Assessing sine and Gaussian fitting for surface feature extraction in real-time wear monitoring of turning operations

Abstract

Due to the challenges in tool condition monitoring, sensor fusion systems and the search for supportive features have gained increasing attention. With advancements in image acquisition and processing, surface image features offer a low-cost, supplementary data source for monitoring applications. This study introduces two novel image features derived from the fitting error metrics of sine and Gaussian functions, applied to column projections of surface images. Unlike conventional texture analysis methods, this approach specifically targets the wave-like patterns characteristic of turned surfaces, enabling localized assessment of surface anomalies. Experiments were conducted with varying cutting speeds, feed rates, and depths of cut to investigate the correlation between the proposed features and tool wear. Static analysis yielded adjusted $R^2$ values of 0.497 for sine fitting and 0.579 for Gaussian fitting, while dynamic analysis demonstrated higher correlations with adj. $R^2$ values of 0.8007 and 0.8197, respectively. Additionally, a cropping analysis was implemented to address potential image acquisition challenges in real-world applications, such as optical distortions and debris interference. Results indicated that focusing on central regions improved static fitting accuracy by up to 28% and dynamic fitting accuracy by up to 16%, underscoring the robustness and practical applicability of the proposed features for localized wear analysis.