Training a reduced order model with a large range of variables speeds up simulation of a fuel cell system by 100 times

Driving automotive innovation

Since 1946, Plastic Omnium has been driving innovation in the automotive industry. It began with using plastics to improve vehicle components and reduce overall weight and has evolved over the past 75 years to meet the challenges of today’s world.

The latest evolution is the Plastic Omnium New Energies Division, which has been created to
advance hydrogen mobility and enable vehicle manufacturers to reduce their emissions. Jurgen Dedeurwaerder, a system engineer at Plastic Omnium, says, “The storage of hydrogen
and low temperature fuel cells are key to the success of this new technology.

“Plastic Omnium acquired companies already active in the development and production of
fuel cell systems, but they worked mainly on demonstrators. We needed to adapt this for the
automotive industry so the fuel cell systems could be mass produced at high quality. To do
this, we needed to master both our development and production processes.”

Enhanced processes and reduced costs

For any new business venture to be successful, cost-effectiveness is vital, and this was no different for the New Energies Division.

“We knew that simulation would not only improve our processes, but also help keep costs down,” says Dedeurwaerder. “So it was essential from the start.”

To accomplish this, the firm adopted Simcenter™ software, part of the Siemens Xcelerator business platform of software, hardware and services. Dedeurwaerder first used Simcenter Amesim™ software to build a complete model of the fuel cell systems, including all its components. However, while this model could be used to optimize the architecture and production cost, the level of detail in the model made exploring all options time-consuming.

To speed this up, Dedeurwaerder took advantage of Simcenter Reduced Order Modeling software. “We trained the original model with a large range of variables such as the piloting of actuators and environmental conditions,” he says. “With those training sets we used Simcenter Reduced Order Modeling to create the reduced order model. This resulted in a much faster running plant model that carried out simulations about 100 times faster than the original model.”

This plant model is then converted to a black box model, which is used by control and software engineers to develop and test their algorithms. The same process is used later in development for hardware-in-the-loop (HiL) testing and to give customers a fast, accurate model they can incorporate into their full vehicle models to test within their own environment.

“We also carry out full vehicle simulations in-house for our customers,” says Dedeurwaerder. “We use Simcenter Amesim to look at different powertrain configurations, battery sizes, vessel sizes and fuel cell power levels. This allows us to interact with our customers to find the optimum design.”

Optimized testing

Dedeurwaerder says that a major benefit of using Simcenter Amesim and Simcenter Reduced Order Modeling is the decrease in the amount of physical testing of the complete system. “It’s not necessarily a reduction in the total amount of testing,” he explains. “But the testing is more oriented toward individual components. So you then build a model with components that you’re confident in. With such an accurate model, much of the analysis that was previously done by full system testing can now be carried out with simulation.

“Simulation also gives us a better understanding of phenomena occurring inside our fuel system. For instance, you might see a sudden unexpected drop in temperature that physical testing couldn’t explain. But by using simulation you discover that it is due to liquid water being present instead of vapor only. And it’s that vaporization of the liquid water that caused the temperature drop.”

He also notes the importance of the flexibility of simulation compared with physical testing. “It makes it much easier to assess performance in different environments. It’s difficult to set up test benches at certain altitudes or temperatures, but with simulation we can choose any conditions we want. We now verify new components with simulation first, so by the time we build the system, we’ve usually got it right the first time, which saves a lot of time and money building further systems.”

Protecting intellectual property

Reduced order models have also proved invaluable when full physics details of components are not available. “In some cases, the physics aren’t fully understood, or the supplier can’t disclose them due to intellectual property rights,” says Dedeurwaerder. “So we stress the component in all directions and record the variable data. This is used to train a reduced order model that we integrate into our system model.”

Similarly, using Simcenter Reduced Order Modeling allows Plastic Omnium to share models with customers and suppliers while still protecting its own intellectual property. “We can allow them to see cell voltages and net power, but they can’t view all the details of the model,” Dedeurwaerder explains. “As we work more closely with partners, we can gradually disclose more details as the relationship progresses, but it’s always within our control. And they can never see everything inside the system model or the specific algorithms we use.”

Dedeurwaerder has been quick to extol the virtues of Simcenter Reduced Order Modeling to partners: “We had a supplier who only worked with physical models. Integrating the physical model inside a complete vehicle model would be too slow to run. So we showed them the benefits of using a reduced order model to explore complex boundary side effects and behave as a physical model would, but in a much shorter time. You end up with a better final product at faster speed as you take all the boundary effects into account.”

From simple to complex

Before adopting Simcenter Amesim and Simcenter Reduced Order Modeling, engineers at Plastic Omnium were using a spreadsheet for a lot of their calculations. “It was used in an intelligent way to get good results,” explains Dedeurwaerder. “But it can’t include all the physics such as gas or liquid properties that evolve with temperature and pressure. The data is static and it’s very inflexible – if you change a component, you must start again from scratch.”

With Simcenter Amesim and Simcenter Reduced Order Modeling, engineers start with a simple model and then increase the complexity as the understanding improves. “The Simcenter products are really easy to pick up,” says Dedeurwaerder. “But you must learn how to generate the training data to be used in Simcenter Reduced Order Modeling. Being able to start simple has allowed us to easily integrate it with our processes and evolve our use as we get to know it better.

“It’s also easy to export and import data to and from other systems. For instance, we’re currently building a plant model with several reduced order models and circuitry around them. We set and test the conditions and then integrate it with the complete model using the Simulink interface. It all works well together.”

Dedeurwaerder believes that without simulation it would be almost impossible to address all the engineering challenges the company faces: “You’d be reliant on testing that is much more expensive. And you can’t afford to make mistakes because that extends the time and cost of development even further.”

Instead, simulation delivers results quicker with fewer resources. “Simcenter Reduced Order Modeling lets us accelerate our simulation models to the point where a detailed fuel cell plant model runs significantly faster than real time with the same accuracy as a full system model,” says Dedeurwaerder. This enables activities such as model-in-the-loop controller development and testing to be done faster, shortening the overall development cycle by around 25 percent. At the same time, it gives us a reliable, IP-protected and cost-effective way to distribute models to other teams, both internally and to our customers to augment the quality of their own products and processes. This results in better quality products being delivered to end users.”

New markets and continuous improvement

With the automotive industry fast progressing toward zero-emission vehicles, Plastic Omnium needs to move with it and take advantage of the latest technology to develop these vehicles. Historically, hydrogen mobility has been used for cars and trucks, but they are now looking to expand to include large ships and vessels, opening a move into the shipping market.

The company also has a separate software division focusing on the likes of autonomous driving and internet of things (IoT) applications. “This will enable live monitoring and a better understanding of how fuel cell systems degrade over time,” says Dedeurwaerder. “Then we can feed this back into development; for instance, changing the way the required humidity is
generated by the electrochemical reaction at the cathode.

“As technology continues to evolve, the digital twin is vital to maintain a competitive edge. That’s why Simcenter Amesim and Simcenter Reduced Order Modeling are so important to us. They (Simcenter Amesim and Simcenter Reduced Order Modeling) enable us to see inside our products and learn from real-time data to make improvements in future iterations. I expect to see more suppliers using this technology, which will help us all develop better products.”