Abstract:Machine-learning performance in laser-induced breakdown spectroscopy (LIBS) is often summarized by a single accuracy, F1 score, R2 value, or root-mean-square error (RMSE), although the same metric can support very different claims depending on how the evaluation data were constructed and accessed. This work proposes a LIBS-specific evaluation-claim reporting framework that records physical independence, instrument/session separation, and evaluation access/custody separately from a factor-specific Domain-Shift Matrix (DSM) describing composition and concentration, physical state/preparation, ambient environment, excitation/geometry, detection/spectral response, and signal quality. Three public-data case studies illustrate the framework. A retrospective EMSLIBS 2019 reanalysis with public-label access examines whether previously reported internal-to-benchmark discrepancies recur under a different fixed protocol; its sample-disjoint variants also alter training composition. In NASA SuperLIBS quantification (997,625 spectra; 2,511 materials), random-shot validation underestimated matched-energy error on unseen in-range and compositionally out-of-distribution (OOD) targets by 0.185 and 0.735 standardized RMSE, whereas target- and split-group-disjoint validation reduced the corresponding absolute discrepancies to approximately 0.028 and 0.060. In a separate controlled lithology-classification study, training rows, labels, weights, preprocessing, and candidate trajectories were held fixed while only the validation view changed; population- and energy-aligned group-disjoint validation reduced median absolute validation-to-test gaps in all eight evaluated situations, but better agreement did not consistently improve locked-test accuracy. These empirical analyses are illustrative case studies rather than inferential estimates of the causal effect of validation design. The framework makes explicit what a reported score estimates and the conditions under which that interpretation is supported.

Abstract:Shot-to-shot fluctuations in laser-induced plasma emission limit the precision of quantitative laser-induced breakdown spectroscopy (LIBS). This study evaluates an ultrasonic-assisted normalization approach in which laser-ablation-induced body waves, measured with a contact piezoelectric transducer, are used as an external indicator of shot-dependent ablation variability. For each analyte, the first-arrival acoustic feature was related to the within-specimen relative deviation of the corresponding emission line using a single model fitted from pooled calibration data; the fitted coefficients were then applied unchanged to the prediction specimens. For Mn, Cr, and Cu, normalization reduced the shot-to-shot relative standard deviation (RSD) from 6.87%, 4.27%, and 4.59% to 2.30%, 2.25%, and 2.87%, respectively. The corresponding average relative errors decreased from 1.41% to 1.13% for Mn, from 2.27% to 2.02% for Cr, and from 1.04% to 0.93% for Cu. For comparison with spectral internal standardization, Fe-dominant reference features were selected from the averaged spectra using spectral non-overlap, monotonic calibration behavior, calibration-set R2 and RMSEC, and calibration-sample RSD; prediction metrics were not used during feature selection. The selected reference features were 391.843, 427.966, and 458.218 nm for Mn, Cr, and Cu, respectively. Both the optimized Fe-ratio and ultrasonic-normalized models improved quantitative performance relative to the raw-intensity models, although the extent of improvement depended on the analyte and evaluation metric. Additional measurements showed that the body-wave response is sensitive to mechanical coupling and specimen thickness, while a synchronous comparison with an airborne-acoustic channel showed distinct temporal and statistical characteristics. Accordingly, the contact body-wave approach can serve as a configuration-dependent external reference when stable mechanical coupling can be maintained.

Abstract:Molybdenum (Mo) is used as a first-wall material in EAST, where laser-induced breakdown spectroscopy (LIBS) is a promising tool for in-situ diagnostics of plasma-wall interactions (PWIs). The large variation in ambient pressure during EAST operation can significantly affect the expansion of laser-ablated plasmas and, consequently, the stability of LIBS signals. In this study, plasma emission spectra were recorded in coaxial and orthogonal detection geometries using a high-resolution Echelle spectrometer, and the differential Doppler shifts between the two geometries were used to investigate the pressure-dependent normal expansion velocities of Mo atoms and ions over an ambient pressure range of 4×10-5-6×102 mbar. At low pressures (4×10-5-3×10-2 mbar), Mo atoms and ions maintained relatively high expansion velocities with weak pressure dependence, reaching approximately 4-6 km/s and 11-15 km/s, respectively. Local maxima in the velocities of both species were observed near 10-1 mbar. At pressures above 1 mbar, the velocities decreased rapidly owing to enhanced collisions and confinement by the background gas. The pressure-dependent trends obtained using the Doppler-shift method agreed well with those determined by time- and space-resolved spectroscopy. These results demonstrate the applicability of Doppler-shift spectroscopy for non-contact velocity diagnostics of laser-ablated Mo plasma and provide experimental guidance for interpreting and optimizing LIBS measurements under EAST-relevant pressure conditions.

Abstract:Spatial confinement is a simple, efficient, and low-cost modulation strategy for enhancing both the signal intensity and shot-to-shot stability of laser-induced plasmas. However, achieving optimal confinement still relies on repeated experiments because existing shockwave models do not account for cavity configurations. In this work, an inclined plate confinement cavity is developed for the spatial modulation of laser-induced plasmas. The confinement effects of cavities with two typical configurations on plumes are compared and analyzed. The spectral intensities are enhanced and the shot-to-shot spectral fluctuation is reduced by using an inclined plate with an obtuse-angle configuration. The mechanism behind the disparity is demonstrated by shadowgraphs of shockwaves. A theoretical formula for the shockwave is developed by incorporating the cavity characteristic parameters into the model. The developed formula enables accurate prediction of the delay time for the reflected shockwave to propagate back to the plasma generation region in inclined plates. The predicted temporal window agrees well with time-resolved spectroscopic measurements, with a mean absolute error of 0.56 μs. Quantitative analysis of the trace element Mg is also conducted using spatially confined LIBS with inclined plates. The limit of detection of Mg is reduced to one-seventh of that with no cavity by using the inclined plate with an obtuse-angle configuration. This work provides a feasible route for controllable plasma modulation and the synergistic enhancement of plasma signal intensity and stability, facilitating precise quantitative analysis by LIBS.

Abstract:Conventional techniques such as X-ray diffraction and scanning electron microscopy coupled with energy-dispersive X-ray spectroscopy are commonly employed for solid oxide fuel cell (SOFC) anode characterisation to provide information on phase composition and local elemental distribution. However, these methods generally involve limited measurement coverage, relatively high analytical costs, and complex operating procedures, restricting their application to the rapid evaluation of large-area SOFC anode samples. In this work, laser-induced breakdown spectroscopy (LIBS), hyperspectral imaging (HSI), and smartphone video imaging (SVI), combined with machine learning, were applied to rapidly quantify the nominal NiO mass fraction in sintered NiO/YSZ and corresponding reduced Ni/YSZ samples. Specifically, LIBS acquired atomic emission spectra that reflected elemental information of Ni and Zr, whereas HSI and SVI recorded reflectance signals and RGB colour responses, respectively. Four regression algorithms, including partial least squares regression, locally weighted partial least squares regression, kernel extreme learning machine, and support vector regression, were employed to establish relationships between the acquired signals and nominal NiO mass fraction. LIBS exhibited the best single-modality performance, with maximum R2 values of 0.9854 and 0.9860 for the sintered and reduced samples, respectively. Further, decision-level LIBS-SVI and HSI-SVI fusion were investigated. Adding SVI improved predictive performance in most model-modality combinations, with lower RMSE values obtained in 14 of the 16 evaluated fusion cases compared with the corresponding LIBS- or HSI-based single-modality models. Overall, SVI provided useful complementary information while introducing little additional equipment cost or measurement complexity.

Abstract:High spectrum-level classification accuracy in laser-induced breakdown spectroscopy (LIBS) may reflect technical-replicate dependence and acquisition-related confounding rather than generalizable material identification. This study therefore evaluates the within-specimen repeatability and internal spectral separability of nine nominal metallic material classes using a validation-aware analytical workflow. The dataset comprised 181 single-shot spectra obtained from nine physical specimens, with one specimen representing each nominal class. Repeatability descriptors, principal component analysis, Ward hierarchical clustering, and random forest classification were combined with label-permutation testing, acquisition-metadata analysis, and leave-one-acquisition-signature-out assessment. Mean within-class spectral correlations ranged from 0.745 to 0.971, and the first two principal components explained 63.55% of the total variance. random forest out-of-bag resampling completely separated the technical-replicate spectra, whereas weighted accuracy decreased to 93.37% when complete acquisition signatures were excluded. Acquisition metadata predicted material class above chance, and the association between material class and acquisition signature was statistically significant. Label permutation produced a mean null accuracy of 11.18% and an empirical p-value of 0.00498, confirming non-random internal spectral structure. The results demonstrate internal separability among the measured specimens but do not establish identification performance for independently acquired materials. The principal contribution is a transparent validation-aware framework that distinguishes within-dataset separation from external material identification and identifies independent multi-specimen testing under uniform acquisition conditions as the necessary next step.

Abstract:Laser-induced breakdown spectroscopy (LIBS) simulation requires accurate Stark-width parameters, yet reliable data remain missing for many spectral lines. Current simulators rely mainly on manually maintained tables, simplified analytical formulas, or the STARK-B database — but manual curation is labor-intensive, simplified formulas carry large errors, and STARK-B covers a limited set of lines on a discrete, pre-computed temperature grid. To bridge these gaps, an automated ion-line Stark-width workflow was developed based on the modified semi-empirical (MSE) method on top of the NIST Atomic Spectra Database, with the required oscillator strengths completed through a tiered hierarchy of sources. The MSE formalism is defined for ionized emitters, and the workflow inherits that domain. Each computed width is returned with an uncertainty estimate and a full data-provenance trace, and widths resting on weak inputs are automatically flagged as high-risk. The workflow is accessible through both a web interface and a Python SDK, making Stark widths available at the scale that spectrum-level LIBS applications require. The workflow is validated against two independent references: the semiclassical STARK-B database and the critically evaluated measured widths of Konjevi? et al. (2002). The MSE formalism itself, applied with hand-curated inputs, carries a stated accuracy of about 50 %; the automated workflow trades part of this precision for coverage, since oscillator strengths are completed from external databases and statistical estimates rather than curated line by line, and sparse level data leave some widths to the approximate lumped term. Within the LIBS-relevant scope of ions II–IV in the optical window, the median ratios of computed to reference widths (W_MSE/W_ref) are 1.15 against STARK-B (802 per-condition comparisons) and 1.05 against the evaluated measurements (1,159 per-condition comparisons). Of these comparisons, 62 % and 58 % respectively fall within a factor of 1.5, and 83 % and 76 % within a factor of two. This provides the basis for obtaining Stark widths at the scale LIBS applications demand, and the identified error sources define clear routes for refining the accuracy. A dedicated treatment of the neutral emitters, the prerequisite for serving a full-spectrum LIBS simulator, is the largest single item of future work.

Abstract:High-repetition-rate fiber-laser-induced breakdown spectroscopy (FL-LIBS) supports rapid surface scanning, yet reliable alloy quantification depends on spectra that are stable and representative of the sampled surface. Here, a spectral acquisition-window optimization strategy was developed for quantifying Cr, Mn, and Ni in Z3CN20-09M steel. Each acquisition window was characterized by pulse accumulation number and acquisition-window length, and CRITIC-TOPSIS integrated multiple spectral-quality metrics to rank the scanning conditions. Three representative acquisition windows were compared using full-spectrum partial least squares regression (PLSR) models calibrated with 11 high-alloy steel certified reference materials. The 10 ms, 4 mm s-1 condition gave the best overall performance, with a 40 μm acquisition-window length that approximately matched the laser spot diameter and a moderate pulse accumulation number of 350. Under this condition, full-spectrum PLSR yielded RMSEPs of 1.170, 0.063, and 1.975 wt.% for Cr, Mn, and Ni, respectively, lower than those under the matched comparison conditions. Subsequent model optimization using mutual information-variable importance in projection PLSR (MI-VIP-PLSR) further reduced the RMSEPs to 0.552, 0.047, and 0.127 wt.%. MI-VIP retained compact subsets of chemically informative variables, improving model interpretability while preserving high prediction accuracy. These results link pulse accumulation and spatial sampling to quantitative accuracy and provide a physically interpretable strategy for rapid, reliable FL-LIBS alloy quantification.

Abstract:Specialty steel is an essential foundational material for equipment manufacturing and large-scale engineering projects. For efficient, rapid and accurate classification and identification of specialty steel, this work proposes a cascade laser-induced breakdown spectroscopy (LIBS) classification model combining the iterative wavelet transform (IWT) and deep Boltzmann machine–backpropagation neural network (DBM-BP). This study investigates seven types of special steel. Initially, the spectral data were subjected to background removal using IWT. Subsequently, principal component analysis (PCA) and DBM were respectively employed to reduce the dimensionality of the preprocessed data. Finally, two hybrid classification models, namely, PCA-BP and DBM-BP were constructed. The experimental results revealed that the signal-to-background ratio (SBR) of the Mn element increased by approximately one order of magnitude after IWT background removal. Furthermore, the DBM-BP model achieved a classification accuracy of 98.53 % on the training set and 96.83 % on the test set for the specialty steel samples, validating the superiority of the DBM-BP model in nonlinear feature extraction. This study demonstrates that the IWT-DBM-BP combined model effectively suppresses the continuous background while preserving critical spectral features, thereby enabling high-precision classification. within the seven types of specialty steel samples investigated. This study offers reliable technical support for practical LIBS applications in rapid steel identification and intelligent scrap metal sorting.

Abstract:Three-dimensional electron diffraction (3DED) has evolved into a mainstream method for nanocrystal structure determination, with widespread applications in structural biology, pharmaceutical research, functional materials, and Earth and planetary sciences. Here we report GIGED-1pro (Guangzhou Institute of Geochemistry Electron Diffractometer, first generation, professional version), a self-developed dedicated instrument for rapid nanocrystal structure determination. The instrument comprises a 120 kV Schottky thermal field emission gun as the electron source, a vertical column geometry, a 4-axis goniometer stage (drift < 200 nm over a ±50° tilt range), and a Dectris Quadro hybrid pixel direct electron detector. It integrates TEM imaging, continuous rotation electron diffraction (cRED) acquisition, and an automated data processing pipeline. Single-crystal structure determination of natural fluorapatite and calcite confirms that the instrument is capable of de novo structure solution of submicrometer mineral crystals. Batch phase screening was validated using Chang’e-6 lunar regolith as a complex unknown system. Unsupervised hierarchical clustering of 43 valid diffraction datasets successfully identified six mineral phases: cristobalite, clinopyroxene, olivine, andesine–labradorite, anorthite, and ilmenite. No priori phase assumption was required, and only trace powder was dispersed on a TEM grid. These results demonstrate that GIGED-1pro, through its dedicated and integrated design, achieves full-workflow coverage from nanocrystal structure determination to unbiased batch phase screening for mineralogical samples, demonstrating potential as an effective tool for low-consumption structural characterization of precious extraterrestrial returned samples and deep-Earth nanominerals. Its core design as a general-purpose nanocrystal electron diffraction platform may potentially be applicable to structural biology, pharmaceutical development, and functional materials research, though experimental validation in these additional fields remains for future work.