Autonomous and data-efficient optimization of turning processes using expert knowledge and transfer learning
Markus Maier, Hannes Kunstmann, Ruben Zwicker, Alisa Rupenyan, Konrad Wegener
Abstract
Process parameters in machining are predominantly selected by following manual tuning procedures. Using data from the system and dedicated performance indicators combined with learning-based approaches enables automating these procedures while reducing the costs of the machining process. This study investigates efficient data-driven approaches for autonomous parameter selection in turning. The number of experimental trials for finding optimal process parameters is reduced by incorporating expert knowledge and transferring knowledge between different tasks. The turning process costs are modeled using Gaussian process models, and the selection of informative experiments is achieved by Bayesian optimization. In this study, all tested methods using expert knowledge or transfer of knowledge reduced the number of experiments by at least 40% compared to a standard approach for parameter selection based on Bayesian optimization without expert knowledge, confirming the efficiency of the applied methods.