CSRR chemical sensing in uncontrolled environments by PLS regression

Abstract

Complementary Split Ring Resonators (CSRRs) have been widely researched as planar sensors, but their use in routine chemical analysis is limited due to dependence on high-end equipment, controlled conditions, and susceptibility to environmental and handling variations. This work introduces a novel approach combining a CSRR sensor with machine learning (ML) to enable reliable quantification of compounds. A low-cost benchtop CSRR system was tested for ethanol concentration prediction in water (10–96%), using 450 randomized measurements. PCA was applied for data exploration, and a PLS regression model with Leave-One-Group-Out cross-validation achieved a 3.7% RMSEP, six times better than univariate calibration (23.4%). The results show that ML can mitigate measurement uncertainties, making CSRR sensors viable for robust, low-cost concentration analysis under realistic laboratory conditions.