Distinguishing chemicals using CMUT chemical sensor array and artificial neural networks

Abstract

Capacitive micromachined ultrasonic transducers (CMUTs) can function as extremely sensitive mass-loading chemical sensors. The resonant frequency of the CMUT changes as mass is added due to chemicals absorbing into a chemical-sensitive layer on the top of the plate. However, these sensors suffer from the problem that they are not selective to a single chemical. As a solution, we present a system of four CMUT chemical sensors with different functionalization layers. Neural networks are used to do pattern recognition on the sensor outputs in order to distinguish different chemicals. The system is capable of distinguishing water, ethanol, acetone, ethyl acetate, methane and carbon dioxide in air at concentrations less than 1% with 98% accuracy. Once the chemical is identified, the concentration can be determined using polynomial regression with an RMS percentage error ranging from 1.1% to 13%, depending on the analyte.