A New Technique of Using a Neural Network Classifier in Microwave Microfluidic Sensors for Purity Detection of Liquids

In this manuscript, a novel microwave microfluidic sensor is presented, which is equipped with a new technique for using neural network in purity detection of liquids. This sensor is composed of two components, a planar semicircular resonator and a PDMS microfluidic holder. The sensing method is based on the shift in the resonance frequency and quality factor of the resonator in presence of liquids with different complex permittivities. Binary mixtures of ethanol-water and some common liquids were investigated to evaluate the performance of the proposed sensor. Experimental measurements show a very high sensitivity equal to 0.52% for complex permittivity in the range of 1-78.4. To overcome the challenges of analyzing the experimental data by the user, a neural network-based classifier has been introduced, which is used for the first time. A set of 20 pure and 20 impure samples' measurement data were used to train the network and a set of 27 measurement data were used to test the neural network for each tested liquid. The obtained results show that the predicted data of the neural network were correct for all cases. The proposed sensor has a simple design, compact size, and easy fabrication process, while it is fabricated on an FR4 substrate, which makes it cost-effective to be use. Also, the detection of the purity or impurity of the liquid under test (LUT) is done by a neural network with a considerable accurate and provides a condition for creating a continuous measurement process.