Title:	LIBS of BaSrTiO3 compounds using ANN based on a stationary state plasma model
Authors:	Amogh, M S Xavier, Sebin Sebastian Jose, Jeena Rose Philip, Reji Biju, P R
Keywords:	LIBS CF-LIBS Laser produced plasmas Artificial intelligence Artificial neural network Machine learning
Issue Date:	Apr-2026
Publisher:	Spectrochimica Acta Part B: Atomic Spectroscopy
Citation:	Spectrochimica Acta Part B: Atomic Spectroscopy, 2026, Vol. 238 (4), AR No. 107480
Abstract:	Laser Induced Breakdown Spectroscopy (LIBS) is widely used for rapid material characterization and quantitative analysis due to its versatility and speed. Linking LIBS spectra to precise quantitative measurements is challenging because of nonlinear effects arising from surface variations, matrix interactions, and self-absorption. As a result, precise quantitative analysis has been a persistent challenge for the LIBS community. We perform stoichiometric evaluation of BaSrTiO₃ (BST) samples with varying compositions using a stationary-state plasma approach. BST is notable for its high dielectric performance and is employed in applications such as non-volatile memories, pyroelectric sensors, and electro-optic devices. We simulate the LIBS spectra using a two-zone plasma model under Local Thermodynamic Equilibrium (LTE) and fit them to experimental spectra to extract stoichiometry, electron density, and plasma temperature. A Controlled Random Search (CRS) algorithm is used to optimize the fit to find the stoichiometry accurately. Further, an Artificial Neural Network (ANN) is trained exclusively on synthetic spectra representing the constituent elements, and when applied to experimental spectra, it accurately predicts the stoichiometry of samples not seen during training. To our knowledge, this work is the first to employ a two-zone stationary-state plasma model to generate synthetic training data for an ANN, which is then successfully applied to predict elemental compositions from experimental spectra. Our findings show that a simple ANN based on one-dimensional plasma modelling enables rapid in-situ classification and stoichiometric analysis, reducing reliance on large experimental spectral datasets for machine learning. This approach could simplify the complexities associated with experimental spectral requirements for training and predicting elemental concentration using ML models.
Description:	Restricted Access.
URI:	http://hdl.handle.net/2289/8684
ISSN:	0584-8547
Alternative Location:	https://doi.org/10.1016/j.sab.2026.107480
Copyright:	© 2026 Elsevier B.V.
Additional information:	Supplementary Material: https://www.sciencedirect.com/science/article/pii/S0584854726000315?pes=vor&utm_source=clarivate&getft_integrator=clarivate#ec0005
Appears in Collections:	Research Papers (LAMP)

Items in RRI Digital Repository are protected by copyright, with all rights reserved, unless otherwise indicated.