Machine learning-assisted wearable sensor array for comprehensive ammonia and nitrogen dioxide detection in wide relative humidity range

Abstract

The fast booming of wearable electronics provides great opportunities for intelligent gas detection with improved healthcare of mining workers, and a variety of gas sensors have been simultaneously developed. However, these sensing systems are always limited to single gas detection and are highly susceptible to the inference of ubiquitous moisture, resulting in less accuracy in the analysis of gas compositions in real mining conditions. To address these challenges, we propose a synergistic strategy based on sensor integration and machine learning algorithms to realize precise NH₃ and NO₂ gas detections under real mining conditions. A wearable sensing array based on the graphene and polyaniline composite is developed to largely enhance the sensitivity and selectivity under mixed gas conditions. Further introduction of backpropagation neural network (BP-NN) and partial least squares (PLS) algorithms could improve the accuracy of gas identification and concentration prediction and settle the inference of moisture, realizing over 99% theoretical prediction level on NH₃ and NO₂ concentrations within a wide relative humidity range, showing great promise in real mining detection. As proof of concept, a wireless wearable bracelet, integrated with sensing arrays and machine-learning algorithms, is developed for wireless real-time warning of hazardous gases in mines under different humidity conditions.