PI: Luis Zuluaga
University: Lehigh University

Air Products and Chemicals, Inc. (AP) is one of the main producers of industrial gases like oxygen (for hospitals), nitrogen (for chemical plants), argon (for the metal industry), and hydrogen (for refineries). In producing and delivering products to their customers, AP uses capital-intensive assets and highly complex processes. These processes operate in a dynamic, competitive, and rapidly changing environment.

Thus, researchers at AP, in collaboration with the principal investigator (PI), are looking to develop novel, decision-support tools that would allow AP to make the best decisions regarding the use of its production plants and delivery systems in real-time fashion. This will be done by continuously evaluating and adapting to different market, resources, and demand conditions thereby maximizing system profits, while maintaining safety and customer satisfaction.

For that purpose, we intend to leverage a detailed mathematical model of AP’s industrial gases manufacturing and distribution activities developed as a result of collaboration between Air Products and Lehigh University (Lehigh), which was supported by a prior PITA grant. Although this detailed model accounts for the complex dynamics of manufacturing and distribution activities, it has a drawback that, even with state-of-the-art optimization solvers, can only be solved in about five minutes. This means that recommendations of the model will be “dated,” given that actual conditions can change in a matter of seconds. Thus, we plan to use an ingenious combination of “sensitivity analysis” and statistical techniques in order to obtain near-optimal recommendations from the previously developed model in a matter of seconds. The sensitivity analysis tools allow us to compute solutions to a model when some parameters change slightly from a nominal solution, while the statistical tools will allow us to decide the set of nominal solutions that need to be computed over time in order to be able to compute the desired real-time recommendations.