Speaker
Description
Accurate predictions of process characteristics in milling, such as tool vibrations, allow for identifying and avoiding unstable cutting conditions that can lead to excessive tool wear, surface defects or tool breakage. Therefore, considering vibrations during process design is essential to ensure dimensional accuracy, surface integrity and the longevity of cutting tools. Common methods, including analytical and simulation-based approaches, often require simplifications and assumptions that limit the accuracy, applicability and scalability in real-world scenarios. In contrast, data-based modeling strategies provide predictions of specific process characteristics with high generalization capabilities. Advancements in engineering technology allow the machining of complex workpiece designs in a single setup, resulting in highly complex milling paths with changing engagement conditions. Hence, for such processes, the demands on the required training data are increased and the resulting prediction accuracy of data-based models can vary significantly for different segments of the milling path. To mitigate this issue, unsupervised learning strategies capable of identifying similar patterns in datasets can be used for manufacturing applications to group machining operations into elementary process sections. The research presented in this paper expands on this idea by training data-based models for each identified section to predict relevant process characteristics, such as tool vibrations, using acquired measurement data and tool path information as features. Thus, a milling process with highly variant engagement conditions was represented by multiple cluster-specific models, leading to an increased prediction accuracy of process characteristics compared to using only a single model.
