RFID-enabled ML-assisted microwave liquid sensor design for complex dielectric characterization of water-methanol mixture

In this study, an RFID tag inspired microwave sensor design is proposed for the dielectric parameter characterization of the water-methanol binary mixture through the RFID tag operating principle with RSSI magnitude and phase output values based on the input variables of operating frequency, RFID reader power strength and sample location in machine learning assisted manner. The proposed microwave sensor design operates at ETSI frequencies of UHF band reserved for RFID. In the characterization of the water-methanol binary mixture by processing the RSSI data received from the RFID reader with machine learning Gaussian Process Regression, the mixing ratios of the liquid components and real and imaginary parts of the complex dielectric constant of the mixture can be conveniently obtained. For the machine learning study, eleven mixtures with 10 % differences, three different power levels, four different frequencies, and four different locations have been combined and carried out with a total of 528 data obtained. Three different machine learning algorithms have been developed using the same input data for three different characterization outputs. R2 values of Gaussian Process Regression method have been obtained as 0.99, 0.99 and 0.98 for the volumetric mixing ratios, real part, and imaginary part of the dielectric constant, respectively.