Metal–Organic Framework-Based Chemiresistive Array Paired with Machine Learning Algorithms for the Detection and Differentiation of Toxic Gases

Abstract

The development of low-power, sensitive, and selective gas sensors capable of detecting and differentiating toxic gases is pivotal for safety and environmental monitoring. This paper describes a chemiresistive sensor array comprising a series of three conductive hexahydroxytriphenylene-based metal–organic frameworks (MOFs) (M₃(HHTP)₂ (M = Ni, Cu, Zn)) capable of detecting and differentiating parts-per-million (ppm) levels of carbon monoxide (CO), ammonia (NH₃), sulfur dioxide (SO₂), hydrogen sulfide (H₂S), and nitric oxide (NO), as well as binary mixtures of SO₂ and H₂S in dry nitrogen at room temperature. This capability arises from variations in the identity of the linking metal and the framework packing pattern across the materials in the array. To visualize the response pattern of the sensor array and map it to a predicted gas composition, principal component analysis and random forest classification are employed. Both machine learning techniques confirm the ability to discriminate CO, NH₃, SO₂, H₂S, and NO analytes as well as binary SO₂/H₂S mixtures at ppm concentrations using the response of the array. Moreover, a feature importance method applied to the classifier assigns importance scores to each sensor in the array to quantify the impact of individual materials on analyte discrimination. Spectroscopic investigations provide insight into how the structural features of the MOFs influence sensing performance and ascertain material–analyte interactions governing sensing selectivity for SO₂/H₂S binary mixtures.