Fraunhofer Institute for Production Technology IPTResearch project "KI4ToolPath"
Artificial intelligence and machine learning for tool path optimization
Additive manufacturing is becoming increasingly attractive for the production of a wide variety of components, especially for single and small batch production of individualized and geometrically complex products. In many cases, not only are additive manufacturing processes significantly more efficient, but they save more resources than conventional, subtractive processes. In addition, they are more flexible from a production planning perspective and offer much greater scope for designing components, for example with hollow or lightweight structures.
The advantages of additive manufacturing are offset by the fact that the processes often do not run stably: Fluctuations occur in the process, which impair product quality. Another disadvantage is that, since they lack digital tools, machine operators have to carry out process and path planning manually, which is time-consuming.
Thus, for additive manufacturing to become a real alternative to conventional processes, shorter process planning phases and consistently high product quality are required. By using extensive process data through machine learning, research could create digital path planning for additive manufacturing processes that automatically compensates for process variations.
New path planning concept shortens manufacturing process by around 20 percent
In the research project “KI4ToolPath – Geometry-supported classification of process states for path planning support using the example of the WAAM process”, a fully digital process planning for Wire Arc Additive Manufacturing (WAAM) was developed. Automated process planning improves the reliability and quality of the WAAM process: the throughput time of the manufacturing process is reduced, the scrap rate drops, as does the number of geometric deviations and the associated inspection and rework effort.
The main focus of the Fraunhofer IPT's work was on the implementation and investigation of Wire Arc Additive Manufacturing (WAAM), process data acquisition, process data preparation and validation. .
Sensor data for artificial intelligence training
In the first stage of the project, the researchers integrated various sensors into the WAAM system. During production, the sensors recorded selected process data, such as the geometry of the weld beads or a temperature profile. This data was then formatted for machine learning (ML) and thus made usable for an artificial intelligence (AI).
To generate sufficient data volume for training an AI system, several demonstrators of varying complexity were defined and constructed by the Fraunhofer IPT at the start of the project and built multiple times in the WAAM plant. In doing so, the limits of the process were also pushed and errors provoked. This allowed process instabilities and interruptions to be recorded, which can make an important contribution when training an AI. Tests of various process and control strategies were used to generate a data field that was as broad as possible.
Since the AI system in the project was to access the largest possible data pool in order to learn efficiently, the project partners not only generated real process data but also used data generated in simulations for training. To do this, the researchers incorporated 3D files from publicly accessible databases over the course of the project.
New interface for transfer between sensors and AI
For the transfer between sensor technology and AI, the researchers designed a suitable interface with an appropriate data structure. To make the data usable, the project partners also considered different options for neural networks, such as auto-encoders for dimension reduction and further processing of the various data formats, transformers or convolutional neural networks. These investigations were important steps in identifying an efficient way to train an AI.
Data-based development of manufacturing strategies
To apply the AI algorithms to the WAAM process, an additional interface for path planning was implemented. In doing so, the researchers laid the foundation for an AI-supported web planning software that is based on large amounts of data from the WAAM process. It develops web strategies that also take into account unwanted process fluctuations. A particular challenge in web planning is the highly fluctuating component temperature, which directly influences the process result. With a digital web planning system, such factors can be taken into account even before production.
Project partners
- Collective mind AG, Leonberg (Coordination)
- Fraunhofer Institute for Production Technology IPT, Aachen
- ModuleWorks GmbH, Aachen
Funding
The research project "KI4ToolPath – Geometry-supported classification of process states for path planning support using the WAAM process" was funded by the German Federal Ministry of Education and Research (BMBF) as part of the research project "KI4ToolPath." Funding code: 01IS22021C
Project Management Agency
DLR Project Management Agency – German Aerospace Center e.V.