The nuclear industry is a strategic high-tech industry and an important cornerstone of national security. It involves various areas, such as ore exploration and mining, uranium extraction, isotope separation, reactor power generation, and spent fuel reprocessing. The uranium content of uranium ores is an important criterion for identifying uranium ore types and evaluating their developmental value. The rapid collection of uranium distribution information is necessary for geographical exploration. In particular, this is true for China, where uranium deposits are scattered and ore bodies are relatively small. Laser-induced breakdown spectroscopy (LIBS) is an atomic emission spectroscopy technique that involves irradiating the sample surface with nanosecond pulse lasers (typically at irradiance levels above GW/cm2). The irradiated material on the sample surface is rapidly heated, melted, vaporized, and partially ionized, forming laser-induced plasma (LIP). The elemental composition of the sample material can be measured by analyzing the emission spectra of the plasma. Fiber-optic LIBS (FO-LIBS) is an LIBS system that utilizes optical fibers for laser transmission and simultaneous collection of plasma emission spectra. It uses flexible, long optical fibers to transmit pulse lasers and spectral signals, which make it more suitable for complex and confined spaces in the field than conventional LIBS. Measurement distances can reach tens of meters. This study addresses the demand for the rapid, in-situ, and on-site detection of uranium in the nuclear industry and establishes a laboratory-based FO-LIBS system for investigating the evolution characteristics of uranium emission spectral lines in plasma under a helium atmosphere. Furthermore, it provides parameter optimization schemes and explores the matrix effects of uranium ore samples. A multivariate calibration method for quantitative analysis is proposed, which effectively improves calibration and prediction accuracy while ensuring model generalization performance. This provides a new approach for the rapid elemental analysis of ores.

We conducted experiments using natural samples and their mixtures to better align the results with practical applications. Spectra of the pressed samples were acquired using the FO-LIBS system. An air-blowing device was used to create a helium atmosphere, and the spectral information in a helium atmosphere was compared with that in an air atmosphere. The detection delay was optimized by comparing the signal-to-noise ratio, signal-to-background ratio, and net spectral intensity of the spectral lines. A multivariate linear calibration algorithm based on an internal standard method was proposed to address the matrix effects caused by the compositional differences among the samples. The model was fitted using partial least squares regression (PLSR) and a constrained genetic algorithm (GA), and the results were compared with calibration results based on spectral net intensity.

Among the U I 356.659 nm, U II 367.007 nm, and U II 409.013 nm lines in the uranium ore, only the U II 409.013 nm line exhibits a higher signal-to-noise ratio and is unaffected by interference from other lines at low mass fraction (Fig.3). In a helium atmosphere, the signal-to-noise ratio of U II 409.013 nm increases by 1.37 times from 13.29 to 31.45. Additionally, the signal-to-noise ratio reaches 8.9 at a mass fraction of 0.0726%. During the study of the variation in the detection delay using FO-LIBS in a helium atmosphere (Fig.4), the signal-to-noise ratio of the characteristic spectral lines remains above 10 until a delay of 1000 ns; however, it rapidly decreases to approximately 5 after a delay of more than 1000 ns. The signal-to-background ratio exhibits a peak of approximately 2.4 at a delay of 1000 ns and continues to increase subsequently when the delay is over 1200 ns, primarily owing to the rapid decay of the background intensity in the later stage of the plasma compared to those of the spectral lines. Therefore, a detection delay of 1000 ns is selected as the optimal value. Finally, a comparison of the results of the univariate calibration, multivariate linear regression using PLSR, and multivariate linear regression using constrained GA (Fig.8) shows that the prediction results obtained using multivariate linear regression are closer to the reference values than those obtained using univariate calibration based on spectral intensity alone. This indicates that the multivariate regression approach can correct for the matrix effects. The R² (coefficient of determination) values of the calibration models based on PLSR and GA have both the training set and leave-one-out cross-validation (LOOCV) greater than 0.99, indicating the accuracy and robustness of these models. A comparison of PLSR and GA shows that the PLSR model exhibits superior calibration accuracy with a higher R² and lower root mean square error in LOOCV. By constraining parameter k to positive values using the GA, the calibration accuracy decreases slightly; however, the relative standard deviation (RSD) decreases, resulting in improved prediction stability. The limits of detection and quantification are estimated as 142 mg/kg and 426 mg/kg, respectively.

This study investigates a uranium detection method based on FO-LIBS to meet the demand for rapid, on-site, and in-situ uranium detection in the nuclear industry. Among the dense spectra containing multiple elements, the uranium spectral line U II 409.013 nm is selected. The enhancement effect of the helium atmosphere on the uranium spectral line is explored. For the sample with a uranium mass fraction of 0.425%, the helium atmosphere improves the signal-to-noise ratio of the spectral line by 1.37 times. In addition, the detection delay of the system is optimized, and a peak in the signal-to-background ratio is observed at 1000 ns, which is determined to be the optimal delay for quantitative analysis. Under optimal conditions, the signal-to-noise ratio of the uranium spectral line is 8.9 in a sample with a mass fraction of 0.0726%. A multivariate linear regression model based on the internal standard method is proposed to address the matrix effect caused by differences in the chemical compositions of the natural samples in the experiments. The spectral lines of the matrix elements are introduced for calibration. The fitting parameters are obtained using PLSR and a constrained GA, with PLSR exhibiting superior quantitative performance in terms of R² and RMSEC. The calibration model achieves an R² of 0.9984 for uranium and an RMSEC of 0.0404%. Furthermore, the limit of detection for uranium using FO-LIBS is estimated to be 142 mg/kg, and the limit of quantification is 426 mg/kg.