Digital twins in the automotive industry: fewer physical tests, more data-driven decisionsIn the aut

As Krzysztof Hyzopski, Simulation Group Leader at ZF CV Systems Poland, emphasizes in an interview with Evertiq, one thing is key: the model must be reliable enough to support a decision.

“The decision-making process, which previously relied largely on experience and risk assessment (based on limited information), can now be supported by a range of verified data. A digital twin allows us to predict how a specific deviation will affect the product’s functionality,” he says.

From a virtual line to a single test

The technology is developing in several parallel directions. In some plants, virtual production lines are being created where the entire process can be simulated — from assembly stations to potential errors and failure scenarios.

“We create models based on geometric data provided by subcontractors. We can run through the entire production process in a virtual environment and verify scenarios even before the line is launched,” Hyzopski explains to Evertiq.

At the operational level, however, the approach is different. At the ZF CV Systems plant in Wrocław, which belongs to ZF Commercial Vehicle Solutions Division in Poland, the digital twin is primarily used in final product testing — where it is necessary to unequivocally confirm that a component or the entire device meets functional requirements.

The model is created based on hundreds of combinations of critical parameters — including those outside standard tolerance ranges.

“We went beyond the permissible deviations to see how the product behaves in extreme cases. This also allows us to assess the quality of deliveries,” he says.

A variance does not always mean a problem

It is precisely at the intersection of manufacturing and the supply chain that one of the most practical benefits becomes apparent.

In the traditional model, a component that does not meet specifications is sent for further analysis and is often even returned to the supplier. This lengthens the process and simply generates costs.

Today, the decision-making process can look different.

“Not every deviation is critical. The digital twin allows us to determine whether a given change will affect the product’s characteristics and whether it will fall outside the acceptable range,” explains Hyzopski.

As a result, some decisions that previously required time and expert evaluation can now be made more quickly based on hard data.

Change without stopping production

A similar mechanism applies when implementing changes on the production line.

Before a new component or configuration goes into production, it is tested in a virtual environment.

“If a supplier or system component changes, we first verify it in the model. Only when we are certain that the change will not negatively affect the product do we halt the line and implement it,” he says.

This strategy not only reduces the risk of downtime but also allows for better planning of when to implement changes.

Time: From Optimization to Preventing Delays

One of the most frequently mentioned benefits of digital twins is the reduction of product development time. In practice, this means reducing the number of physical tests, and sometimes even eliminating them entirely.

“When it comes to vibration testing, we’re talking about days or weeks. For durability testing — months,” emphasizes Hyzopski.

However, the greatest impact is seen where the limitation isn’t the test itself, but access to the infrastructure.

“It happens that a test bench isn’t available for another six months. Eliminating such a test can mean avoiding a real project delay,” he adds.

In such cases, the digital twin ceases to be an optimization tool and begins to determine the delivery date.

More data, more responsibility

At the same time, the use of digital twins does not simplify the development process; quite the opposite.

For the model to replace physical testing, it must be thoroughly verified. This means more measurements, more analyses, and a greater emphasis on risk assessment.

“Creating a digital twin requires a tremendous amount of work. We need to understand the product and all possible scenarios of its operation very thoroughly,” says Hyzopski.

It is crucial not only to model normal operating conditions but also potential failures and their sequence.

“The digital twin must account for everything that a real-world test verifies,” he adds.

AI complements, but does not replace

In this process, the importance of data analysis and AI-based tools is growing, yet their role remains complementary, not dominant.

They do not replace classical physical models, but rather support them where simulation proves insufficient or too imprecise.

“We are not able to simulate everything. In such cases, we generate data and build predictive models,” explains Hyzopski.

In practice, this means combining various approaches — from classical simulation to regression models and tools that detect anomalies or specific patterns.

At the same time, every such solution must be properly verified.

“In the automotive industry, we cannot afford false data. Everything that comes out of the model must be checked and confirmed in reality,” emphasizes the expert.

Policies as a turning point

The biggest change, however, may come not from technology, but from regulatory policies.

In Europe, work is underway to allow virtual testing and certification in homologation processes.

“This could be one of the biggest regulatory changes in recent years,” Hyzopski assesses.

If the new approach is implemented, digital twins will cease to be merely an optimization tool and will become a formal part of the product approval process.

For now, however, the industry is proceeding in stages.

Digital models today cover selected elements — from the supplier’s production process, through assembly, to product lifecycle analysis and predicting its wear and tear over time.

“We are only at the beginning of this journey. Looking at the big picture, this is a prospect of several, perhaps a dozen or so years,” concludes Hyzopski.

Where simulation ends and decision-making begins

Contrary to appearances, a digital twin is neither “artificial intelligence” nor a simple simulation.

Its foundation remains physics, measurement, and testing. It is only on this basis that models are built, allowing us to predict a product’s behavior even beyond scenarios that can be directly replicated.

Artificial intelligence comes into play where these scenarios begin to fall short. It helps fill in the gaps, but it does not eliminate the need for verification.

In the automotive industry, this distinction is not merely academic. It determines whether a decision can be made without physical testing and whether it can be justified.