Selective Detection of Mixtures via a Single Nonselective Sensor—Making the Unworkable Sensor Workable by Machine Learning

Abstract

Highly selective and sensitive detection of chemical mixtures by a single sensor consistently suffers from overlapping response signals. Previous researchers have to make a detour to avoid overlapping through preparing selective sensor materials, tedious preseparation, and time‐consuming operation. Nowadays, machine learning (ML), as one remarkable branch of artificial intelligence (AI), possesses advantages to modernize pathways to approach chemical challenges and provides a novel short cut for chemical mixture detection. Herein, a challenging example of a single nonselective cataluminescence (CTL) sensor with nonequilibrium and especially nonlinear responses is taken for chemical mixture detection. Intelligent recurrent neural networks (RNNs) with gated recurrent units are applied to directly predict mixture component contents from overlapping signals and achieve 100% and 92.22% of prediction accuracy for known and unknown samples, respectively. Furthermore, the physical law of response overlapping is revealed to support the mechanism of the proposed approach. The introduction of RNN into selective sensing not only initiates a distinctive and simple pathway for selective quantitative detection of mixtures via a single nonselective sensor but also can inspire future intelligent detection approaches in analytical chemistry, biological analysis, environmental monitoring, healthcare, clinical therapy, etc.

Keywords

• gas sensors
• machine learning
• recurrent neural networks
• selective detection
• signal overlap