Application of machine learning for improved surface quality classification in ultra-precision machining of germanium

Abstract

Ultra-precision machining is the primary process for producing optics for infrared systems including those for thermal imaging, night vision, and other applications. Materials used in infrared optics are typically brittle and include germanium, zinc selenide, zinc sulfide and chalcogenide glass. It has been shown that under certain conditions, these materials can be machined so that they deform by shear and not fracture, and the surface produced is suitable for optical applications. However, the physical mechanisms that govern the material behavior and the surface generation are complex and not well understood, hence leading manufacturers to choose conservative machining parameters to ensure surface and subsurface quality. These parameters often compromise productivity (material removal rate and consequently manufacturing time). In this work, we apply machine learning to the ultra-precision diamond machining of single-crystal germanium to develop a model that provides manufacturers with a go/no-go answer for a given set of parameters. Further, we demonstrate that by also including surface metrology parameters in the model, improvements in fracture prediction can be achieved. With this information, manufacturers can choose less conservative and hence more productive parameters that reduce machining time and cost without degradation of optical performance. This type of modeling is expected to be of even greater value as optical surfaces and the associated manufacturing processes become more complex, such as with freeform optics.