Metal–semiconductor sulfide bilayer based SPR fiber optic sensors for the detection of methanol adulteration of ethanol using machine learning algorithms: an investigative study

R. Sreekanth,S. Abraham Sampson

DOI: https://doi.org/10.1007/s11082-024-07579-x

IF: 3

2024-09-29

Optical and Quantum Electronics

Abstract:The study explores the use of surface plasmon resonance (SPR)-based fiber optic sensors, coated with alternating layers of metal and semiconductor sulfide materials, for detecting methanol adulteration of ethanol. Gold, silver, and copper are chosen as the metal layers for the excitation of SPR, with cadmium sulfide (CdS)/zinc sulfide (ZnS) as the semiconductor sulfide materials. Sensors with varying thicknesses of metal layers, i.e. 40 nm, 45 nm and 50 nm, are considered while the thickness of the material varies from 0nm (only the metal layer) to 3 nm in 0.5 nm increments. The sensor with a 40 nm gold layer and 3 nm CdS layer showed a maximum sensitivity of 1.47 nm/vol.% of methanol adulteration in ethanol. Apart from sensitivity, the figure of merit (FOM) and detection accuracy (DA) of all the sensors are computed and compared. Multiple machine learning approaches, such as logistic regression (LR), K-nearest neighbours (KNN), support vector machine (SVM), linear discriminant analysis (LDA), random forest (RF), and decision tree (DT), with the stratified K-fold cross-validation method of train/test split, are used for the detection of methanol adulteration of ethanol using the sensor outputs. The CV splits and the hyperparameters are tuned to achieve maximum performance for the models. DT achieves an accuracy of 91% of prediction with 87.1% of F1_score at max_depth=5 and CV=7, when trained with the outputs of all the sensors, and KNN shows 92% accuracy of prediction with 86.6% F1_Score at n_neighbours= 15 and CV=10 for the selected sensors.

engineering, electrical & electronic,optics,quantum science & technology