Exploring the Dimensions of Digital Twins: Types, Abstraction Levels, and System Interactions

In the first deep dive, we explore the various types of digital twins, from simulative to prescriptive, each offering unique capabilities and benefits. This section breaks down how digital twins are categorized based on their core functionalities:

• Simulative Digital Twins: Used primarily for virtual testing and experimentation before physical implementation, such category is actually a “light” digital twin, as it is missing a data connection and the interaction elements, as the object does not exist yet in the real world. Typically, a simulative digital twin is more rapidly deployed and more flexible than other types of strong data connection and interaction, due to their testing nature and scope, and lack of heavy system integration effort. Nontheless, based on certain algorithms, they can be particularly relevant in understanding the physical object, in sizing it (i.e. what is the correct size of each productive asset to remove bottlenecks?), and in defining and optimizin its behavior (i.e. how does a robotic arm or crane should move? What is the optimized way to manage it?).
• Monitoring Digital Twins: Focused on real-time data collection and analysis to mirror the current state of a physical entity or process, such digital twin type is not forward looking and does not predict anything about the future state of the object, as it merely report the current status based on the available data via a centralized single point of access to informations. Anyhow, the monitoring digital twin is the basis for both the predictive and prescriptive digital twins as they both rely on current states to start predictions.
• Predictive Digital Twins: Predictive Digital Twins leverage historical and real-time data to forecast future states, enabling proactive decision-making. By applying machine learning and AI to defined rules and historical performance data, these twins focus on anticipating potential future scenarios. They often include decision-making tools that allow users to analyze the impact of various actions on key performance indicators (KPIs). This capability makes predictive digital twins inherently exploratory, supporting quantitative “what-if” analyses that drive faster and more effective decision-making. They are particularly valuable in scenarios where understanding the potential outcomes of decisions is critical to optimizing performance.
• Prescriptive Digital Twins: Going a step further, this type of digital twin is an evolution of the predictive digital twin. While the former rely on static logics, defined by the business mostly as “common sense decisions or rule of thumbs”, in the latter, an artificial intelligence, typically a reinforcement learning agent, is applying optimized dynamic decisions over time by himself. If properly configured and trained, this solution can provide outstanding results, unmatcheable with any other methods.

For each type, we will discuss its applications across different domains—such as product design, manufacturing, and urban planning—illustrating the value they add. By comparing these types, this section will help you identify which digital twin best suits your specific needs.