The term “Digital Twin” is increasingly being used to describe a model-based system representation that is coupled with the real system itself. The term was originally used in production engineering. The model could provide more continuous and complete information on the state of equipment than would be possible with sensors, which could be then used to guide decisions. Information provided by sensors is far less complete in the environment than in production facilities and yet many decisions must still be made. In this blog post I will briefly discuss the digital twin concept in the context of environmental management.
A Digital Twin applied to environmental systems joins physical and virtual representations of the environment and infrastructure for collective exploration of the system in question using simulation and optimisation. I ultimately envisage digital twins of the environment to lead to more efficient infrastructure that both meets human needs while minimising environmental impacts. Digital twins of the environment have not yet found application in practice. I believe that barriers to application are best examined by considering the underpinning models and the coupling with the physical system.
Three central tensions push against each other in a way that makes dragging and dropping models into a Digital Twin from the wide body of environmental research difficult. The first is that the model must represent all parts of the system that are essential to the decision-making process. This tension will prefer a broad model that captures a diverse range of information but with a more focussed representation of key locations that are important to the decision in question. The second is that the decision guiding information will not be useful if the model is not sufficiently representative of reality. This tension will prefer detailed and sophisticated models that can produce accurate predictions. The third is that the model must be sufficiently fast to be useful in a real-time decision-making context. It must clearly be faster than real time and probably much faster to enable on-the-fly uncertainty exploration. I suspect a sort of parsimony is the only way to achieve this. Fast enough but no faster. Detailed only where required and broad elsewhere. Accuracy if possible and quantification of accuracy if not.
The coupling between digital and real should be considered as soon as it is apparent that models of this calibre are possible. The definition above specifies only that some form of coupling need exists and not how detailed it is. At one extreme is an incredibly tight coupling that enables the digital twin to make all operational decisions in the real world. Early definitions of digital twins were primarily focussed on tight coupling with a utopia of fully automated complex systems just over the horizon. At the other extreme is a loose coupling that simply provides curated state variable information to operators for their own interpretation. Loose coupling of digital twins in this sense is a form of model-aided monitoring system. How loose or tight a coupling might be will depend on the amount of trust that the operators are willing to put in the modelled representation of their real system and its decision-making capabilities. The current capabilities of most environmental models suggest that the types of digital twin we might aim to create are on the looser end. We should aim to foster an appropriate amount of trust in our models and ideally trust that is based on a decision-maker’s understanding of the underlying digital twin and the real system. We do not want to find ourselves in the unfortunate position of one environmental consultant:
“We find that the [decision-aiding information] we produce [for our clients at water companies] is either followed rigidly or not at all; we would prefer that it is incorporated with a wider understanding of the water resources system in question.”
Written by Barney Dobson for the Centre for Systems Engineering and Innovation (CSEI) blog