The Rauschert Group, a manufacturer of technical ceramics, moulded plastic parts, components and assemblies, is the global leader in the development and production of ignition systems for heating systems, such as ignition and ionization electrodes for gas and oil burners. In the production of ceramic bushings, a metal retaining plate is joined to a ceramic bushing using a press. A considerable pressing force is required to guarantee the load-bearing capacity of this connection. However, the ceramic bushing must not be damaged in the process. Although Rauschert already has an excellent understanding of the pressing process, test series are often required to optimize the process. These are costly, as damage to the ceramic component can usually only be detected with the aid of an electronic test method. There are no non-destructive methods to test the bond strength.

In order to gain a deeper insight into the pressing process, Rauschert carried out several test series using an electric press, which measures, in high resolution, process variables such as the progression of the pressing force over time. The aim is to minimize scrap and guarantee a tight joint between the ceramic and the retaining plate. However, test series are costly and time-intensive, and so machine learning has been used to predict and assess the pressing process.

 

Approach

As training labels are given to all models of this project through the measured data, all models were trained under supervision. To check which parameters influence the process, support vector machines and neural networks were trained to distinguish process parameters based on process variables. Additional neural networks were trained to predict and assess the process variables.

Partner