Physics-informed Bayesian inference for milling stability analysis

Abstract

The accurate prediction of chatter stability in milling operations is a continuous pursuit of manufacturing engineering. Physical-based stability analysis methods suffer from inaccurate model parameters, while data-driven methods lack generalisability and physical interpretability. This study proposes a physics-informed Bayesian inference framework for milling stability analysis. The framework leverages experimental data to infer the distribution of model parameters for probabilistic stability lobe diagram (SLD) computation, thus maintaining the generalisability and interpretability of the physical model in a hybrid-driven manner. By defining a novel likelihood function based on the Floquet theory, the underlying connection between the model parameters and experimental cutting data is established. The uncertainties and variations of the model parameters can then be represented by the inferred probability density function, which can be used to generate a reliable probabilistic SLD. The experiments indicate that the proposed method has significant potential for improving the accuracy of milling stability prediction.