*Important notice: This news reports on an unedited version of an accepted paper and is awaiting final editing. Therefore, the paper should not be regarded as conclusive or treated as established information.
Laser-induced breakdown spectroscopy (LIBS) has been investigated as a rapid alternative for quantifying phosphorus pentoxide (P2O5) in solid phosphate rock. The researchers developed a calibration framework that integrated spectral preprocessing, statistical analysis, and machine-learning techniques to improve prediction accuracy across a broad range of phosphate concentrations. Their findings were published in Scientific Reports.
Study: Quantitative analysis of P2O5 in solid raw phosphate matrices using laser-induced breakdown spectroscopy (LIBS) for online mining industry monitoring. Image Credit: B Brown/Shutterstock.com
Why Phosphate Mining Needs Faster Ore Analysis
Phosphate rock is an important raw material for fertilizer production, supporting agricultural productivity worldwide. It also serves industries such as animal feed, food processing, and detergent manufacturing.
As high-grade deposits become depleted, mining companies must find ways to process lower-grade ores efficiently while maintaining consistent product quality. Flotation plays a major role in phosphate beneficiation, accounting for more than 80% of global phosphate production. Its efficiency depends on the composition of the incoming ore.
Phosphorus pentoxide (P2O5) concentration serves as the standard indicator of phosphate rock grade and helps operators assess feed quality and optimize processing conditions. However, conventional methods such as inductively coupled plasma optical emission spectrometry (ICP-OES) require extensive sample preparation and laboratory analysis, often taking several hours to deliver results.
These delays limit continuous monitoring and timely process adjustments in ore mining sites.
In this study, the researchers combined spectral preprocessing, statistical analysis, and machine-learning techniques to develop a calibration framework for phosphate grade prediction.
The study examines how different data-processing methods, wavelength-selection strategies, and regression models affect prediction accuracy, aiming to establish a foundation for rapid phosphate-grade monitoring in mining and mineral processing operations.
Developing a LIBS Method for Phosphate Analysis
The researchers developed a laboratory laser-induced breakdown spectroscopy (LIBS) system to analyze pelletized phosphate samples. A Q-switched neodymium-doped yttrium aluminum garnet (Nd:
YAG) laser generated 1064 nm pulses, each delivering 25 mJ of energy over five nanoseconds.
A high-resolution echelle spectrometer equipped with an intensified charge-coupled device detector recorded the plasma emission across wavelengths of 200–900 nm.
The calibration dataset comprised 84 samples prepared from certified phosphate rock (NIST Standard Reference Material SRM 694). The researchers diluted the material with high-purity silica to obtain P2O5 concentrations ranging from 6.4% to 24.8% by weight. Inductively coupled plasma–optical emission spectroscopy (ICP-OES) provided the reference concentrations.
The team compressed two grams of powdered sample into pellets at 300 MPa without binding agents. Twelve replicate measurements per pellet produced 1008 spectra. To improve data quality, the researchers applied principal component analysis (PCA), Savitzky–Golay smoothing, and standard normal variate (SNV) normalization. Three outlier-detection methods helped distinguish technical errors from natural mineralogical variations.
Finally, they evaluated 30 calibration configurations involving different regression algorithms, wavelength-selection methods, and replicate-handling strategies. The team compared partial least squares (PLS) regression with alternative machine-learning models. Independent training and test datasets, alongside 10-fold cross-validation, helped assess prediction accuracy and model reliability.
Spectral Processing Improves Phosphate Grade Predictions
The results show that spectral preprocessing plays an important role in improving LIBS accuracy. Raw spectra contain information about elemental composition, but plasma fluctuations, background radiation, and sample variability can interfere with analysis. Standard Scaler normalization performed less effectively because it increased the influence of noisy spectral channels.
Principal component analysis (PCA) showed that the first four components explained 69.9% of the total spectral variation, but their loading patterns contained fluctuations that did not clearly correspond to characteristic emission peaks. Standard normal variate (SNV) normalization produced clearer patterns.
After normalization, the first two components explained 62.75% of the variation, while the first five accounted for 89.65%. The loading patterns revealed specific features associated with phosphorus and calcium emissions, while sample scores showed a consistent progression as P2O5 concentrations increased. Savitzky–Golay smoothing further reduced noise while preserving important spectral features.
Download a free copy of this page here!
The researchers also combined three methods to distinguish technical errors caused by unstable plasma formation or sample surface conditions from natural variations in mineral composition. This approach identified eight confirmed technical outliers, which they removed from the original 1008 spectra.
The team then compared models using the full spectrum with those selecting specific wavelength regions associated with phosphorus and calcium. They also evaluated different approaches to handling replicate measurements. The best configuration combined partial least squares (PLS) regression, targeted wavelength selection, and averaging predictions after analysis.
The optimized model achieved a test-set coefficient of determination (R2) of 0.963 and a root-mean-square error (RMSE) of 0.966 percentage points of P2O5. Ten-fold cross-validation produced an R2 of 0.949 and an RMSE of approximately 1.16 percentage points.
LASSO and support vector regression also performed competitively, while random forest, XGBoost, and convolutional neural networks showed weaker generalization or greater overfitting. These findings highlight the importance of effective spectral preprocessing, careful wavelength selection, and appropriate model complexity for accurate phosphate grade prediction.
Toward Faster Phosphate Monitoring in Mining Operations
The study demonstrates that LIBS, combined with appropriate spectral processing and statistical modeling, can predict P2O5 concentrations in solid phosphate rock with high accuracy.
The approach could benefit mining and mineral processing by helping operators assess incoming ore, detect changes in feed composition, and adjust beneficiation conditions effectively. Integrating LIBS with conveyor-based or in-line monitoring systems could reduce dependence on time-consuming laboratory analysis and improve raw material consistency for fertilizer production.
However, the researchers tested the method using prepared pellets under controlled laboratory conditions: industrial applications introduce challenges such as variable moisture, uneven surfaces, changing mineral composition, and instrument contamination.
Further testing across different phosphate deposits and actual production streams will therefore be necessary to establish the method's reliability. Researchers must also assess long-term instrument stability and verify measurements against established laboratory techniques.
Overall, the study highlights the potential of combining laser spectroscopy with carefully designed calibration models to accelerate phosphate grade assessment. Successful industrial validation could enable more continuous ore monitoring, faster processing decisions, and improved resource utilization in phosphate mining and mineral processing.
Journal Reference
El Kohen, M., El Asri, M., et al. (2026). Quantitative analysis of P2O5 in solid raw phosphate matrices using laser-induced breakdown spectroscopy (LIBS) for online mining industry monitoring. Scientific Reports. DOI: 10.1038/S41598-026-72879-Z. https://www.nature.com/articles/s41598-026-72879-z.
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.