.02), while G8 (paints and coatings) presented the lowest value (US$ 5.05). Materials are classified into three tiers: low risk (Γioniz ≤ US$ 30, E ≤ 0.3 mSv y-1), moderate risk (US$ 30 < Γioniz ≤ 60, 0.3 < E ≤ 0.6 mSv y-1), and high risk (Γioniz > US$ 60, E > 0.6 mSv y-1). The Γioniz framework offers regulators and industry stakeholders a financially grounded, ALARA-aligned instrument for regulatory impact analysis and informed material selection in residential construction.

Gamma-ray spectrometry is an essential analytical technique in environmental studies, providing valuable insights into radionuclides in building materials. This method enables researchers to assess potential radiation exposure and ensure compliance with safety regulations. By identifying specific isotopes, scientists can evaluate the radiological impact on public health and the environment. It uses the energy spectrum of emitted gamma rays to determine the type and quantity of isotopes present in samples. This study applies gamma-ray spectrometry to quality testing through case studies of commonly used building material samples in Brazil, using ISO and ABNT NBR standards as parameters. The main results discussed here demonstrate the potential for the development of gamma-ray spectrometry in the civil construction field. The identification of radioisotopes, the relationship between radionuclides and the standard composition of the samples, and the simplification of processes for obtaining results were fundamental to ensuring the quality and application of the technique. Although it has limitations, such as the need for large amounts of data, skilled labor, and proper sample preparation, the technique represents a significant improvement over current analytical methods, offering advantages in non-destructiveness and multi-elemental simultaneous detection compared to X-ray Fluorescence (XRF), Energy Dispersive Spectroscopy (EDS), and X-ray Diffraction (XRD), while also providing radiological characterization not accessible through those techniques, as well as in environmental sciences. In this context, the results showed that some building materials—such as sands, mortars, granites, enamel paints, and bricks—exhibit significant compositional differences from standard materials, as detected through gamma spectra and peak counting. These anomalies may compromise structural integrity and pose risks in Brazilian dwellings.

To assess radiation risks in building materials, this study presents an analysis framework that considers dose, exposure time, and frequency of material use, and develops a risk assessment and management model based on a management table and the temporal formula for material use in dwellings. Additionally, a cost-benefit analysis was employed to validate the risk assessment and management approach. First, the risk matrix classifies materials according to dose limits; then, usage time indicates the maximum exposure period during which a person could be exposed for up to 80% of their lifetime. The mean activity concentrations of 226Ra, 232Th, and 40K in the samples ranged from 1.3 ± 0.4 to 94.3 ± 2.5, 0.74 ± 0.4 to 91.7 ± 0.9, and 1.2 to 1023 ± 44 Bq kg-1, respectively. The effective dose ranged from 0.011 ± 0.001 to 1.094 ± 0.02 mSv y-1, with a mean value of 0.342 mSv y-1, which is above the recommended exclusion value of 0.3 mSv y-1 and below the dose limit of 1.0 mSv y-1 for most materials. However, regular monitoring of radioactivity levels in geological materials from sources is recommended before their use in construction. The results indicate that sand, cement, crushed stone, bricks, and some porcelain tiles present the highest risk under the established conditions. A cost-benefit analysis indicated that glass sand can replace river sand commonly used in dwellings, reducing radiation exposure risk and providing positive environmental impact by reusing materials and lessening the effects of river sand extraction. An appropriate agency should be established to consistently monitor radiation and prevent related health problems. The constructed model can offer new insights for radiation risk assessment in environmental analysis. These results are significant for the safety of materials in the Brazilian construction sector and may provide information to support radiological risk management.

This study aims to provide a thermophysical characterization of crushed stone, sand, brick, and cement due to radioactive decay. The 28 sand samples, 23 cement samples, 30 crushed stone samples, and 29 brick samples were analyzed by X-Ray Fluorescence (XRF) and Scanning Electron Microscope (SEM) coupled with Energy-Dispersive X-ray Spectroscopy (EDS) to screen samples for the presence of uranium, potassium, and thorium. Also high-purity germanium (HPGe) gamma spectrometers were applied to estimate 226Ra (as a proxy for the 238U decay series), 40K, and 232Th contents in ppm and to evaluate heat generation generated in the samples. For the samples, 226Ra ranges from 0.68 ± 0.04 to 21.7 ± 0.47 ppm, 40K ranges from 428 ± 92 to 37172 ± 1651 ppm, and 232Th ranges from 0.86 ± 0.11 to 171 ± 0.24 ppm. 21 sand samples, 19 cement samples, 29 crushed stone and stone dust samples, and 13 brick samples presented values ​​above the recommended average. The IH varied from 6.27×10-4 to 0.34, and the radiogenic heat production between the materials varied from 0.04 to 0.88 μW m-3, exceeding the average value ​​of 0.5 μW m-3 observed by Hansi Junior & Kusznir (2025). The temperature generated (Ttrmc) due to the radioactive decay of these materials in use in a standard room in a house was 7.54×10-5 °C. The value of the thermal power generated in the wall, the heat generation produced by the consumed mass, the linear power density, and the heat flux that passes through the wall surface from these materials were 3.65 μW, 1.98 nW kg-1, 0.46 µW m-1 and 0.23 µW m-2, respectively. The primary chemical components in the samples media are SiO2, Fe2O3, K2O and Al2O3. The crystal structures of samples typically consist of alternating Si-O tetrahedral layers and Al-O octahedral layers. The findings from this research will be useful to assess the radiation hazards of building materials in humans.