Machine Learning Regressor for Impedance Parameter Estimation of QCM Sensors for Liquid Media Characterization.

This paper presents a Machine Learning (ML) technique for extracting some meaningful impedance parameters of quartz crystal microbalance (QCM) sensors operating in liquid media. A regressor based on a low-complexity shallow Artificial Neural Network (ANN) is trained to estimate the resistive and inductive changes induced by the liquid in contact with the resonator from the measured impedance around the resonance. The training and test datasets are simulated by the modified Butterworth-Van Dyke model assuming the contact condition with a fluid half-space and varying the resistive and inductive contributions of the fluid in accordance with the theoretical models for Newtonian fluids and with the behavior of real fluids observed during tests. Experimental results obtained with different fluids (e.g., air, ultrapure water, and aqueous solutions at different glucose concentrations) demonstrated the validity and the accuracy of the proposed approach.