An Integrated Platform for Multi-Model Digital Twins

The notion of Digital Twin is known as a means to access otherwise dispersed lifecycle data of industrial devices, and enabling advanced reasoning on top of the data via various kinds of models (e.g. machine learning, simulation). Despite many studies on digital twins, there is still a need for common architectures, platforms and information meta-modelling that enable defining various lifecycle data in a harmonized way, as well as integrating the information with machine learning and simulation models; a gap that is filled by this paper. Our approach for the integration of various digital twin models addresses three known technical debt in machine learning systems: data pipeline jungle, undeclared/unstable data dependencies and undeclared consumers. Adopting such an integrated digital twin platform can reduce the required time and effort to develop and maintain digital twin-based solutions, as well as laying a foundation to support a variety of digital twin-based use cases.