Gas-Sensor Drift Counteraction with Adaptive Active Learning for an Electronic Nose

Area of Science:

• Sensor Technology
• Artificial Intelligence

Background:

• Gas sensors are crucial for electronic noses (E-noses) but suffer from drift, degrading performance over time.
• Online calibration for sensor drift is challenging, necessitating new approaches.

Purpose of the Study:

• To explore an effective online approach for managing gas sensor drift in E-nose systems.
• To propose and evaluate a novel active learning method, AL-ACR, for dynamic drift counteraction.

Main Methods:

• Utilized active learning (AL) to provide reliable labels for online instances, overcoming limitations of conventional AL methods.
• Developed AL on adaptive confidence rule (AL-ACR) to dynamically address online drift data by selecting instances evenly distributed across categories.
• Evaluated AL-ACR using two E-nose drift databases and compared it against reference methods.

Main Results:

• AL-ACR demonstrated higher accuracy compared to reference methods on both E-nose drift databases.
• The study analyzed the impact of labeling number on AL-type methods' performance.
• Introduced the Labeling Efficiency Index (LEI) to numerically assess labeling contribution.

Conclusions:

• AL-ACR effectively counters online gas sensor drift in E-nose systems.
• AL-ACR achieves superior performance with optimal cost-effectiveness among tested AL methods.
• The proposed method offers a robust solution for maintaining E-nose accuracy in the presence of sensor drift.