Adsorptive Gas Sensor Response Forecasting to Enable Breath-by-Breath Analysis

MOS gas sensors have proven to be useful in electronic noses, which utilize these sensors to detect volatile organic compounds in human breath to detect various lung diseases. Unfortunately, the long settling time of MOS gas sensors is ill-suited to measuring human breath, where complete breathing cycles are often shorter than 5 s. Existing studies circumvent this limitation by collecting gas samples and injecting them into a sealed chamber to react with the sensors. However, it would be convenient if breath-by-breath analysis could be conducted without the need to store breath samples. To accomplish this, we present a novel forecasting methodology to predict the final value t∞ of a gas sensor's response based on its initial transient behavior. To do this, we present and validate a second-order mathematical model of the sensors' response characteristics, which we then use in our preliminary work using neural networks to predict the final sensor value. Although some challenges were encountered, the initial results are encouraging, and we plan to extend our study in the future to collect a more expansive dataset and explore the use of other types of machine learning algorithms for this application.