Accuracy-Improved and Low-Cost Material Characterization Using Power Measurement and Artificial Neural Network

A machine learning (ML) approach is proposed to provide an accurate retrieval of the dielectric properties of materials. In our earlier work, based on a semi-analytic solution of transmission lines (TLs), we proposed a method for characterizing dispersive materials having a known dispersion model using amplitude-only transmission measurements. In this work, instead of using a semi-analytic solution, the characterization method uses a multilayered artificial neural network (ANN). The proposed characterization method relies on training an ANN that uses the full-wave simulation results of a coaxial line loaded with different materials under test (MUTs) as its training set. The ANN accurately extracts the dispersion model’s parameters, and consequently, the MUT’s complex permittivity profile. For experimental verification, several chemicals were used as MUTs to investigate the utility of the proposed method within the 0.3–3-GHz frequency band. We show that the ML approach introduced here reduces the error by up to approximately 30 times compared with our previous work.