NSF Award
2421968
This project aims to computationally design, then fabricate and test, an electronic nose for classifying complex gas mixtures. Broadly, the gas sensing technology developed in this project could enable the design and fabrication of electronic noses for indoor and outdoor air quality monitoring, crop monitoring in agriculture, food quality assessment, and health assessment via breath composition analysis. Specifically, the hardware of the electronic nose will constitute an array of porous materials, metal-organic polyhedra (MOPs), coated as thin films on microbalances. Each sensor in the array functions by virtue of gas adsorbing into the film of material, which is registered by the microbalance. The software of our electronic nose will constitute a machine learning algorithm that parses the response pattern of the sensor array and makes a prediction about the gas composition that produced the response. A key advantage of MOPs is their tunability: the pore size and shape and surface chemistry of the MOPs can be tuned to arrive at a highly diverse set of MOPs that interact differently with each species in the gas phase. As a consequence, the response pattern of the MOP-based sensor array will provide a lot of information about the composition of the gas. While first tested for its ability to discriminate between pure analytes, the electronic nose in this project will be tailored and tested for classification of plant oils and, as a lofty goal, grades of olive oil, to counter fraud. For outreach, the project includes development of YouTube videos to educate the public about MOPs and gas sensors and hands-on learning modules with a gas sensor and an Arduino microcontroller. <br/><br/>This project will design, fabricate, and test an electronic nose, consisting of (1) a gas sensor array employing diverse metal-organic polyhedra (MOPs) as gravimetric sensing elements paired with (2) a supervised machine learning model, to discriminate complex mixtures (eg. plant oils). As nanoporous, stable, recyclable, and tunable materials, MOPs may serve as sensitive and selective sensing elements for the next generation of gas sensors. Coating a thin film of MOP on a quartz crystal microbalance (QCM) gives a gravimetric sensor whose response is the mass of gas adsorbed in the MOP film. Mimicking olfactory systems in living organisms, the response pattern of an array of multiple QCM-MOPs—chemically and structurally diverse MOPs—will be interpreted by a supervised machine learning model to discriminate complex mixtures. The computational design of the QCM-MOP sensor array constitutes: (i) construct a database of candidate MOP structural models, (ii) conduct molecular simulations of adsorption of a portfolio of volatile organic compounds in each MOP, (iii) employ dimension reduction algorithms to embed the MOPs into a latent space wherein MOPs with similar adsorption properties congregate, then (iv) use a diversity selection algorithm to curate the most diverse set of MOPs for the array. Next, the investigators will synthesize and characterize the computationally-curated MOPs and employ surface deposition techniques to attach them to QCMs. To test the efficacy of different strategies to inject diversity into MOPs, the investigators will construct three generations of QCM-MOP arrays, wherein the MOPs differ by: (1) metal only, (2) functional group only, and (3) topology, metal, bridging ligand, and functional group. The reversibility, cyclability, and stability of the MOPs will be tested. Finally, the QCM-MOP sensor array will be tested for discrimination of (1) pure compounds, (2) plant oils, and (3) grades of olive oil. Dimension reduction algorithms will aid in exploring the discriminatory capability of each QCM-MOP array, then supervised machine learning algorithms will map its response pattern to a predicted compound/mixture. The design of the 3rd-generation QCM-MOP array will be evolved by replacing the least-informative MOP, flagged by interpreting the machine learning model, with the next MOP in the computational design queue. Finally, to make for a robust QCM-MOP array, the temperature and humidity will be varied for a context-sensitive classifier.<br/><br/>This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
