Nondestructive Electrical Permittivity Detection Using ANN-Enhanced Microstrip Sensors

This study introduces a novel methodology for nondestructive estimation of electrical permittivity using microstrip sensors enhanced with artificial neural networks (ANNs). The proposed approach enables accurate and efficient material characterization without dependence on conventional measurement techniques. Frequency shifts induced by materials with varying permittivity values placed on microstrip patch antennas are analyzed, and an ANN-based model is used to estimate permittivity. A dataset comprising 528 samples, generated via CST Microwave Studio simulations across the frequencies at 1.6, 2.4, and 3.2 GHz bands, is used for model training and validation. Optimal performance is achieved using 80% of the data for training, the Levenberg–Marquardt (LM) algorithm, the tangent sigmoid (TS) activation function, three hidden layers, and 20 neurons. The model yields a mean squared error (mse) of 0.0031, mean absolute error (MAE) of 0.0347, and a coefficient of determination ( $R^{2}$ ) of 0.9995, indicating excellent prediction accuracy with high repeatability and low uncertainty. This ANN-based sensor model marks a substantial advancement in nondestructive testing (NDT), offering a robust and noninvasive method for precise permittivity estimation applicable to a wide range of materials.