A Survey on Digital Twin: Definitions, Characteristics, Applications, and Design Implications
A Survey on Digital Twin: Definitions, Characteristics, Applications, and Design Implications
ABSTRACT When, in nineteen fifty-six, Artificial Intelligence was officially declared a research field, no one would have ever predicted the huge influence and impact its description, prediction, and prescription capabilities were going to have on our daily lives. In parallel to continuous advances in Artificial Intelligence, the past decade has seen the spread of broadband and ubiquitous connectivity, embedded sensors collecting descriptive high dimensional data, and improvements in big data processing techniques and cloud computing. The joint usage of such technologies has led to the creation of digital twins, artificial intelligent virtual replicas of physical systems. Digital Twin technology is nowadays being developed and commercialized to optimize several manufacturing and aviation processes, while in the healthcare and medicine fields this technology is still at its early development stage. This paper presents the results of a study focused on the analysis of the state-of-the-art definitions of Digital Twin, the investigation of the main characteristics that a Digital Twin should possess, and the exploration of the domains in which Digital Twin applications are currently being developed. The design implications derived from the study are then presented: they focus on socio-technical design aspects and Digital Twin lifecycle. Open issues and challenges that require to be addressed in the future are finally discussed.
I. INTRODUCTION
I. INTRODUCTION
In nineteen fifty-six, John McCarthy organized a summer workshop, entitled the "Dartmouth Summer Research Project on Artificial Intelligence", which is now considered by many the seminal event where Artificial Intelligence was officially declared a research field. At the workshop, researchers from several disciplines met to clarify, define ideas and establish a research program concerning "thinking machines". They chose the name "Artificial Intelligence" for its broad sense, to avoid restricting the interests of this field to subjects such as cybernetics, automata theory and complex information processing.
Today, Artificial Intelligence concerns the theory and development of computerized systems able to imitate and simulate human intelligence and behavior.
essentially being human-like rather than becoming human, and performing tasks normally requiring human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages.
Since nineteen fifty-six, Artificial Intelligence researches have succeeded in developing intelligent systems allowing machines doing not only all of the physical work, but also the reasoning, the predicting and the subsequent decision-making. Rather than trying to achieve a perfect replica of the human mind, Artificial Intelligence systems exploit processes emulating human reasoning as a guide to provide both aiding tools and better services.
For this reason, and thanks to the continuous advances in the computational power, in Big Data processing, and in the machine learning and pattern recognition fields, Artificial Intelligence applications are becoming a fundamental part of our everyday life, providing surprising benefits in several fields. Examples are researches in the medical fields, where
Artificial Intelligence algorithms are developed with the aim of discovering novel biological relations and treatments. Similarly, Artificial Intelligence algorithms modeling biological structures and human reasoning are integrated either to develop Computer Aided Diagnosis Systems, aiding clinicians during their everyday diagnostics procedures, or to study organs' functioning and reaction to pharmacological treatments, eventually uncovering the hidden patterns and information encoded by the data, by reducing the data dimensionality to remove redundant information.
In the past twenty years, the advent of the Internet of Things is changing the way data are exchanged among different sources. Indeed, the diffusion of technologies such as embedded sensors and actuators connected through the Internet, allows a continuous exchange of Big Data. This term refers to data Volume (having high dimensionality and requiring the storage of large amounts of data), Variety (data with heterogeneous nature, belonging to different sources), Velocity (the speed of production and acquisition, opposed to long processing time), and Value (the significance of the information carried by data). Luckily, scientific advances in data fusion techniques, high-dimensional data processing, big data analytics and cloud computing allow to store and elaborate Big Data to obtain important knowledge and improve the performance of physical systems.
More specifically, the integration of AI models (of physical objects) and Big Data Analytics for processing IoT data [15]-[17] motivates one of the latest, and prob- ably one of the most important advancement in the field of technology, that is, the Digital Twin (DT). DT models are gaining more and more interest for their potentials and strong impact in application fields, such as manufacturing, aerospace, healthcare, and medicine.
Despite successful Digital Twin technologies are now being investigated in the scientific field and are massively spreading in the corporate and business environments, literature works have never described in detail the characteristics of a generic Digital Twin. Indeed, each state-of-the-art paper concentrates on the development of few components of Digital Twins.
In this work, we searched for answers to three specific research questions, related to the state-of-the-art definitions of Digital Twin technology, the main characteristics that a Digital Twin should possess, and the domains in which Digital Twin applications are currently being developed. After the presentation of the research background (Section Two) and the methodology used for the study (Section Three), three sections present the answer to each of the research questions (Section Four, Five, and Six). In Section Seven, we discuss some design implications that emerged from our analysis, while Section Eight presents the open issues and main challenges that still exist in this field of research.