Artificial intelligence in additive manufacturing

Artificial intelligence, or AI for short, is on everyone's lips and is already being used in numerous applications. Well-known examples include voice assistance systems such as Siri, Alexa and Google Assistant. Additive manufacturing can also benefit from AI technologies, for example when it comes to developing new materials or monitoring processes. But how much work can AI take away from us and to what degree can we trust the decisions of AI systems? We talked about this with Professor Dr.-Ing. Johannes Schilp from the University of Augsburg. He is head of the Chair of Production Informatics with a focus on digitalization and networking in production, head of department at Fraunhofer IGCV and a member of the expert panel of the Additive Manufacturing Coordination Office of Bayern Innovativ.

Mr. Schilp, additive manufacturing is already a digital technology. Let me ask you provocatively: What still needs to be digitized?

Prof. Schilp: Let me formulate a counter-question to this. For topical reasons: what would happen if a manufacturing company really sent all employees except the production workers on the store floor to the home office ? It would probably quickly become clear that there is definitely digitization potential in the order flow process chain. And that's despite the fact that some companies certainly consider themselves to have a high level of digitization.

Even companies that want to successfully use digital production technologies such as additive manufacturing must ask themselves the following questions: How well does my data integration work, i.e. the vertical but above all the horizontal networking within the process chain? Are there the necessary end-to-end data interfaces between the individual process steps or machines from different suppliers? The individual flexible production modules, as I call them, must be continuously networked with each other in the automated process chain so that the use of relevant data can be successfully implemented and meaningful decisions can be made on their basis. This should be supported in the future by KI .

What is needed to integrate additive manufacturing into production processes? And where is it still stuck?

Prof. Schilp: The production processes must be even better integrated into the existing production systems, not only in terms of data technology, both horizontally and vertically. Let's first look at horizontal networking: here, there are still interruptions in data forwarding, as the entire process chain is usually not mapped digitally in a continuous manner. The lack of transparency across the process chain makes for greater susceptibility to errors, even when integrating with relevant, already existing production controls.

The vertical networking in order processing must also take into account the special features of additive manufacturing . In addition, additive manufacturing is currently used mainly for small batch and medium batch production. As batch production, there are certain intervals, so-called discrete production intervals, which must be converted into a continuous flow in series production. This represents a challenge both in terms of production times and in a technical sense in terms of automation.

In which additive change processes is AI already used and how does AI specifically support optimizing these processes?

Prof. Schilp: From a production technology perspective, we are already familiar with learning algorithms from learning processes, for example from applications such as intelligent image processing. Here, the artificial intelligence method toolbox supports the evaluation of measured data and either leads to autonomous decisions or supports a production employee in making the right decisions, for example regarding processes or production quality.

In the future, AI will certainly find much broader use - not only in process chains, but in entire value networks. There are already initial application examples in learning processes, including in additive manufacturing . Process data is intelligently processed, for example, in self-organization, autonomous process control as well as production planning, which are supported by AI methods. AI methods should find their way into the production portfolio in order to be able to take the right planning steps, respectively, to then arrive at an efficient and ultimately robust and, of course, effective production.

To train AI, you need a large amount of high-quality data. How do you get at this data?

Prof. Schilp: The prerequisite for collecting high-quality data is an intelligent sensor network in the production system. Our project experience shows, however, that not only the collection of the data is the problem, but also to bring them into a usable form. From a scientific perspective, we approach this issue by defining a Digital Twin. The Digital Twin is a process model that is enriched by real production data in order to perform simulations with it. This makes it easier for companies, for example, to make valid predictions for future production orders.

To what extent can we trust the decisions of AI systems?

Prof. Schilp: In order to be able to trust the data, a valid database and also a realistic model are important. This does not necessarily have to be a purely physical model. Data-based models are also playing an increasingly important role. For trust, there must also be a basic understanding of how the AI system works.

And who bears responsibility for these decisions: humans or machines?

Prof. Schilp: Artificial intelligence cannot take decisions away from us - but it can support us in decision-making. AI systems can, for example, show a production employee various options for action, but he must make a final decision himself. The use of AI also eliminates the need for time-consuming preparatory work. This has the advantage that employees can concentrate more on their core tasks. At the moment, responsibility still clearly lies with humans, but this issue will have to be clarified legally in the future.

How well positioned is Bavaria as a location when it comes to using AI for additive manufacturing processes?

Prof. Schilp: In Bavaria, we are well positioned both in the field of artificial intelligence and in additive manufacturing . This is certainly due to the fact that both research fields are supported by various funding programs and initiatives. Above all, it is positive that AI and additive manufacturing are not considered separately in current calls for proposals, but that the focus is on the interplay of both topics.

What role do Bavarian research institutions and funding play in the future of AI in additive manufacturing processes?

Prof. Schilp: Here, too, we are very well positioned and promote further progress in this innovative topic area. It is positive that the initiatives are not concentrated regionally, but are anchored at various science and university locations throughout Bavaria . A large number of projects funded by the Bavarian Research Foundation, for example, have already been completed and the results are quite promising. I am confident that we will be able to continue this success story and that, in the long term, innovations will emerge that are successfully implemented by companies.

Do you also work with small and medium-sized enterprises (SMEs)? If so, what topics do you address in these projects?

Prof. Schilp: Yes, we often work with SMEs. The companies involved in our projects come from very different industries, for example, from the consumer goods sector, the construction industry and plant and mechanical engineering. The aim of the corresponding projects is often to lay the foundations for using artificial intelligence at the participating companies. This does not mean that all companies are not fit enough when it comes to digitization . But a certain basic understanding of the subject, coupled with the right infrastructure, are basic prerequisites for the successful use of AI and the implementation of other digitization aspects.

To make themselves fit for the use of AI and other digitization aspects, companies often also turn to the Chair of Production Informatics or Fraunhofer IGCV for training. Here, employees can be familiarized with digitized processes in learning factories, i.e. exemplary production environments, and thus also taught a new understanding of their roles within the value chain.

Thank you for the exciting conversation on a very promising topic!