Consistency Matters: Building Consistent Digital Twin Virtual Entities

The concept of Digital Twin (DT) encapsulates a real-world entity (RE) and one or more virtual entities (VEs) bidirectionally connected to this RE. These VEs mimic certain aspects of the RE to facilitate various use-cases, such as predictive maintenance and optimization. DTs typically encompass various models, often developed by experts from different domains using diverse tools. To maintain consistency among these models and ensure the continued functioning of the system, effective identification and resolution of any consistency issues are imperative. In this thesis, we study consistency management in the context of DTs and develop a framework to address this challenge.

Based on our interview study with 19 DT researchers and practitioners, we discovered that various consistency issues are often encountered during the development and maintenance of DTs. There exist tools and methods for addressing these issues. However, these are limited in terms of capability and applicability. Therefore, the topic of consistency management remains largely unexplored in the context of DTs. In this thesis, we address this gap by developing a consistency management framework focused on managing the consistency of DT models, specifically the ones contained within VEs. To develop the framework, we first perform a detailed analysis of the characteristics of DT models, identified and categorized through a review of the current literature. Based on these characteristics, we elicit a set of essential requirements that must be fulfilled by any system aiming to manage the consistency of DT models. We evaluate our consistency management framework by implementing it in three case-studies and assessing its ability to address the consistency management requirements. These case-studies demonstrate the framework’s ability to address these consistency management requirements. We also identified several limitations of the framework and discuss possible ways to address them.

The dynamic, data-driven, and interconnected nature of DTs introduces various consistency challenges beyond model-level ones. A key example is behavioral inconsistency between the RE and its VE, detecting which is essential for ensuring equal outcomes, i.e., a mimicry relation. We present a case-study implementing a DT using machine learning (ML) and 3D models to detect inconsistencies between the expected behavior (encapsulated with a VE) and the actual behavior of an autonomous soccer robot (RE). This demonstrates the feasibility of using DTs for automated inconsistency detection.

Finally, despite being a well-known research topic, consistency management has not been sufficiently explored in the context of DTs. Our research on DT model consistency management and RE-VE inconsistency detection addresses this research gap. We demonstrate the relevance and applicability of our methods with various casestudies. Furthermore, we strongly believe that continued research is essential in identifying further possible consistency issues in the context of DTs and in developing methods for identifying and addressing them. The concept of DTs have enabled a wide range of use-cases and will continue to drive new ones. To fully realize their potential, DT engineering must be supported by appropriate tools and methods, with consistency management as a foundational concern throughout the lifecycle.