Sensitivity Improvement in the Detection of Trace Elements With Magnetic Field Confined Laser‐Induced Breakdown Spectroscopy Using Machine Learning

ABSTRACT

Laser‐induced breakdown spectroscopy is a versatile analytical technique for rapid multi‐element detection; however, its sensitivity is often limited due to low emission intensity and high detection limit for trace elements. In this work, the enhancement of LIBS performance is systematically studied with magnetic field confinement combined with machine learning‐based analysis for trace elements in Aluminum alloys. A magnetic field of 0.5 T is applied to confine the plasma, and its effect on emission characteristics and analytical sensitivity is evaluated in comparison with unconfined (0 T) conditions. Six certified Al alloys were used to collect the emission spectra with and without magnetic field confinement. A significant enhancement in emission intensity is observed under magnetic field confinement, indicating effective plasma plume confinement. This improvement led to enhanced plasma excitation conditions and improved LIBS detection sensitivity. Furthermore, the limit of detection (LOD) of trace elements is evaluated using the calibration curve method to establish the relationship between spectral emission and elemental concentrations. The results demonstrate that the magnetic field confinement significantly reduced the LOD values in all selected trace elements. The LOD values of Mg (II) 279.55 nm and Fe (II) 288.25 nm were significantly reduced from 0.0645 to 0.0446 and from 0.0829 to 0.0673, respectively, under magnetic field confinement. The reduction of LOD under magnetic field confinement is attributed to the plasma confinement, resulting in more uniform radiation. Furthermore, to assess the quality and reliability of the LIBS signal, three machine learning models, namely Partial Least Squares Regression (PLSR), Support Vector Regression (SVR), and Gaussian Process Regression (GPR), were used. The comparison between the actual and predicted concentrations in regression plots revealed improved agreement in the presence of magnetic field confinement, as evidenced by increased values. Additionally, among the applied regression models, GPR achieved the superior predictive performance based on the evaluation parameters, RMSEP, ARE, and , owing to its strong capability to model nonlinear relationships and reduce prediction uncertainty. The proposed technique, integrating magnetic field‐assisted plasma enhancement with machine learning, provides an effective approach for improving the detection capability of LIBS and offers strong potential for high‐sensitivity trace element analysis.