Abstract
A multivariate dominant factor based non-linearized PLS model is proposed. The intensities of different lines were taken to construct a multivariate dominant factor model, which describes the dominant concentration information of the measured species. In constructing such a multivariate model, non-linear transformation of multi characteristic line intensities according to the physical mechanisms of lased induced plasma spectrum were made, combined with linear-correlation-based PLS method, to model the nonlinear self-absorption and inter-element interference effects. This enables the linear PLS method to describe non-linear relationship more accurately and provides the statistics-based PLS method with physical backgrounds. Moreover, a secondary PLS is applied utilizing the whole spectra information to further correct the model results. Experiments were conducted using standard brass samples. Taylor expansion was applied to make the nonlinear transformation to describe the self-absorption effect of Cu. Then, line intensities of another two elements, Pb and Zn, were taken into account for inter-element interference. The proposed method shows a significant improvement when compared with conventional PLS model. Results also show that, even compared with the already-improved baseline dominant-factor-based PLS model, the present PLS model based on the multivariate dominant factor yields the same calibration quality (R2=0.999) while decreasing the RMSEP from 2.33% to 1.97%. The overall RMSE was also improved to 1.05% from 1.27%.
Introduction
Laser-induced Breakdown Spectroscopy (LIBS) has emerged as a versatile tool for rapid chemical analysis across various industries. However, its quantitative accuracy is often hampered by the complex physical phenomena inherent in laser-generated plasmas. Specifically, self-absorption and inter-element interference (matrix effects) create non-linear relationships between spectral line intensities and elemental concentrations. While statistical methods like Partial Least Squares (PLS) are commonly employed to handle high-dimensional spectral data, their linear nature often fails to capture these intricate physical distortions, leading to significant errors in measurement.
Methodology: Integrating Physics with Statistics
The researchers propose a novel hybrid modeling approach: a Non-linearized PLS Model based on a Multivariate Dominant Factor. This method bridges the gap between physical theory and statistical regression through three primary innovations.
1. Physically-Grounded Non-linear Transformation
Instead of treating the spectral data as a "black box," the model applies a non-linear transformation to the intensities of characteristic lines. By utilizing a Taylor expansion of the physical equations governing self-absorption, the authors transformed the intensities of multiple copper (Cu) lines into polynomial variables ($I, I^2, I^3$). This "pre-linearization" allows the subsequent linear PLS model to account for physical non-linearities more effectively.
2. Multivariate Dominant Factor Construction
The model moves beyond single-line analysis by constructing a "multivariate dominant factor." This factor incorporates intensities from multiple atomic and ionic lines of the target element (Cu) as well as interfering elements like Lead (Pb) and Zinc (Zn). By using products and quotients of these intensities, the model explicitly accounts for the inter-element interference that characterizes complex matrices like brass.
3. Secondary PLS Correction
To further refine the results, a secondary PLS step is applied. This step uses the entire spectrum to model the residuals—the differences between the primary model's predictions and the actual values. This "mops up" remaining uncertainties caused by experimental fluctuations or noise that the initial physical model might have missed.
Experimental Setup and Datasets
The study utilized 14 standard brass alloy samples provided by the Central Iron and Steel Research Institute of China. The dataset was split into a calibration set of 10 samples and a prediction set of 4 samples (ZBY906, ZBY907, ZBY924, and ZBY927). Notably, the prediction set included samples with concentrations both inside and outside the calibration range, providing a rigorous test of the model's robustness and ability to extrapolate.
Results and Discussion
The proposed multivariate dominant factor model demonstrated superior performance compared to both conventional PLS and a previously established baseline dominant-factor model.
Key performance metrics include:
- Calibration Quality: The model achieved a near-perfect $R^2$ of 0.999.
- Prediction Accuracy: The Root Mean Square Error of Prediction (RMSEP) was reduced to 1.97%, a significant improvement over the 2.33% achieved by the previous baseline and the 5.25% error seen in standard PLS.
- Overall Error: The total RMSE improved to 1.05% from 1.27%.
- Robustness in Extreme Cases: For the sample with the highest copper concentration (ZBY927), the absolute relative error was remarkably low at 0.49%, compared to 3.30% for the previous best model.
The results underscore a critical insight: physical grounding is essential for non-linear modeling. The authors found that "random" non-linear transformations (e.g., arbitrary polynomial degrees) actually degraded model performance. Only transformations rooted in physical mechanisms, such as the Taylor expansion of self-absorption laws, provided meaningful accuracy gains.
Conclusion
By combining the physical insights of spectroscopy with the statistical robustness of PLS, this research provides a powerful framework for overcoming the quantitative limitations of LIBS. While the specific transformations used in this study were tailored to brass alloys, the methodology offers a roadmap for improving chemical analysis in other complex materials. Future work may focus on refining these physical transformations to further reduce the reliance on secondary statistical corrections, moving closer to a truly "physics-first" quantitative LIBS model.
Summary updated