Speaker
Description
Designing stable milling operations is crucial to ensure a high surface quality of the machined workpieces and reduce rejects during production. Stability lobe diagrams can be used to identify stable conditions. Analytical approaches or simulation techniques can be used to reduce the experimental effort for stability evaluation for different process parameter values. However, complex cause-effect relationships, such as concept drift due to tool wear, and simplifying assumptions required to ensure sufficient simulation run-times limit their applicability. In contrast, machine learning models can provide real-time predictions of process characteristics based on different process parameter configurations with high accuracy and generalization. However, a large number of experiments are still necessary to establish the required database. In this context, labeling the data, i.e., evaluating the resulting stability for each parameter configuration, can be time-consuming. To this end, sensor data collected during the process can be used to automate the stability assessment. However, sensors that can be efficiently integrated into the working area without interfering with the experimental setup, e.g., acoustic emission sensors, are often susceptible to noise, which makes algorithmic analysis challenging. In this paper, a framework for automatic evaluation of milling stability based on statistical tests using time series data acquired by acoustic emission sensors is presented. The proposed framework also considers model uncertainty and has been validated on controlled synthetic and noisy real data sets. In addition, insightful analyses of the model hyperparameters are given for efficient model performance.
