A Model-driven Approach for Knowledge-based Engineering of Industrial Digital Twins.

Digital twins are heralding a new paradigm in the process and manufacturing industries by providing real-time decision support for a range of key problems. Engineering digital twin solutions is a knowledge and effort intensive activity. In-depth knowledge of manufacturing plant, process operations, underlying physics, data, and technical problems is essential in the design of digital twin application for a given business objective. Industry domain experts, data scientists and solution developers must collaborate to build it with the required features and functionalities. Currently, it is not an easily scalable process. For each industry vertical and even for the same plant type, the development process has to be repeated manually. To address this, we present a model-driven knowledge-based approach where knowledge can be captured in a machine processible form and reasoned with, to systematically guide the solution development process. Knowledge is modelled at three levels of abstraction, namely, meta, plant type and plant instance. We describe how this helps in generalizing the knowledge, in driving the knowledge acquisition process, and in contextualizing knowledge to specific problem instances. We then outline how the captured knowledge can be exploited to guide various digital twin engineering tasks such as problem definition, model building (physics or data-driven), data preparation, algorithm selection, etc. In particular, we discuss how the approach helps in arriving at a detailed digital twin specification starting from a high-level business problem statement. We present a case study demonstrating the approach on a real-life industrial problem.