ABSTRACT

Decision making occurs in all facets of the human experience, and in almost every case requires the consideration of tradeoffs between multiple goals that cannot be fully satisfied simultaneously. For critical chemical manufacturing infrastructure, system design and operation decisions require a balance between, for example, supplying an affordable and reliable stream of products to customers, providing well-paying jobs to the community, maintaining safe performance of all production units, and reducing detrimental environmental impacts. Mathematical tools that allow for the identification of decision-making tradeoffs are essential to ensuring that US industries can meet these varied goals while remaining economically competitive. Unfortunately, existing rigorous methods for doing so do not scale well to problems with many (greater than four) objectives. The proposed research program aims to address this challenge by developing a computational framework that systematically reduces the number of objectives in decision making situations with many criteria by identifying sets of objectives that are correlated, or give similar solutions when considered individually, and grouping them into a single objective. Tools also will be developed that efficiently identify a single decision that provides a high-quality compromise between competing objectives. Through the proposed integrated educational activities, this program also will provide training to the next generation of scientists and engineers in multi-criteria decision-making.

The goals of this research are to develop generalizable methods for objective reduction in many objective optimization problems (MaOPs) which preserve maximum tradeoff information and provide orders of magnitude reduction in the required solve time, and to apply these methods to representative decision-making problems in the chemical process industries. In particular, this project aims to develop methods that (1) provide a first of its kind approach for systematically reducing high dimensional MaOPs a priori to solving the problem, (2) apply machine learning methods and develop a stochastic community detection approach for finding objective groupings that work well in use cases with dynamically evolving or uncertain parameters, such as real time operation and strategic planning, and (3) use a novel robust single objective optimization approach for a priori identification of knee points on many objective tradeoff curves. Publicly available software to implement the above methods will be developed and shared with the broader academic and industrial chemical process systems community for use in furthering research and education in this area, as well as improving industrial process outcomes. The methods developed will be tested using timely chemical systems applications, such as in electrified chemical production, green fertilizer production, and chlorine manufacture and distribution. However, the methods developed in this project will be highly generalizable and applicable to other fields, including but not limited to artificial intelligence, finance, medicine, and robotics. Furthermore, the project team will pursue integrated educational activities such as the developing a set of educational modules for multi-criteria decision making for inclusion throughout the undergraduate chemical engineering curriculum, teaching underrepresented K-12 students about multi-criteria decision making through games, and mentoring K-12 students for academic competition programs in STEM areas.

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.

Please report errors in award information by writing to: awardsearch@nsf.gov.