Machine learning-based deterministic plasma jet machining of optical surfaces - PlasmaJetControl

• Applicants: Professor Dr. Thomas Arnold; Dr. Martin Rudolph
• Subject Area: Coating and Surface Technology ; Synthesis and Properties of Functional Materials ; Metal-Cutting and Abrasive Manufacturing Engineering
• Term: since 2026
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 577831116

The ultra-precise form generation and correction of optical surfaces such as lenses or mirrors by means of locally acting fluorine-containing atmospheric pressure plasma jets is based on a dry etching process that relies on chemical reactions between the reactive species formed in the plasma (fluorine radicals) and the atoms of the substrate surface, resulting in the formation of volatile products for silicon compounds such as Si, SiO2, SiC. However, this technology has severe limitations with regard to efficient use in ultraprecision machining due to the thermodynamically determined etching mechanisms, which depend on the surface temperature of the substrate, and the associated and not yet fully understood complexity of the underlying chemical removal mechanism. These disadvantages arise mainly from the non-linear dependence of the etch rate on the spatial and temporal surface temperature distribution, since the plasma jet is not only a source of reactive species but also a local heat source. This results in a lack of process stability and insufficient process convergence, which, when using plasma jet technology for ultra-precise figuring and figure error correction of optical surfaces, leads to a high number of iterations of process steps and thus causes a high consumption of environmentally harmful, chemically aggressive process media and energy. The aim of the project is therefore to gain a deeper understanding of the plasma jet processing procedure and, on this basis, to develop new simulative and experimental methods to significantly increase process convergence. By investigating the temperature-etch rate relationship, a data basis for the parameterization of FEM models for process simulation will initially be created. The developed FEM models, which are then to be validated experimentally, will be used to generate process data that will serve as training data for neural networks. With the help of these networks, an adaptive machine-learning-based process control system is to be created that compensates for the nonlinearities of the interaction of the plasma jet with the surface, thus enabling stable deterministic process control for ultra-precision machining.

• DFG Programme: Research Grants