⚡ electrical engineering

A Method for Real-Time Milling Monitoring Based on Tool Engagement Analysis in Hybrid Micro-Manufacturing

This paper presents a computationally simple, calibration-free method for real-time monitoring of tool engagement and crack detection in hybrid micro-manufacturing, addressing the challenges of unstable cutting conditions caused by poor surface integrity from additive processes.

Original authors: Marcin Gołaszewski

Published 2026-08-07

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Marcin Gołaszewski

Original paper licensed under CC BY 4.0 (https://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 a factory where robots build things in two very different ways. First, they use a "3D printer" style method, stacking up layers of hot metal like a very precise, high-tech cake. This is great for making weird shapes and hollow structures, but the surface of this "cake" is often rough, bumpy, and sometimes even cracked. Then, to make the part smooth and perfect, a second robot comes in with a spinning metal cutter to shave off the rough bits. This is called "hybrid manufacturing."

The problem is that the second robot is flying blind. Because the first robot left a bumpy, cracked surface, the second robot doesn't know exactly how deep to dig. It might hit a hard spot, then suddenly fly over a crack, or get confused by a bump. If it digs too deep, it breaks; if it doesn't dig deep enough, the part is ruined. Engineers need a way to tell the cutter exactly what it is touching right now, instantly, so it can adjust its path before things go wrong. This paper tackles that exact challenge: how to give the cutting tool "super-senses" to feel the surface as it works.

The "Silent Detective" for Metal Cutters

In this study, Marcin Gołaszewski from the West Pomeranian University of Technology in Szczecin, Poland, proposes a clever new way to watch a tiny metal cutter in action. Think of the cutter as a high-speed spinning drill bit, no wider than a pencil eraser (specifically, 1.5 mm), zooming across a piece of aluminum that was just 3D printed. The goal is to figure out two things in real-time: Is the cutter actually touching the metal, and is there a crack or a hole in the path ahead?

Usually, to know what a cutter is doing, you need complex sensors and heavy math that takes a long time to calculate. But this paper suggests a much simpler trick. Imagine the cutter is a drummer hitting a drum. When the drumstick hits the skin, it makes a loud noise (a force). When it hits the air, it's silent. The author's method is like a very smart microphone that just counts how many times the drumstick hits the skin versus how many times it misses in the air.

How the Method Works
The researchers used a special machine to measure the "push" (force) the cutter feels as it spins. They didn't try to map the whole 3D shape of the cut or do complicated transformations. Instead, they set a simple "noise threshold." If the force is tiny (just the vibration of the machine), they assume the cutter is in the air. If the force is big, the cutter is biting into the metal.

By counting the "hits" (where the force is high) versus the total time the cutter spends spinning, they calculated a "tool engagement ratio." It's like a score from 0 to 1. A score of 1.0 means the cutter is fully buried in the metal (like a slot milling). A score of 0.5 means it's only half-buried. If the score suddenly drops to 0.5 or lower while the cutter is supposed to be in a solid block, the system knows, "Hey! There's a gap or a crack here!"

What They Found
The team tested this on a custom machine using a tiny 1.5 mm cutter spinning at 21,000 revolutions per minute. They created different scenarios:

Solid Metal: When cutting a smooth, solid block, the "score" stayed high, matching what they expected.
Different Depths: When they changed how deep the cutter went, the score changed accordingly, showing the method could track the depth of the cut.
Fake Cracks: They cut across artificial gaps of different widths (from 0.25 mm to 1.75 mm). The method successfully spotted gaps that were 0.5 mm wide or larger. The "score" would dip, and the length of the dip matched the width of the gap.
Real Cracks: They even tested on a piece of metal that had a natural crack (about 1.8 mm wide) from the 3D printing process. The method spotted the crack and even gave clues about the shape of the damaged area, like a "porous" island in the middle of the break.

The Limits of the Detective
However, the paper is honest about where this detective fails. The method is not perfect for tiny things. If a crack is smaller than 0.5 mm (specifically, they tried 0.25 mm), the cutter moves so fast that it doesn't have time to "feel" the gap. The forces don't drop low enough to trigger the alarm, so the tiny crack goes unnoticed.

Also, the method gets a bit confused if the cutter is moving in a way that changes the direction of the push force. In those specific situations, the "score" might not be as accurate, though it can still show that something is changing.

Why This Matters
The biggest win here isn't just that it works; it's how simple it is. Other methods often require massive datasets, complex training, or heavy computers to figure out what's happening. This method just needs a simple "noise filter" and a counter. It's so fast (taking about 1.8 milliseconds to calculate) that it could run on a standard computer while the machine is working, allowing the robot to fix its path instantly.

The author suggests that while this isn't a magic bullet for every single tiny defect, it's a powerful, low-cost tool for hybrid manufacturing. It allows factories to catch cracks and adjust for bumpy surfaces without stopping the machine, keeping the production line moving and the parts high-quality. The paper concludes that while more research is needed to handle even smaller cracks and different tool sizes, this "counting the hits" approach is a solid step toward smarter, self-correcting manufacturing.

Technical Summary: A Method for Real-Time Milling Monitoring Based on Tool Engagement Analysis in Hybrid Micro-Manufacturing

Problem Statement
Hybrid manufacturing, which integrates additive (AM) and subtractive processes, faces significant challenges in the subtractive stage due to the inherent characteristics of additively manufactured surfaces. These surfaces often exhibit poor quality, including cracks, delamination, and dimensional inaccuracies caused by residual stresses and layered structures. In micro-machining (tool diameters < 1 mm), these surface irregularities lead to unstable tool engagement, fluctuating cutting conditions, and potential "air milling" (loss of contact). While process monitoring exists for additive stages, there is a need for in-situ monitoring during the subtractive phase to detect surface defects and compensate for workpiece misalignment without interrupting the integrated manufacturing process. Existing monitoring methods often rely on complex calibration, extensive training datasets, or high computational costs, making them less suitable for real-time implementation in small-batch or single-piece production.

Methodology
The paper proposes a computationally simple method for estimating the degree of tool engagement ( $p_{real}$ ) based on cutting force signals. The approach is designed for real-time operation and relies on the following principles:

Force Signal Analysis: The method analyzes cutting forces in the feed direction ( $X$ ) and the direction normal to the feed ( $Y$ ). Axial forces ( $Z$ ) are excluded due to low magnitude and high noise susceptibility.
Threshold-Based Detection: A force threshold ( $F_{th}$ ) is established to distinguish between cutting and non-cutting states. A sample is considered "non-cutting" if the force in either the $X$ or $Y$ direction falls below this threshold (accounting for measurement noise and drift).
Engagement Calculation: The method calculates the ratio of the actual time the tool is engaged in cutting ( $t_{real}$ $t_{r e a l}$ ) to the theoretical engagement time ( $t_{en}$ $t_{e n}$ ) for a single cutting edge during a half-rotation. This is achieved by counting the number of valid samples ( $n_{th}$ $n_{t h}$ ) within a buffer of size $n_{dyn}$ $n_{d y n}$ (corresponding to the theoretical engagement time).
- $p_{real} = n_{th} / n_{dyn}$
Geometric Interpretation: The calculated $p_{real}$ value is mapped to the spatial domain to estimate the radial depth of cut and detect the width of surface discontinuities. The method assumes a two-flute tool with a zero helix angle for simplified kinematic modeling, though experiments utilized a tool with a 46° helix angle.
Calibration: The only required calibration is the determination of the force threshold based on the noise level of the dynamometer when the spindle runs without cutting. No complex force characterization or machine learning training is required.

Experimental Setup
The method was validated on a custom-built machine using 1.5 mm diameter solid carbide end mills machining aluminum alloy 6061 produced via additive manufacturing. Experiments covered five categories:

Continuous slot milling.
Radial immersion milling at varying depths.
Milling over artificially created discontinuities (0.25 mm to 1.75 mm).
Milling over natural delamination defects (~1.8 mm).
Variable radial depth of cut scenarios.

Cutting forces were measured at 204.8 kHz, and surface topography was verified using an optical measurement system.

Key Results

Radial Depth Estimation: The method successfully estimated tool engagement for radial immersion milling. However, discrepancies were observed at higher radial depths (e.g., 750 μm and 1050 μm) where the mean absolute error reached 0.13. This is attributed to changes in the direction of the feed force crossing the threshold band, leading to an underestimation of engagement. The method showed better agreement (error ~0.06) at lower radial depths where force direction remained constant.
Discontinuity Detection: The method effectively detected surface cracks and delaminations wider than 0.5 mm. For discontinuities of 0.5 mm to 1.75 mm, the mean relative error in width determination was 11.14%.
Limitations in Sensitivity: Discontinuities smaller than 0.5 mm (specifically 0.25 mm) were not detected under standard threshold settings because the tool did not have sufficient time to return to an undeformed state before the force signal dropped below the threshold. Increasing the threshold allowed detection of the 0.25 mm gap but significantly reduced the accuracy of engagement estimation for continuous surfaces.
Real-Time Performance: The computational time was approximately 1.8 ms on a standard CPU, significantly faster than neural network-based approaches (~20 ms).

Significance and Claims
The paper claims that the proposed method offers a practical solution for in-process quality control in hybrid micro-manufacturing by:

Simplifying Calibration: Eliminating the need for extensive datasets or complex training procedures required by statistical or neural network methods. Calibration is limited to a simple noise-based threshold.
Direct Physical Interpretation: The indicator $p_{real}$ directly corresponds to the physical degree of tool engagement, unlike statistical indicators that require mapping.
Real-Time Feasibility: The low computational complexity enables immediate implementation in monitoring systems to detect surface defects (cracks, delamination) and estimate geometric deviations (radial depth variations) without disrupting the manufacturing flow.

The author acknowledges limitations, specifically the inability to monitor axial depth of cut, reduced sensitivity to very small discontinuities (<0.5 mm), and decreased sensitivity in regions where the feed force direction changes. Future work is suggested to address these sensitivity issues and extend the method to a wider range of tool diameters and defect types.

The "Silent Detective" for Metal Cutters

More like this