Development and Experimental Validation of an Artificial Neural Network Model of a Microwave Microstrip Resonator for Humidity Sensing

The focus of this paper is on the modeling of a gas sensor based on a microwave microstrip resonator aimed for humidity detection. This is because humidity sensors have been widely applied in different fields, like healthcare, environmental monitoring, meteorology, and industrial processes. The propagative structure of a microwave resonator sensor is covered with a humidity sensing layer whose frequency dielectric properties vary with the change of humidity, causing changes in the sensor microwave properties. The frequency- and humidity- dependent behavior of the reflection coefficient of the studied sensor is modelled by using artificial neural networks (ANNs). To achieve a model which will reliably and accurately predict the reflection coefficient, the prior knowledge input (PKI) approach is implemented. The data used for the model development have been acquired by measuring the reflection coefficient in the frequency range (3.4 divided by 5.6) GHz and for different relative humidity values, in the range (0 divided by 83) %rh. The ANN-based model has been developed and experimentally validated, allowing an accurate reproduction of the measured properties of such sensor under test and prediction even at an operating condition not used during the ANN training. This demonstrates the good capabilities of the achieved model to learn and generalize.