Digital twins: a practical guide to digital twins in industrial technology

In recent years, the term “digital twin” has taken a firm place in the lexicon of industrial technology. The promise is great: to virtually replicate a physical system in order to simulate, predict and optimize its behaviour. But people have been talking about it for a long time. So why are digital twins not yet a concrete and everyday presence in engineering departments?

The answer is not just technological. It concerns the way in which we understand, build and adopt these solutions.

The first distinction that needs to be made is between descriptive and predictive digital twins. It is the transition from descriptive to predictive solutions that marks the quantum leap in most concrete industrial applications.

What is a predictive digital twin (really)?

A predictive digital twin is a dynamic model that represents a real physical system in order to monitor, analyze or predict its behavior. It is not a simple simulation, but a model that lives and is updated.

The main types of digital twin

Today, there are different types of predictive digital twins, which differ in the way they are built and updated.

• Physical (equation-based): Modeling based on physical equations and technical knowledge. They are accurate but complicated to create and update.
• Data-driven: created by analyzing data collected from the actual product (tests, sensors, logs, etc.). It is faster and more flexible, but depends on the quality and availability of the data.
• Hybrid: combines both logics by integrating a physical model with data-driven corrections or enhancements.

Obstacles to widespread acceptance: where the digital twin gets stuck

Many engineers would like to build or use a Digital Twin in their own process, but encounter concrete obstacles. The most common are:

1. The skills to create a model

Whether equation-based or data-based twin, skills are required to create them, engineering skills on the one hand and data science skills on the other. What is certain is that software houses are working on simplifying their products to create user interfaces that are increasingly simple and usable by non-experts.

2. The available data is not perfect

Industrial data sets are often fragmented, incomplete and unsynchronized. This makes it difficult to apply classic, purely data-driven methods.

3. Models are quickly outdated

Equation-based twins in particular work well for redesign or optimization, but have difficulty adapting to real-world operating conditions: Deterioration, changing environments, real-world usage. Keeping a rigid model up to date is costly and not very scalable.

The result? In many cases, digital twins in companies have remained limited to individual projects, pilot projects or isolated process steps without becoming an everyday tool.

A new approach is needed

For digital twins to become truly operational, they need

• A more flexible approach
• A technology that also works with little data
• Simple tools that can also be used by non-experts

In the next article, we will take a closer look at data-driven digital twins.

We will see how their strategic use throughout the product development and lifecycle, either independently or integrated with equation-based siblings, provides a concrete answer to these obstacles. And how lightweight platforms like Neureco make it possible to build predictive models even when little data is available.