Planar microwave sensor with machine learning analysis for non-destructive quantification of methanol contaminants in distilled spirits

This paper presents a dual-port planar microwave sensor designed for the non-destructive detection and quantification of methanol contamination in distilled alcoholic beverages. The proposed sensing structure is based on a compact U-shaped microstrip line integrated with a Defected Ground Structure (DGS) that combines a Complementary Split-Ring Resonator (CSRR) and a Complementary Interdigital Capacitor (CIDC). This hybrid DGS configuration enhances the local electric field distribution and strengthens the electromagnetic coupling between the resonant elements, thereby improving the overall sensitivity of the sensor. Consequently, the device exhibits a stable and reliable response over a wide range of liquid dielectric permittivities, from 1 to 100, with an average percentage sensitivity of 1.39% and consistent, traceable resonance behavior across the entire range. Additionally, the proposed approach achieves low error in methanol concentration estimation (below 5%), confirming its accuracy and reliability. The performed measurements conducted at ambient conditions, combined with a geometrically robust design and enhanced electromagnetic field confinement, mitigate the effects of environmental fluctuations and fabrication tolerances, ensuring stable and reliable methanol detection. Machine learning methods were employed to correlate the sensor resonance frequency with the methanol concentration, using 240 real samples obtained from three types of distilled beverages at eight concentration levels. The proposed approach achieved coefficients of determination (R 2 ) exceeding 0.95, demonstrating the potential of the developed sensor and data-driven methodology for accurate and non-destructive quality assessment of distilled spirits.