High-accuracy prediction of microcantilever sensor for fluid detection: simulation to deep learning

Abstract Viscosity and density measurements with small volume samples of fluids are required in several industrial applications. The surrounding fluid impacts the vibrating microstructures of microcantilever micro-electro-mechanical-system sensors, influencing their dynamic behavior. In this study, the behavior of microcantilever sensors immersed in fluids through simulation and deep learning methods is used to simultaneously estimate the density and viscosity of fluids from the resonance frequency and quality factor of the microcantilever. The simulation is carried out by creating an appropriate environment to produce the frequency and quality factor values of the microcantilever. Then, the simulation results are used to train a deep neural network that maps the input data to the desired parameters. This deep learning method shows high prediction accuracy as long as the input data is sufficiently independent, does not show bias in the prediction values, and provides instant results. The resulting error values, namely the mean error, max error, and mean standard error of generalization on viscosity, are 0.5%, 0.8%, and 1.46, respectively. The mean error, max error, and mean standard error in the density results are 1.76%, 7.92%, and 0.01, respectively.