Decomposition Techniques for Multi-Scale Structured Product Design: Subspace Optimization

摘要

Recent developments in the area of integrated process and product design have shown that products can be designed in terms of their properties without committing to any specific components a priori. Although current techniques make use of group contribution methods (GCM) to design molecules, there are many properties, atomic arrangements, and structures that cannot be represented using GCM. One approach to expand the capability of GCM to handle a more diverse range of solutions is to combine property clustering with decomposition techniques in a reverse problem formulation. This approach first utilizes multivariate characterization techniques to describe a set of representative samples, and then uses decomposition techniques such as principal component analysis (PCA) and partial least squares (PLS), to find the underlying latent variable models that describe the molecule's properties.