Backed by vast amounts of data
When researchers conduct experiments, they investigate, for example, what happens when different parameters are adjusted in a manufacturing process. This could be the composition of metals in an alloy, or what temperatures or pressures a material should be exposed to. Similarly, sensors can be installed in an existing production line if you want to know something about it.
"With data-driven modelling, you collect large amounts of data and use it to create statistics. This makes these models good at finding correlations between changes in the process and changes in the result," says Jesper Hattel.
The advantage of data-driven models is that they are based on specific situations, which makes them quick to work with, but also only able to say something about similar situations, meaning they are often quite problem-dependent. Therefore, data-driven models are not always sufficient.
"In data-driven models, there is often only correlation and not causation. This is where physics-driven models come in, because they not only tell us that something happens, but also why it happens," says Jesper Hattel.
Based on theoretical understanding
Sometimes knowing why something happens, and not just that it happens, can provide the best conditions for making meaningful and optimizing changes to a process.
With the physics-driven approach, you use your knowledge and theoretical considerations about the laws of physics to create a model that can mimic reality. This is where Jesper Hattel's expertise comes in.
With an understanding of heat transfer, fluid mechanics, solid mechanics and materials science, he can digitally model a manufacturing process using advanced mathematical models. This allows him to predict what will happen without setting a single machine (except the computer) in motion.
"The physics-driven models can say something much more general and contain more factors. However, they are very time-consuming and therefore not suitable for situations where you need to make quick adjustments," says Jesper Hattel.
The best of both worlds
Both data and physics-driven models have their advantages and disadvantages. In the MADE initiative, which aims to strengthen manufacturing in Denmark, Jesper Hattel has been working to combine the two approaches, which is also something he sees more and more projects working on in his role as a reviewer of research applications.
The combination of large amounts of data and an understanding of advanced physics can make the models usable in real time. If a model is fast enough to respond to an impact calculation, it will make it possible to react to deviations from desired quality in a production as soon as the deviation occurs.
"Many researchers are trying to combine data-driven models with physics-driven models. If we succeed, we will have something really powerful," says Jesper Hattel.
