In the "ProBayes" research project, the SKZ and the Fraunhofer IPA have jointly developed a process monitoring and control system for injection molding based on Bayesian networks. Using a fully networked injection molding cell in the SKZ's technical center, the researchers are demonstrating a system in live operation that independently detects deviations in product quality, identifies the most likely cause, and provides the machine operator with specific recommendations for action.
After two years of joint research, the SKZ Plastics Center in Würzburg and the Fraunhofer IPA in Stuttgart are pleased to announce the successful completion of a development project funded by the German Federal Ministry of Economics and Technology. The project developed Bayesian networks - statistical methods for modeling processes in which quality-relevant process parameters and their dependencies are connected by nodes and edges. Conditional probability distributions are assigned to nodes so that dependent process variables can be accurately predicted.
Direct Feedback on Part Quality
The power of the method was demonstrated on a fully networked injection molding cell at SKZ. This included measurement systems, peripherals and the injection molding machine itself, which were networked using OPC UA and MQTT. Thanks to sophisticated inline measurement technology, direct feedback on part quality can be recorded and stored for each cycle. As a side effect, the companies involved in the project were able to benefit greatly from the knowledge gained in the area of interfaces and networking.
Validated in live operation
The developed Bayesian network models the dependencies of the part weight on the setting and process variables of the injection molding machine and enables a root cause analysis if the part weight deviates from the target value. As a result, the performance of the Bayesian network trained on real process data was validated on the injection molding cell in live operation and demonstrated to interested companies. The Bayesian network was not only able to detect when the part weight was outside the adjustable tolerance. It was also able to identify the most likely cause and provide the machine operator with targeted recommendations for action.
"Economically highly relevant results"
"The results are economically highly relevant - that's why I'm more than satisfied," says Jonathan Lambers, project manager and senior scientist at SKZ. "I would like to thank everyone involved for the fantastic collaboration - especially KraussMaffei Technologies GmbH for providing the injection molding machine and support with data collection, Erium GmbH for their support with Bayesian networks, and codecentric AG for their help with data science issues." Christoph Kugler, Group Leader Digitalization at SKZ, is also pleased with the results: "We were able to create immensely valuable groundwork for the digitalization of injection molding production in general and lay the foundation for the profitable integration of the generated results into production environments for companies."
Interested companies can view the project results live at the SKZ and have their suitability for their own processes evaluated.
After two years of joint research, the SKZ Plastics Center in Würzburg and the Fraunhofer IPA in Stuttgart are pleased to announce the successful completion of a development project funded by the German Federal Ministry of Economics and Technology. The project developed Bayesian networks - statistical methods for modeling processes in which quality-relevant process parameters and their dependencies are connected by nodes and edges. Conditional probability distributions are assigned to nodes so that dependent process variables can be accurately predicted.
Direct Feedback on Part Quality
The power of the method was demonstrated on a fully networked injection molding cell at SKZ. This included measurement systems, peripherals and the injection molding machine itself, which were networked using OPC UA and MQTT. Thanks to sophisticated inline measurement technology, direct feedback on part quality can be recorded and stored for each cycle. As a side effect, the companies involved in the project were able to benefit greatly from the knowledge gained in the area of interfaces and networking.
Validated in live operation
The developed Bayesian network models the dependencies of the part weight on the setting and process variables of the injection molding machine and enables a root cause analysis if the part weight deviates from the target value. As a result, the performance of the Bayesian network trained on real process data was validated on the injection molding cell in live operation and demonstrated to interested companies. The Bayesian network was not only able to detect when the part weight was outside the adjustable tolerance. It was also able to identify the most likely cause and provide the machine operator with targeted recommendations for action.
"Economically highly relevant results"
"The results are economically highly relevant - that's why I'm more than satisfied," says Jonathan Lambers, project manager and senior scientist at SKZ. "I would like to thank everyone involved for the fantastic collaboration - especially KraussMaffei Technologies GmbH for providing the injection molding machine and support with data collection, Erium GmbH for their support with Bayesian networks, and codecentric AG for their help with data science issues." Christoph Kugler, Group Leader Digitalization at SKZ, is also pleased with the results: "We were able to create immensely valuable groundwork for the digitalization of injection molding production in general and lay the foundation for the profitable integration of the generated results into production environments for companies."
Interested companies can view the project results live at the SKZ and have their suitability for their own processes evaluated.
