Engineering a Digital Twin for Manual Assembling.

The paper synthesizes our preliminary work on developing a digital twin, with learning capabilities, for a system that includes cyber, physical, and social components. The system is an industrial workstation for manual assembly tasks that uses several machine learning models implemented as microservices in a hybrid architecture, a combination between the orchestrated and the event stream approaches. These models have either similar objectives but context-dependent performance, or matching functionalities when the results are fused to support real-life decisions. Some of the models are descriptive but easy to transform in inductive models with extra tuning effort, while others are purely inductive, requiring intrinsic connection with the real world.