WEB Inversion of the Structure-Property Mapping using Machine LearningWednesday (01.01.2020) 00:15 - 00:30 Part of:
Microstructure design approaches require the exact determination of the structure-property relation. Such approaches highly benefit from a compact mathematical description of microstructures. This is achieved via a representation by suitable features. The relationship between structure and property descriptors can be generalized from the concrete data by regression models (e.g. by artificial neural network) in the form of a Structure-Property Mapping (SPM). One approach consists of a Structure-Property Mapping and an optimizer (e.g Genetic Algorithms), whose objective function represents the desired properties. Alternatively, Generative Model methods can be developed for this purpose. In this context, Generative Models can be learned to represent the conditional probabilities of microstructures, given dedicated properties. In our presentation, we present our development of machine learning based Structure-Property Mapping and Inversion methods in the context of microstructure design.