Why do we use Machine Learning?
Machine learning provides a state-of-the-art tool to analyze experimental data and infer key properties of systems under observation. The Hanson Research Group has been generating high-quality spectroscopic and kinetic experimental data for many years. Our current efforts combine contemporary machine learning models with our precise measurement techniques for inference of important properties of fuels, molecules, and molecular spectra. The methodologies used by our team members are successful in predicting 20+ hydrocarbon properties and understanding intricate multi-species chemical processes. Our group provides expertise and insight in the accurate prediction of molecular properties that play a significant role in developing next-generation energy systems and fuels.
What techniques do we apply?
Fuels are complex and our group has utilized in-house convex optimization, regularization, support vector regression, and quadratic programming techniques to understand their properties. Researchers from our group apply novel algorithms on sparse and dense datasets obtained through a wide variety of shock tube, spectroscopy, and flame experiments to characterize and predict key properties of various species simultaneously using convex optimization. Similarly, regularization and generalized linear models are being implemented to unravel the complexities of functional groups in the infrared region.
 Y. Wang, W. Wei, Y. Zhang, R. K. Hanson, “A new strategy of characterizing hydrocarbon fuels using FTIR spectra and generalized linear model with grouped-Lasso regularization,” Fuel, Vol. 287 (2021) 119419. DOI: 10.1016/j.fuel.2020.119419
 Pinkowski, N. H., Ding, Y., Johnson, S. E., Wang, Y., Parise, T. C., Davidson, D. F., & Hanson, R. K.(2019). “A multi-wavelength speciation framework for high-temperature hydrocarbon pyrolysis,” Journal of Quantitative Spectroscopy and Radiative Transfer, 225, 180–205. DOI: 10.1016/j.jqsrt.2018.12.038