Improved quantification of naphthalene using non-linear Partial Least Squares Regression

TL;DR

This paper enhances naphthalene quantification using non-linear PLSR variants, achieving improved linearity and resolution with a WO3 sensor across a range of concentrations.

Contribution

It introduces a double-logarithmic transformation and LW-PLSR to better model non-linear sensor responses for naphthalene detection.

Findings

Double-logarithmic PLSR improves linearity and resolution.

LW-PLSR outperforms ordinary PLSR at higher concentrations.

Resolution of 4 ppb in 0-20 ppb range.

Abstract

A test dataset is generated using temperature cycled operation with a WO3 metal oxide semiconductor (MOS) gas sensor. Six concentrations of naphthalene from 0 to 40 ppb are measured and, subsequently, used to evaluate the performance of three variants of Partial Least Squares Regression (PLSR). Ordinary PLSR produces highly non-linear models due to the non-linear response of the sensor. Double-logarithmic data results in a model with much better linearity which has a resolution of 4 ppb in the range from 0 to 20 ppb. The more complex Locally Weighted PLSR (LW-PLSR) produces an even better model, especially for higher concentrations, without making any assumptions for relationships in the underlying data.