Industry collaboration yields top Optimization and Engineering paper

As soon as he was introduced to supply chain optimization in a master's course taught by Chrysanthos Gounaris, Dev Kakkad ('22) wanted to "go deeper into the weeds" of the subject. The following semester, he took the advanced graduate-level optimization course taught by Ignacio Grossmann. Kakkad also joined Grossmann's research group in the Center for Advanced Process Decision-making (CAPD) and set out to address a practical problem for industry partner and CAPD sponsor Aurubis: how to make optimization models more explainable.

The resulting paper has been recognized as the best paper published in the journal Optimization and Engineering in 2024. Kakkad, who led the research as a master's student at Carnegie Mellon, and Grossmann, the Rudolph R. and Florence Dean University Professor of chemical engineering, share the Howard Rosenbrock Prize with their co-authors, Bianca Springub, Christos Galanopoulos, Leonardo Salsano de Assis, Nga Tran, and John Wassick. The citation reads: "The proposed method combines methodological novelty with practical applicability, offering both theoretical insights and significant potential for application in industrial optimization."

In industry, linear programming models are commonly used to optimize limited resources and make informed decisions. Supply chain management and optimization is one such application. For employees without technical training, however, optimization models can be opaque. Real-life operations have practical considerations that may not be explicit in the model. Some problems may have multiple optimal solutions. Evaluating alternate solutions or the next-best solution can help a decision-maker understand the results of an optimization model more fully.

Grossmann, Kakkad, and their co-authors propose an algorithm that, instead of finding only one solution to a linear programming problem, finds a pool of alternate solutions. For instance, it can find the optimal, second best, third best solutions, and so on. Their paper describes their iterative mixed integer linear programming (MILP) algorithm and applies it to several supply chain optimization problems.

For example, a company may want to optimize the flow of raw materials to its processing plants. If they already have in place long-term contracts for the raw materials, then they may prefer a solution that requires fewer changes. Using Grossmann and Kakkad's algorithm, the company would be able to compare the financial gain from alternate solutions with the number of changes required to the status quo.

Once Grossmann and Kakkad determined that their algorithm could successfully generate alternate solutions to an optimization problem, they needed to test its scalability. "In real-world applications, the supply chain models are quite large," says Kakkad. Aurubis, a leading global copper recycler and provider of non-ferrous metals, provided anonymized data for two case studies using their supply chain structure. Grossmann and Kakkad generated alternate solutions, which were evaluated by optimization model users at Aurubis. "There are many real-world events that could disrupt a supply chain. In cases like that, it's extremely useful to have a ready list of alternate solutions," says Kakkad.

Kakkad's research in the Center for Advanced Process Decision-making at Carnegie Mellon came with networking opportunities that have helped him continue to pursue his interest in optimization. He is now a research engineer on the process systems engineering team at KeyLogic, a contractor for the National Energy Technology Laboratory. He works on both traditional chemical engineering applications and operations research applications, such as carbon capture technologies and supply chain optimization.

For media inquiries, please contact Lauren Smith at lsmith2@andrew.cmu.edu.