Online Learning for Active Odor Sensing Based on a QCM Gas Sensor Array and an Odor Blender

In this work, a previously proposed procedure of active odor sensing was improved. During this active sensing measurement procedure, an odor blender was controlled to reproduce accurately the target odor of a previously measured blended odor. The final blended odor was the result of the active sensing measurement procedure. The odor was measured by a sensor array composed of four quartz crystal microbalances (QCMs) coated with different sensing films to obtain frequency changes and resistance changes. The odor blender mixed two volatile organic compounds, ethyl valerate and propionic acid. The procedure was robust to some degree against sensor drift or changes of ambient conditions such as humidity because of relative measurement. However, a recalibration was needed under changing environment. In this article, we present an improvement of the procedure with online learning of the parameters that avoid recalibrations. During the active measurement, the model parameters are updated by recursive least squares and the control loop is redesigned. This procedure was tested in different scenarios, sensor drift, and humidity. The results show that the system can adapt itself to sensor parameter changes improving the robustness of the active sensing procedure using the online learning.

Automated summary

This work improves the previously proposed procedure of active odor sensing by introducing online learning of parameters to enhance the robustness. The procedure involves measuring odor using a sensor array and reproducing the target odor using an odor blender. The results demonstrate that the system can adapt to sensor parameter changes, improving the robustness of the active sensing procedure.