Digital twins, shadows, and models: Understanding the spectrum
2 min
Digital twins have become a widely discussed concept in process engineering, yet the term is often used inconsistently. The term digital twin originated in aerospace and manufacturing. It originally referred to a high-fidelity, evolving virtual model of a specific physical asset, like a jet engine, tightly linked to real-time sensor data to mirror its performance, wear, and behaviour.
In process engineering, what is often called a digital twin is really a digital model: a dynamic simulation used to test how a process responds to changes in feed, temperature, or equipment performance. These models can be highly valuable, but they typically are not tied to a specific physical asset in the way a true digital twin is. Understanding the terminology can help choose the right modelling approach for a given application. Digital representations of processes exist along a spectrum of complexity and integration:
Digital model
A standalone physics-based, data-driven, or hybrid model. It simulates how a process responds to changes in inputs such as feed composition or temperature, but they do not receive live data from the physical process. These models are widely used for design, scenario analysis, optimisation, and training.
Digital shadow
A model that is connected to live data and that reflects the process in real-time. Unlike a digital twin, it does not send information back to the process. Digital shadows are useful for monitoring, prediction, and performance assessment, providing insight without directly influencing operations.
Digital twin
A model of a specific physical asset that receives live data and can send information back to the asset. It evolves with the asset and can support or even automate operational decisions. True digital twins are typically reserved for high-value assets where the investment in integration and fidelity can be justified.
Selecting the appropriate level of model complexity and integration ensures the right balance between investment and operational impact. Regardless of where a model sits on the spectrum, three considerations ensure it delivers value:
Fidelity: The model should accurately capture the relevant phenomena required to reflect process behaviour.
Responsiveness: The model should solve quickly enough the support decision-making or real-time monitoring, if required.
Adaptability: The model should be updatable (offline or online) as new data become available, allowing it to remain stable over time.
In conclusion, digital twins represent the most integrated point on a broader modelling spectrum. The goal is not to build the most complex model possible, but the most appropriate one for the decision at hand. When fidelity, responsiveness, and adaptability are balanced correctly, digital representations become practical tools that support understanding, improve decisions, and evolve alongside the process. In this way, digital twins and their counterparts move from abstract concepts to applied, value generating assets.
