Decision Analysis Services (DAS) is an independent professional services company. We provide professional services globally to clients within the Energy, Defence, Government, Health, and Transport sectors. We specialise in bridging the gap between strategy and operational execution.

DAS contributes to research, via our sponsored PhD and MSc students, using AI and ML tools to interrogate data and provide foresight to decision makers. This, combined with the work undertaken by our Innovation Lab places us in the position of being able to capitalise on, and incorporate AI, ML and Data Science developments in work for our clients.

Introduction

Over the past two decades, digital twins have emerged as powerful tools in domains such as predictive maintenance and performance optimisation and have significantly enhanced decision-making capabilities.

Digital twins comprise of a virtual representation of a physical system, process or entity which is updated using real-time data. These virtual representations can take the form of physics-based models, simulations, statistical models, etc. which mirror the characteristics and behaviours of their real-world counter parts. Traditional digital twins, however, often rely heavily on physics-based models (models of real-world systems that follow the laws of physics), which can be limited by the availability of accurate data and the complexity of the system being modelled.

This article will explore innovative approaches, centred around system dynamics (SD) and Machine Learning (ML), which push the definition of digital twins. These methods may allow the concepts of digital twinning to be applied to a wider range of problems, where constructing a fully comprehensive model of the physical counterpart may not be possible. By integrating these methods, we aim to provide pioneering, versatile and adaptable digital twins that can deliver actionable insights across many industries.

BACKGROUND

Digital twins are defined as virtual models which mirror the behaviour and performance of their physical counterparts through real-time data integration and simulation. There are three components to a digital twin: virtual model, which enables the system to be represented and integrated with; the refresh rate, which provides information from the real system to update the model; and simulation capability, which allows for the testing and evaluation of alternative scenarios (such as future states, alternative configurations, etc.). The key technologies enabling digital twins include Internet of Things (IoT) devices for data collection, advanced analytics, and simulation software.

Two particularly interesting developments in digital twinning are the use of ML only digital twins, and system dynamics based digital twins. By harnessing their power, organisations can not only apply digital twinning across a wider range of applications and domains, but also significantly enhance decision-making processes. In turn, leading to the development of more proactive strategies, optimised performance, and measurable improvements in efficiency and innovation.

MACHINE LEARNING ONLY DIGITAL TWINS

The virtual model of many digital twins in physical engineering systems includes detailed physics-based models, where each part and interaction is clearly understood and represented. This allows operators to simulate changes by adjusting inputs and processes to determine the impact on key factors of the system. However, this approach requires extensive knowledge of the system. In settings where this level of understanding is not possible, machine learning (ML) models can predict the impact on key factors without explicitly modelling all the intermediate processes. This capability enables digital twin approaches to be applied to complex systems where robust modelling of every process and subsystem would not be feasible.

For instance, in industrial maintenance, an ML-based digital twin can predict optimal maintenance intervals by analysing sensor data, detecting patterns, and forecasting failures, reducing downtime and costs. In healthcare, ML-enhanced digital twins of patients can integrate data from wearables and health records to predict condition progression, enabling personalised treatment plans and timely interventions, which is particularly valuable in managing chronic conditions. In urban planning, ML-driven digital twins can model traffic, energy consumption, and environmental impacts, helping planners optimise infrastructure and enhance sustainability. By simulating various scenarios, such as new transportation routes or climate change impacts, these models support more informed decisions. In the energy sector, ML-based digital twins can predict power grid failures, optimise energy distribution, and better integrate renewable energy sources by forecasting output and balancing supply and demand. Similarly, in manufacturing, ML-enhanced digital twins can optimise production by predicting equipment wear, streamlining supply chains, and reducing waste, leading to increased productivity and lower costs.

These examples demonstrate how ML-powered digital twins can be applied across diverse industries, offering predictive insights and optimising decision-making, even in complex, dynamic environments where traditional models may fall short.

SYSTEM DYNAMICS-BASED DIGITAL TWINS

System dynamics is another robust method for developing digital twins, especially for complex systems characterised by numerous interacting components and feedback loops. Unlike physics-based models that focus on detailed physical interactions, system dynamics models emphasise the relationships and feedback within the system, making them ideal for simulating long-term behaviour and strategic planning. System dynamics models can also incorporate social dynamics, offering insights into how behaviours and interactions influence system-wide outcomes over time

For instance, in project management, a system dynamics-based digital twin can analyse progress by considering factors such as scope creep, rework rates, staffing availability, staff morale and experience. System dynamics models allow a structural representation of activities of an organisation or system and can be tuned (calibrated) using actual data. By regularly updating the model with real-time data, project managers can gain insights into potential delays and resource bottlenecks, enabling informed decisions to keep the project on track. Additionally, these models can simulate the impact of various interventions, helping organisations to optimise resource allocation and improve project outcomes.

By integrating machine learning and system dynamics into digital twin frameworks, DAS is pioneering more flexible and intelligent models that can handle a wider range of applications. These methods allow digital twins to adapt to changing conditions and provide actionable insights, ultimately bridging the gap between strategy and operational execution.

CONCLUSION

Digital twins are revolutionising various domains by providing real-time insights, predictive analytics, and optimisation capabilities. They have had a significant impact on sectors such as energy, defence, healthcare, government, and transport, driving efficiency and innovation. Machine learning and system dynamics are enhancing the versatility and accuracy of digital twins. Other emerging technologies such as edge computing and augmented reality could further amplify these capabilities. For example, edge computing can improve the responsiveness and efficiency of digital twins by processing data closer to the source. Augmented reality, on the other hand, can enhance visualisation and interaction with digital twins, providing more intuitive and immersive experiences.

By expanding the methods and applications of digital twins through advanced technologies like machine learning and system dynamics, we are unlocking new possibilities for data-driven decision-making. These innovations are set to transform how we design, manage, and optimise systems across various industries, paving the way for smarter, more resilient operations.