Digital Twins: State of the art theory and practice, challenges, and open research questions

Highlights

• We study various research and industrial works and identify the reasons for delay in widespread adoption of digital twin despite its potential. A major shortcoming identified is the lack of universal consensus on the definition and components of digital twin. To this end, we present a conceptualisation and discuss the best practices for its implementation.

• We discuss the current state of machine learning and big data in digital twin, and find that the advancement in these technologies has impacted the adoption and ideal implementation of digital twin.

• We assess the implementation of digital twins in various domains, and find that the architecture is highly impacted by the application domain. This calls for a specialised digital twin for each domain. However, the basic concept of digital twins should still be universal, it is only the prioritising of components that is affected.

• Since digital twins might be implemented across various collaborators or industries, regulations and techniques for data sharing and security need to be implemented. This area is not yet well researched and explored.

• Evaluation and resilience metrics are required to fulfil the ‘self-evolution’ property of digital twins. This is again domain-dependent, and lacking in the current literary works.

Abstract

Digital Twin was introduced over a decade ago, as an innovative all-encompassing tool, with perceived benefits including real-time monitoring, simulation, optimisation and accurate forecasting. However, the theoretical framework and practical implementations of digital twin (DT) are yet to fully achieve this vision at scale. Although an increasing number of successful implementations exist in research and industrial works, sufficient implementation details are not publicly available, making it difficult to fully assess their components and effectiveness, to draw comparisons, identify successful solutions, share lessons, and thus to jointly advance and benefit from the DT methodology. This work first presents a review of relevant DT research and industrial works, focusing on the key DT features, current approaches in different domains, and successful DT implementations, to infer the key DT components and properties, and to identify current limitations and reasons behind the delay in the widespread implementation and adoption of digital twin. This work identifies that the major reasons for this delay are: the fact the DT is still a fast evolving concept; the lack of a universal DT reference framework, e.g. DT standards are scarce and still evolving; problem- and domain-dependence; security concerns over shared data; lack of DT performance metrics; and reliance of digital twin on other fast-evolving technologies. Advancements in machine learning, Internet of Things (IoT) and big data have led to significant improvements in DT features such as real-time monitoring and accurate forecasting. Despite this progress and individual company-based efforts, certain research and implementation gaps exist in the field, which have so far prevented the widespread adoption of the DT concept and technology; these gaps are also discussed in this work. Based on reviews of past work and the identified gaps, this work then defines a conceptualisation of DT which includes its components and properties; these also validate the uniqueness of DT as a concept, when compared to similar concepts such as simulation, autonomous systems and optimisation. Real-life case studies are used to showcase the application of the conceptualisation. This work discusses the state-of-the-art in DT, addresses relevant and timely DT questions, and identifies novel research questions, thus contributing to a better understanding of the DT paradigm and advancing the theory and practice of DT and its allied technologies.