Quantification of micro-pullwinding process as basis of data mining algorithms for predictive process model
Cost efficient customization of goods is a very important field of interest in many areas of the manufacturing sector today. New research tries to extend the concept of customized products to the field of medical devices. In particular, the focus is set on the production of minimally invasive disposables. In this context, fiber reinforced plastics (FRP) do not only provide compatibility with all relevant medical imaging technologies but can also be used to realize mechanical customization. Currently, the physician has to choose from the available product range predefined by the manufacturers. However, to maximize the ease of use of the medical devices and, thus, to optimize the outcome of the intervention a customized product would be the perfect solution. The involved devices are disposable items and todays production technology is optimized to produce high amounts of these devices with low variability at low costs. Also an individualized version of these devices would be a disposable. Therefore a still cost efficient and at the same time adaptive production system would be needed to manufacture such individualized disposables. One solution to overcome this challenge would be to link the necessary adaptive production systems in a continuous production line. On the one hand, the product can be produced cost efficiently in an endless process with minimum handling operations, on the other hand the adaptiveness of the involved production systems allows for customized manufacturing. However, the complexity of such combined continuous production system calls for new ways to control and optimize the process using data mining technologies. In this context, this work presents the initial steps to create a model of a micro-pullwinding process. The micro-pullwinding process as part of the linked production system will be used to produce FRP wires with customized mechanical properties. A process analysis reveals the crucial process parameters and their relationships, which are investigated in more detail afterwards. The results of this quantifying investigation can be used to train a process model based on self-learning algorithms.