🔬 applied physics

Effects of Tool Wear on the Surface Texture in Turning: A Feature Characterization Approach Based on ISO 21920-2

This study demonstrates that applying ISO 21920-2 feature characterization to turning surface profiles, specifically using the groove-level mean maximum absolute gradient ( $\overline{R_\mathrm{dt}}_\mathrm{groove}$ ), enables robust and physically interpretable estimation of tool wear by isolating deterministic wear-induced changes from stochastic surface noise.

Original authors: Alexander Müller, Maximilian Berndt, Hagen Schmidt, Lars Müller, Matthias Eifler, Eberhard Kerscher, Benjamin Kirsch, Jörg Seewig

Published 2026-08-13

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Alexander Müller, Maximilian Berndt, Hagen Schmidt, Lars Müller, Matthias Eifler, Eberhard Kerscher, Benjamin Kirsch, Jörg Seewig

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are a detective trying to solve a mystery, but the culprit never leaves a fingerprint. Instead, they leave behind a tiny, invisible scratch on a piece of metal every time they walk by. In the world of manufacturing, this "culprit" is a cutting tool, and the "scratch" is the surface texture of the part it just made. For decades, engineers have tried to guess how worn out a tool is by looking at the roughness of the metal it cut. They used tools like a magnifying glass that only gave them a single, blurry number—like saying a surface is "rough" or "smooth." But this is like trying to guess the shape of a mountain just by knowing its average height; you miss the peaks, the valleys, and the specific shape that tells you exactly what happened.

This paper dives into a smarter way to read these scratches. It uses a new, standardized rulebook (called ISO 21920-2) that treats the surface not as a blurry cloud, but as a collection of distinct, individual "features"—like the tiny, repeating grooves left by the tool's edge. Think of it as switching from looking at a blurry photo of a forest to counting and measuring every single tree. The researchers wanted to know: Can we look at these specific grooves to tell exactly how much the tool is worn out, and can we even figure out where on the tool the damage is happening? This matters because if a factory knows a tool is about to break, it can swap it before it ruins a part, saving time, money, and preventing disasters.

The Story of the Worn-Out Tool

In this study, the researchers acted like forensic scientists for metal-cutting tools. They took twelve brand-new cutting tools (made of super-hard tungsten carbide with a shiny aluminum titanium nitride coating) and used them to turn a steel rod, just like a potter shaping clay on a wheel. They didn't stop after one cut; they kept going until the tools were completely worn out, checking them at nine different stages of their "life." At each stage, they measured the tiny grooves left on the steel rod and compared them to the actual damage on the tool itself.

The Big Discovery: One Number to Rule Them All
The team found that the old way of measuring—looking at the whole surface and giving it one average score—wasn't very good at spotting the specific damage. It was like trying to hear a single violin in a full orchestra by just listening to the volume of the whole room.

Instead, they used a clever digital trick called "watershed segmentation." Imagine the surface grooves as a landscape of hills and valleys. This trick acts like a smart flood, filling up the valleys until the water level hits a specific height, effectively separating each individual groove from its neighbors. Once they isolated these grooves, they realized something amazing: the wear on the tool didn't change the whole surface randomly. It changed one very specific part of the groove.

They discovered that as the tool got worn, the "trailing flank" of the groove (the back wall of the little trench the tool digs) got steeper and steeper. It's as if the tool's edge got a tiny, sharp notch, and that notch was being stamped perfectly onto the metal every time it cut. By measuring just the steepness of this specific back wall, they found a single number (called $R_{dtgroove}$ ) that was a superstar detective. This one number explained between 83% and 90% of the changes in how worn the tool was. That is a massive amount of clarity from just one measurement!

What They Ruled Out
The researchers were very careful not to get fooled. They tested over 920,000 different combinations of measurements and complex computer models to see if they could find a "super-recipe" that was better than their simple steepness number. They tried mixing and matching dozens of different math formulas and even used fancy machine learning algorithms (like XGBoost) to find hidden patterns.

The result? Nothing beat the simple steepness number. The complex models didn't give any better answers. The paper explicitly argues against the idea that you need a complicated, multi-parameter system to predict tool wear in this specific setup. The "noise" or random bumps in the metal (the stochastic part) didn't matter much; the real story was entirely in the shape of the groove (the deterministic part).

How Sure Are They?
The team is very confident in their findings, but they are also honest about the limits. They proved their theory by taking the "average groove" they calculated from the metal and comparing it directly to a high-tech 3D photo of the actual tool edge taken with a confocal microscope. The shapes matched almost perfectly, with a tiny error of only 0.50 micrometers (that's about 1/200th the width of a human hair). This confirmed that the "steepness" they saw on the metal was indeed caused by a physical "notch wear" on the tool's edge.

However, they are careful to say this specific "steepness" rule applies to the exact type of steel and tools they used (AISI 1045 steel and AlTiN-coated inserts). If you switch to a different metal, like titanium, or a different tool shape, the "fingerprint" might look different. But for the setup they tested, the evidence is solid: the shape of the groove tells the whole story of the tool's health.

The Takeaway
This paper shows that you don't need a super-complex computer to know if a tool is dying. You just need to look closely at the tiny, repeating grooves it leaves behind and measure how steep the back wall of those grooves has become. It turns a messy, confusing surface into a clear, readable map of the tool's life. This discovery could help factories build better sensors that watch the tool in real-time, swapping it out the exact moment it starts to get a little notch, keeping production smooth and parts perfect.

Technical Summary: Effects of Tool Wear on the Surface Texture in Turning: A Feature Characterization Approach Based on ISO 21920-2

Problem Statement
Tool wear is a critical determinant of process stability, workpiece quality, and cost-efficiency in machining operations like turning. While the surface texture of a turned component acts as a "fingerprint" of the cutting edge geometry and wear state, traditional global surface parameters (e.g., $R_a$ , $R_q$ ) are insufficient for diagnosing wear. These field parameters average the entire profile, obscuring spatially resolved, process-induced deterministic structures (feed grooves) and failing to distinguish between wear-induced geometric changes and stochastic microstructural deviations. Consequently, empirical findings regarding the correlation between roughness parameters and wear are often contradictory or non-transferable across different setups.

Methodology
The study utilizes a database of roughness profiles from 12 AlTiN-coated tungsten carbide inserts (CNMG120408) machining AISI 1045 steel. Measurements were taken across nine wear states throughout the tool life, with three replicate profiles per state. Wear was quantified via crater wear area ( $A_C$ ) and mean flank wear width ( $V_B$ ), alongside cutting time ( $t_c$ ).

The methodology employs a hierarchical analysis based on ISO 21920-2 feature characterization:

Standard Correlation: Initial screening of standard ISO 21920-2 field and feature parameters against wear indicators.
Watershed Segmentation: Adaptation of the ISO 21920-2 watershed segmentation to isolate individual feed grooves. A width-based pruning threshold (0.15 mm) and significance filter (0.19–0.21 mm) were established to robustly extract tool grooves corresponding to the 0.2 mm feed rate.
Mean-Feature Decomposition: A novel approach decomposes the surface profile into a deterministic component (reconstructed from the mean of normalized, segmented grooves) and a stochastic component (the residual). This separates the geometric imprint of the cutting edge from random surface noise.
Validation: Confocal microscopy of the cutting edge was used to validate that the reconstructed deterministic profile accurately images the engaged cutting edge geometry.
Exhaustive Evaluation: An exhaustive search of over 920,000 feature characterization combinations and custom descriptors was conducted, alongside multivariate modeling (OLS, PLS, XGBoost) using Leave-One-Tool-Out (LOTO) cross-validation to prevent data leakage.

Key Results

Deterministic vs. Stochastic: Wear-induced changes are almost entirely carried by the deterministic component of the surface profile. The stochastic component shows weak correlation with wear ( $r \leq 0.41$ ).
Localization of Wear: Within the deterministic component, the wear-induced changes are localized almost exclusively to the trailing flank of the cutting groove (the last 0.03 mm of the feature). This region exhibits a steepening trend as wear progresses.
Physical Mechanism: The steepening of the trailing flank is directly attributed to notch wear on the secondary cutting edge. Confocal measurements confirm that the mean feature reconstructs the engaged cutting edge geometry, validating the "profile sharpening effect" where notch wear transforms blunt profile tips into sharp trailing flanks.
Superior Parameter: The groove-level mean maximum absolute gradient ( $R_{dt}^{groove}$ ) was identified as the single most informative parameter. It achieves Pearson correlation coefficients of $r = 0.91$ ( $A_C$ ), $0.92 $($ V_B$), and $0.95 $($ t_c$).
Model Performance: In LOTO cross-validation, $R_{dt}^{groove}$ alone explains 83–90% of the variance ( $R^2_{LOTO} = 0.82–0.89$ ) of the wear indicators. Exhaustive searches for other single parameters and multivariate models (including nonlinear approaches) yielded no statistically significant improvement over the simple linear relationship with $R_{dt}^{groove}$ .

Significance and Claims
The paper claims that the standardized feature characterization of ISO 21920-2, when applied at the groove level, provides a physically interpretable and robust method for wear estimation. By isolating the deterministic groove geometry, the approach overcomes the limitations of global roughness parameters. The primary contribution is the demonstration that a single, physically motivated parameter ( $R_{dt}^{groove}$ ) is sufficient for robust wear estimation, rendering complex multivariate models unnecessary for this specific machining configuration.

The authors emphasize that these findings are specific to the investigated combination of tool material, workpiece, and cutting parameters. While the method successfully links surface texture to cutting edge geometry and identifies notch wear as the dominant mechanism in this context, the transferability to other material systems (e.g., titanium or nickel alloys) where different wear mechanisms dominate requires further investigation. The work establishes a spatially resolved database intended to validate grey-box models and serves as a foundation for future inline monitoring strategies using scattered light sensors.

The Story of the Worn-Out Tool

More like this