Sensor drift compensation for gas mixture classification in batch experiments

Abstract

Sensor drift in batch experiments is a well-known problem in mixed gas classification. In batch experiments, gas sensors can be easily affected by environmental covariates that hinder mixed gas classification. To address this problem, we propose a novel end-to-end deep learning model comprising a drift-compensation module and classification module. Utilizing the nonlinear relationship between sensor readings and environmental covariates, the drift-compensation module corrects the drifted sensor readings in batch experiments by minimizing a scatteredness-based fitness function. The corrected values are then fed into the classification module. To train the proposed model, which involves optimizing two different objectives simultaneously, the hypernetwork-based optimization approach with the stochastic gradient descent is employed. We validated the effectiveness of the proposed method for mixed gas classification using synthetic and real gas mixture data collected from the UCI machine learning repository.