Complex Permittivity Extraction for Ethanol-Water Mixtures Characterization using Artificial Neural Networks

Dielectric characterization of liquids is of importance in biomedical applications, such as microwave imaging and cancer detection for the optimization of the diagnostic process. This paper proposes a way to characterize ethanol-water mixtures using a regression model applied by an artificial neural network. This approach is beneficial for characterization when only a limited number of samples is available. A neural network is constructed using a finite set of measurement data for mixtures with 20, 40, 60, and 80 percent of ethanol as a training set. The complex permittivity values for mixtures with 30, 50, 70, and 90 percent of ethanol are deduced using this neural network with an average error of 2.17% for the real part and 3.49% for the imaginary part, respectively. The results indicate a good agreement between the model's predicted values and expected values obtained from coaxial probe measurements. The use of machine learning in this context reduces the number of samples that need to be prepared as well as the number of experiments that have to be conducted.