A hybrid linear–nonlinear model with two-stage feature selection for TiO2 content prediction in ilmenite based on XRF spectroscopy

X-ray fluorescence (XRF) spectroscopy provides an efficient and non-destructive means for determining the Titanium Dioxide (TiO2) content in ilmenite. However, the accuracy of this technique can be compromised by matrix effects, and its industrial application is often bottlenecked by the limited ava...

Full description

Saved in:
Bibliographic Details
Published in:Minerals engineering Vol. 236; p. 109929
Main Authors: Dai, Wei, Duan, Mingzi, Nan, Jing, Wang, Lanhao
Format: Journal Article
Language:English
Published: Elsevier Ltd 01.02.2026
Subjects:
Geometric control strategy
Hybrid model
Ilmenite
Stacked structure
X-ray fluorescence
Hybrid model
X-ray fluorescence
Geometric control strategy
Stacked structure
Ilmenite
ISSN:0892-6875
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:X-ray fluorescence (XRF) spectroscopy provides an efficient and non-destructive means for determining the Titanium Dioxide (TiO2) content in ilmenite. However, the accuracy of this technique can be compromised by matrix effects, and its industrial application is often bottlenecked by the limited availability of labeled samples. To address this issue, this paper proposes an efficient method (L(FeTi)-CMI-SICA) for predicting the TiO2 content in ilmenite. The paper initially proposes a two-stage feature selection method based on physical mechanism analysis and mutual information statistics, aimed at addressing the issue of inaccurate feature selection due to matrix effects. Subsequently, this paper introduces a Stacked Interpretable Constructive Algorithm (SICA) for resource-constrained industrial scenarios, which achieves precise modeling of the complex nonlinear relationships in XRF data through a multi-layer feature abstraction mechanism. Finally, this paper employs SICA as the nonlinear component of the model, combining it with linear least squares regression to construct a linear–nonlinear hybrid model. Extensive experiments on the ilmenite XRF dataset demonstrate that our method not only achieves superior performance across key metrics but also exhibits remarkable robustness under the constraint of small-sample conditions. •A two-stage feature selection method is proposed to accurately identify key features for TiO2 content prediction in ilmenite.•A hybrid model is developed to effectively correct matrix effects and enhance prediction accuracy.•Stacked Interpretable Constructive Algorithm (SICA) employs a multi-layer feature abstraction mechanism to precisely model complex nonlinear relationships in XRF data.•Extensive experiments on industrial dataset show that our method outperforms existing approaches in prediction accuracy.
ISSN:0892-6875
DOI:10.1016/j.mineng.2025.109929