How to choose the right Digital Twin: equation-based, data-driven or hybrid?

Jun 20, 2025 | NeurEco

In the first article, we identified three concrete obstacles to the widespread adoption of Digital Twins: the complexity of building models, the difficulty of working with small data sets, and the obsolescence of models that cannot be updated. This article delves into how different approaches to Digital Twins can address these issues and help you choose the right strategy for your engineering process.

Most engineers are familiar with the concept of Digital Twin. But when it comes to practice, the real questions emerge:

  • What kind of twin makes sense in my process?
  • Do I have enough data to use an ML model?
  • Can I leverage the physical models I already have?
  • Can I build something that adapts over time?

The answer comes through technical thinking: there is no single type of Digital Twin. And each one works best in certain contexts.

The three approaches compared

 

Twin Equation-based Twin Data-driven (ML) Twin Hybrid
Source Physical equations Real data and ML algorithms Physics + ML

Requirements

Simulation tools, engineering expertise Historical datasets, ML knowledge

Both, but integrated

Advantages High interpretability, accuracy Speed, self-adaptability

Robustness, flexibility

Limitations Long lead times, rigid models Need data quality, less transparent More complex to design
Ideal scope Design, validation Monitoring, optimization Integrated processes, complex systems

When to use what: application cases

 

  • Initial design: if you are at a stage where you have no real data yet, and you want to design a new technology by virtually evaluating its performance, a virtual, equation-based prototype is the optimal strategy. Once you have developed and validated the equation-based model, however, you can leverage its results to build a lighter, faster data-driven twin that allows you to run additional simulations and predict additional operational scenarios.
  • Control: use a data-driven, lightweight model of the product that can run in real time to test control logic in various scenarios and debug.
  • Testing and validation: a data-driven twin speeds up test cycles using predictive models based on data from experimental campaigns that have already been run. Also ideal for virtual sensors.
  • Monitoring and maintenance: this is where the data-driven twin shines, especially if it can continually update itself. In many cases it can supplement a physical model with data that keeps it aligned with operational reality.

How to choose

The question is no longer “to do or not to do a twin,” but which approach to take at each stage. Some useful criteria:

  • Do you already have equation-based simulation models? Model results can be a good starting point for generating lighter data-driven models.
  • Do you have real data from testing or operational, even if limited? You can build a data-driven model.
  • Do you need scalability and upgradeability? Hybrid is often the most robust choice.

 

Neureco’s role

Neureco is a tool designed to speed up simulation, testing, operations and product maintenance activities.

It generates data-based, predictive Digital Twins even from limited datasets.

It can work alongside existing equation-based simulation models to correct, extend, or update the Digital Twin according to real data collected over time, or be used on its own, with a pure data-driven approach.

It is designed specifically to be applied in the R&D/Technical team, by those who do not have a machine learning team, but want a tool to improve design, make controls more robust, speed up prototype testing, and better monitor the product in operation.

Download the fact sheet or contact us to see an example of a concrete application in your industry.

Condividi: