The project team members want to open up established monitoring approaches from large-scale production by using cross-process learning for monitoring small batch sizes and enable the detection of complex fault patterns with the help of AI methods.
In order to ensure that high-wage countries like Germany remain attractive as production locations and that the challenging transformation towards a sustainable economy succeeds without a loss of prosperity, it is necessary to realise an economic and at the same time resource-saving production. To achieve this, production plants must be efficient and highly automated. An important role is played here by process monitoring systems that detect unacceptable deviations in production, so-called anomalies, at an early stage and automatically. This enables timely intervention in the process: Production downtimes, costly consequential damage and rejects are avoided. Monitoring systems therefore ensure process reliability and are indispensable for fully automated production.
The performance of current monitoring systems is process-specific and is determined by the set monitoring limits. For the best possible performance, optimal limits must first be found. This leads to a high parameterisation workload and requires corresponding know-how. Especially for medium-sized companies with changing processes, the parameterisation effort is usually not economical. AI-based methods provide a solution. Scientist Krüger: "If sufficiently large data sets are available, it is possible with AI-based methods to realise a self-parameterising system with high detection rates and low false alarm rates at the same time."
For small batch sizes, the use of AI-based methods for monitoring has not been useful so far, as there is only little data available that is needed as a reference. For small and medium-sized enterprises (SMEs), which often produce smaller batch sizes, these AI-based methods are thus not applicable. As a result, the achievable level of automation is limited for SMEs, while potential consequential damage and higher scrap lead to additional economic risks.
Transfer learning is the approach that the scientist and his industrial partner are taking in the project to make AI-based monitoring methods applicable to small batches: Knowledge already gained from similar processes is used in an abstracted form to generate the missing reference processes. Krüger: "In this way, AI models will also be applicable in small batch and single-part production."
In the RoPro research project, the cooperation partners IFW and UTTec are developing a robust and autonomously parameterising inspection system with diagnostic function for the machining of small series. The envisaged overall system consists of the three sub-modules "autonomous signal segmentation", "real-time capable anomaly detection" and "fault cause diagnosis".
For segmentation, a method is being developed to break down signal sequences completely autonomously into individual recurring standard elements, such as grooves or pockets. In this way, piece-by-piece comparability is achieved even for highly individual processes. The module is first used to build up a database containing signal sequences recorded at different process parameters. This provides similar sections, which are then used to generate the required reference curves and on the basis of which a knowledge transfer can take place by means of transfer learning.
"With the help of further AI-based methods, we are implementing real-time anomaly detection, which detects unacceptable deviations during the production process," explains scientist Krüger. By diagnosing the cause of the error, machine operators can in turn quickly identify and eliminate the cause of the error following the occurrence of an anomaly. This reduces downtimes and achieves higher availability. The increased process reliability also contributes to the avoidance of rejects, as damage to the workpieces due to process errors is prevented. This leads to a higher quality rate while at the same time using fewer resources. Krüger: "We thus achieve a significant increase in overall equipment effectiveness for small batch production."
This Project is supported by the Federal Ministry for Economic Affairs and Climate Action (BMWK) on the basis of a decision by the German Bundestag. The IFW and the UTTec GmbH & Co. KG are grateful for the financial support of the project.
Contact:
For further information, please contact Maximilian Krüger, Institute of Production Engineering and Machine Tools at Leibniz Universität Hannover, via telephone +49 511 762 18068 or by e-mail (krueger@ifw.uni-hannover.de).
