10/17/2022
By Murat Inalpolat

You are all invited to attend a master's thesis defense in Mechanical Engineering on the "A Data-Driven Design Optimization Framework for Vanadium Redox Flow Batteries with Improved Electrode Microstructure Characteristics.”

MS Candidate: Alina Berkowitz
Date: Friday, Oct. 28, 2022
Time: 1 to 2:30 p.m. (US EST)
Location: This will be a virtual defense via Zoom. Those interested in attending should contact MS candidate Alina_Berkowitz@student.uml.edu and Advisor Murat_Inalpolat@uml.edu at least 24 hours prior to the defense to request access to the meeting.

Committee Chair: Murat Inalpolat, Ph.D., Associate Professor, Mechanical Engineering, UMass Lowell

Dissertation Committee Members:

  • Ertan Agar, Ph.D., Associate Professor, Mechanical Engineering, UMass Lowell
  • Xinfang Jin, Ph.D., Assistant Professor, Mechanical Engineering, UMass Lowell

Abstract:
Present-day large-scale stationary renewable energy sources lack the design flexibility, reliability, and storage capabilities that the vanadium redox flow battery (VRFB) can provide. Although the relatively high cost of this promising technology has hindered widespread implementation in the beginning, many researchers have realized that significant cost reduction can be achieved by improving the performance of cell-level components. The system-level efficiency of large-scale VRFBs is heavily influenced by the performance and reliability of the flow-cell stack-up. One of the key components in this stack-up is the electrode. The electrode is made of porous carbon media, and heavily affects cell performance as it is responsible to mass transport as well as providing active surface area for reactions to occur.

Previous research has identified that structured carbon fiber arrangements improve electrode efficiency. However, operating conditions vary from user to user, and thus electrode functionality differs for each set of operating conditions. Therefore, there will be a different set of optimal electrode parameters for every set of operating conditions. This requires a significant amount of data to be generated for conclusive analyses and decision-making on the battery design.

In this study, a high-fidelity sampling plan is used to mitigate the costs associated with data-generation while also providing the space-filling qualities necessary to train a supervised machine-learning model on a relatively small training dataset without compromising the prediction accuracy. The best performing machine-learning models are used to create a surrogate model (“meta-model”) along with the objective function for multi-objective optimization using a non-dominated sorting genetic algorithm II (NSGA-II). Finally, the optimizer is employed to explore the design space and find an improved set of electrode features for the set of operating conditions specified.