Akademik Veri Yönetim Sistemi
9TH INTERNATIONAL ANTALYA SCIENTIFIC RESEARCH AND INNOVATIVE STUDIES CONGRESS, Antalya, Türkiye, 20 Kasım - 23 Aralık 2025, ss.600-609, (Tam Metin Bildiri)
Introduction and Purpose: Real-time monitoring of tool wear in turning is important for process reliability and economic efficiency. Turned-surface images are known to reflect wear, and various features extracted from these images have been used in the literature for real-time monitoring. In most studies, features have been computed from the entire image acquired after machining; however, in practical settings, coolant and chips may prevent uninterrupted fullsurface imaging. Therefore, it is necessary to evaluate features performance across different resolution regions. This study investigates how the grayscale mean (G) feature previously used on full images reflects the wear trend when vertical resolution is reduced.
Materials and Methods: Turning experiments were conducted on AISI 4140 at five different cutting speeds, and surface images were acquired after each pass. Vertical resolutions were progressively narrowed starting from the full image, and G was recomputed for each resolution. Fit was assessed experiment-base (for each speed setting) by relating G–time trends to wear progression; in addition, the dynamic relationship was examined by comparing the logarithmic wear trend observed with increasing speed to the dynamic behavior of G. And Adj. R² values was reported.
Results: Across the five cutting speed settings, experiment-base consistency was evaluated. As vertical resolution narrowed, the relationship between G and the wear trend generally weakened both experiment-base and dynamically; however, the decrease was not strictly linear, and local maxima/minima were observed at mid-range coverage levels, depending on illumination and acquisition conditions.
Discussion and Conclusion: When considered in terms of the G feature, vertical resolution should be treated as a tunable analysis parameter: very narrow resolutions may yield weaker fits, whereas adequate coverage can lead to more stable trend estimation. Given potential partial occlusions in real applications, reporting feature behaviors across multiple resolution levels is recommended.