Objective Residual uranium (U) present in uranium-bearing slag is a byproduct of nuclear fuel processing and nuclear waste management. Its rapid and accurate quantification is critically important in several domains, including nuclear resource assessment, environmental monitoring of contaminated sites, and the safe disposal of radioactive waste. Residual U not only reduces the efficiency of nuclear material utilization but also poses significant radiological risks if released into the environment through leaching or improper storage. Therefore, the development of analytical techniques that ensure both high accuracy and operational safety is of paramount relevance to the nuclear industry. Conventional laboratory-based methods, such as inductively coupled plasma mass spectrometry (ICP-MS) and X-ray fluorescence (XRF) spectroscopy, are widely regarded for their high sensitivity and precision. However, their application is often limited, as extensive sample preparation is required, expensive instrumentation must be employed, and real-time or remote detection under hazardous conditions is not supported. To overcome these limitations, laser-induced breakdown spectroscopy (LIBS), particularly in its remote configuration (R-LIBS), has emerged as a promising alternative. As a rapid, nearly non-destructive technique, LIBS enables simultaneous multi-element detection without sample preparation. Its capability for non-contact, variable-distance measurements makes it well-suited for radioactive environments where safety and efficiency are essential. Nevertheless, the complex and heterogeneous composition of slag introduces challenges such as plasma–matrix effects, background fluctuations, and spectral interference, all of which can compromise analytical accuracy. Accordingly, a method integrating remote LIBS (R-LIBS) with partial least squares (PLS) regression is proposed to enable quantitative determination of U in slag at varying detection distances. This approach aims to enhance the analytical reliability and expand the applicability of LIBS for on-site, remote analysis of nuclear waste materials under complex environmental conditions.

Methods To evaluate the feasibility and performance of the proposed approach, twelve slag samples with varying U concentrations were prepared (Tab.1). R-LIBS experiments were conducted under two distinct detection distances, 2 m and 7.2 m, representing short- and long-range measurement conditions. The laser system, spectrometer, and optical setup were configured to ensure stable plasma generation and efficient light collection (Fig.1). Four different types of spectral input were investigated to assess the impact of preprocessing on PLS modeling: raw spectra (Raw), whole normalization (Whole), Savitzky–Golay smoothing (SG), and standard normal variate (SNV) transformation. Each spectral dataset was used to construct a PLS calibration model, and the models were evaluated in terms of coefficient of determination (R²), root mean square error (RMSE), and relative error (RE) of prediction. This systematic design allowed for a direct comparison of preprocessing strategies under different detection conditions, providing insights into their role in improving prediction robustness and accuracy.

Results and Discussions At the 2 m standoff distance, the PLS model based on raw spectra exhibited the best stability and predictive accuracy, with an R² of 0.998 and a root mean square error of prediction (RMSEP) of 67.64 µg/g (Fig.7). The corresponding RE for two validation samples were 0.2% and 3.2% (Fig.8). At a longer standoff distance of 7.2 m, an improved prediction performance was achieved using the Whole-preprocessed spectra, yielding an R² of 0.997, an RMSEP of 90.29 µg/g (Tab.2), and RE values of 9.0% and 10.2% (Fig.8). Further enhancement in accuracy was obtained with SG and SNV preprocessing, reducing RE to 2.5% and 6.0%, respectively (Fig.8). A comprehensive comparison across all models confirms that R-LIBS combined with PLS regression provides robust quantitative performance under both short- and long-range conditions. Notably, the adaptability of spectral preprocessing enables the analytical workflow to be tailored to specific measurement environments. These findings further demonstrate that, even in the presence of complex slag matrices and signal degradation at extended distances, appropriate preprocessing combined with chemometric modeling can effectively recover prediction accuracy to levels suitable for practical field deployment.

Conclusions This study demonstrates the feasibility of integrating R-LIBS with PLS regression for the remote quantitative analysis of U in complex slag. An internal standardization approach based on the U II 385.95 nm emission line was initially evaluated but exhibited limited effectiveness in compensating for signal attenuation and matrix interference across varying standoff distances. By introducing preprocessing strategies such as Whole, SG, and SNV, the predictive performance of the PLS models was significantly improved, particularly under long-distance conditions where raw spectra became less reliable. The comparative results confirm that raw spectra are sufficient for accurate modeling at short distances, while tailored preprocessing is indispensable at longer distances to maintain robustness and accuracy. Overall, the proposed method proves to be a reliable and efficient solution for on-site detection of U in slag under remote and complex environments, with strong potential applications in nuclear waste management, resource assessment, and environmental monitoring.