Multi-physics Topology Optimization with Applications to Swing Adsorption Systems

The performance of many engineered systems is sensitive to their geometrical design. To remove the human element from the design process and create next-generation systems with optimized performance, geometrical layouts can instead be automatically discovered using computational approaches. Firstly, in this thesis, we explore a novel geometry optimization strategy which utilizes a neural network. Compared to existing techniques, the method is able to discover better performing designs in a more computationally efficient manner. Secondly, we optimize the geometries for components of various systems to obtain configurations that far outperform currently adopted designs.