A Methodology for Acceleration Signals Segmentation During Forming Regular Reliefs Patterns on Planar Surfaces by Ball Burnishing Operation

In the present study, an approach for determining the different states of ball burnishing (BB) operations aimed at forming regular reliefs’ patterns on planar surfaces is introduced. The methodology involves acquiring multi-axis accelerometer data from CNC-driven milling machine to capture the dynamics of the BB tool and workpiece, mounted on the machine table. Following data acquisition from an AISI 304 stainless steel workpiece, which is subjected to BB treatments at different toolpaths and feed rates, the recorded signals are preprocessed through noise reduction techniques, DC component removal, and outlier correction. The refined data are then transformed using a root mean square (RMS) operation to simplify further analysis. A Gaussian Mixture Model (GMM) is subsequently employed to decompose the compressed RMS signal into distinct components corresponding to various operational states during BB. The experimental trials at feed rates of 500 and 1000 mm/min reveal that increased feed rates enhance the distinguishability of these states, thus leading to an augmented number of statistically significant components. The results obtained from the proposed GMM based algorithm applied on compressed RMS accelerations signals is compared with two other methods, i.e., Short-Time Fourier Transforms and Continuous Wavelet Transform. The results from the comparison show that the proposed GMM method has the advantage of segmenting three to five different states of the BB-process from nonstationary accelerations signals measured, while the other tested methods are capable only to distinguish the state of work of the deforming tool and state of its rapid (re-)positioning between the areas of working, when there is no contact between the BB-tool and workpiece.