Modeling Actual Feedrate Delay Based on Automatic Toolpaths Segmentation Approach Using Machine Learning Methods in Ball Burnishing Operations of Planar Surfaces

This paper presents a novel approach using machine learning methods for the automated segmentation of acceleration signals measured during ball burnishing (BB) operations performed on a computer numerical controlled (CNC) milling machine. The study addresses the challenge of accurately finding actual feedrates in that finishing operation, which often deviate from programmed values due to various dynamic reasons. The method involves a two-stage process: first, an automatic signal segmentation algorithm employing Gaussian Mixture Modeling (GMM) and K-means clustering is applied to the ball burnishing (BB) process and acceleration data. Second, a Taguchi L9 experimental design is used to assess the influence of some regime parameters on the actual feedrate and the BB’s cycle duration. Results show successful segmentation of the toolpaths based on X-axis accelerations and deforming force data, with the Calinski–Harabasz Index confirming good cluster separability. Programmed feedrate and the number of toolpath points were identified as the most significant factors affecting the percentage delay between programmed and obtained feedrates. The main contribution is the development and testing of a new method for segmenting different toolpath states in ball burnishing operations, based on measured accelerations and momentary deforming force magnitudes. The present work offers valuable insights into autonomous monitoring and control in BB operations.