We investigate a simulation methodology for systematically optimizing air cooling in an existing battery system by placement of passive components. The goal in such retrofit optimization is to achieve design improvement by making as few and cheap changes in the original system as possible. Our methodology utilizes CFD for fluid flow and heat transfer modeling and machine learning for cause-effect assessment across binary design variables, such as wall placement for passive flow control. As an application, we consider computational optimization of air cooling in a scaled-down electric bus charging station battery system.