NSF Award
2404384
With support from the Chemical Measurement and Imaging Program in the Division of Chemistry, the research groups of Dr. Karl Booksh at the University of Delaware and Dr. Barry Levine at Oklahoma State are developing new computational tools and methods for “soft sensing” in chemical industry applications. “Soft sensing” refers to the application of data science techniques to infer key process indicators (KPIs) and critical process parameters (CPPs) when these indicators and attributes cannot be directly or conveniently assessed by physical sensors. Advanced applications for soft sensing are being developed and explored through collaborations with industrial partners at Merck and Arkema. Merck is interested in developing new soft sensing technologies for understanding and optimizing the production process for biopharmaceuticals, targeting the recombinant expression of virus-like particles (VLPs) in yeast-based bioreactors for vaccines and the immobilization of biocatalytic enzymes on resin and polymer beads. Arkema is interested in the production process of polymers, using soft sensing to detect contamination in polymers assuring final product quality. The benefits of soft sensing are projected to extend beyond industrial applications. For example, soft sensing can be applied to environmental and health related monitoring to estimate KPIs that cannot be directly measured. The project addresses the persistent need for highly trained professionals in chemical data sciences. Graduate and undergraduate students will develop state of the art chemical data science and machine learning tools and apply them to chemical problems presented by the industrial partners.<br/><br/>The approach being devenoped in the Booksh and Lavine labs combines advancement of multiway chemical data science methods – machine learning tools designed to exploit the structure of data cubes or higher order tensors – with industry-driven needs to maximize understanding and control of chemical processes. Multiway methods, under the broad category of Tucker models, promise to better extract the chemical information imbedded in the interactions among measurement modes. Specifically, the team is developing excitation-emission matrix fluorescence (EEMF) as an on-line or in-line soft sensor for KPIs and CPPs of VLP production processes, targeting significant improvement over currently available multivariate sensing options. They are also developing improved methodology for collection and analyses of hyperspectral images for soft sensing of polymer contaminants and biocatalytic enzymes immobilized on resins. Methods that determine statistically meaningful confidence limits, especially when the associated error matrix is structured and non-Gaussian, are being developed. Additionally, modified versions of the popular multivariate curve resolution approach (MCR) are being developed to better extract KPIs/CPPs associated with minor components from hyperspectral images.<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.
