Dreaming neural networks for adaptive polishing

Kategorien Konferenz (reviewed)
Jahr 2020
Autorinnen/Autoren Dittrich. M.-A., Rosenhahn, B., Magnor, M., Denkena, B., Malek, T., Munderloh, M., Kassubeck, M.:
Veröffentlicht in Proceedings of the 20th international conference of the european society for precision engineering and nanotechnology (euspen's), June 8th - 12th June 2020, Virtual Conference, Geneva, S. 263-266.

Mechanical polishing is an important step within the process chain of manufacturing workpieces with high requirements regarding the surface quality, e.g. for optical components. The resulting surface quality depends on several parameters, e.g. the process parameters, the workpiece material, the initial surface roughness and the tool condition. Thus, finding process parameters that lead to the desired surface roughness can be regarded as a complex optimization problem. For this purpose, an artificial neural network (ANN) has been designed and trained with data from polishing experiments. Using a dreaming network approach, the ANN has been enabled to suggest appropriate process parameters under consideration of the initial roughness of the workpiece and the tool condition. The validation experiments showed that the process parameters suggested by the neural network led in 72% to the target roughness within a standard deviation.