A novel approach to oil-water mixture analysis: Microstrip antenna-based sensor combined with GMDH neural network

Abstract

Microstrip antenna sensors are gaining recognition in the oil industry for their compact design, cost-effectiveness, and versatility in measuring complex fluid mixtures. This study presents a novel microstrip antenna-based sensor featuring a quasi-circular resonator, designed specifically for measuring the volumetric content of oil in oil-water biphasic mixtures. Operating at three distinct resonance frequencies (5.88 GHz, 6.96 GHz, and 8.1 GHz) and housed within a 3D-printed PLA cylindrical structure, the sensor accurately analyzes spectral responses for 21 different volumetric oil-water ratios within a fixed volume of 20 mL. To address the challenge of reproducibility often associated with microstrip sensors, each mixture was tested five times and the data was fed into a Group Method of Data Handling (GMDH)-type neural network. This approach enhances the neural network's resistance to minor input variations, ensuring more reliable predictions. The network successfully predicted the volumetric oil content with an Root Mean Square Error (RMSE) of less than 0.93 and an R² value of 0.99. With an average sensitivity of 4.26 MHz/ε_r, the sensor demonstrates high effectiveness in fluid characterization, particularly in oil industry applications where precise, real-time monitoring is essential. This study not only advances microstrip antenna sensor technology but also highlights its practical utility in enhancing operational efficiency and decision-making processes within the oil sector.