1. Introduction
Understanding the health and capabilities of manufacturing assets is critical to optimize performance, maintain adequate part quality, and to aid in data-driven decision making in the manufacturing industry [
1]. CNC machines are highly utilized in industry as they can carry out various types of cutting processes, such as end-milling, drilling, slotting, grinding, and turning processes. The market for milling machines alone accounts for USD 17.25 billion and is expected to reach USD 21.55 billion by 2026 in the United States [
2]. The automotive and aerospace sectors employ the highest number of milling machines by far compared to other industries. Milling can be used to manufacture parts with a wide range of slots, grooves, and threads, and many more complex designs. The tools used in milling processes undergo excessive wear over the duration of the cutting process and can fail prematurely if cutting parameters and tool wear are not accounted for.
As the cutting parameters vary with each process, it is critical to monitor and detect tool wear. Tool breakage can cause unscheduled maintenance stops, which leads to machine downtime. In fact, tool failure accounts for almost 7–20% of total machine downtime and tool changes can account for 3–12% of the total machining costs [
3]. To circumvent such losses, an effective tool wear detection and tool replacement strategy needs to be implemented. Tool Condition Monitoring (TCM) has emerged as one of the most effective techniques to monitor the condition of cutting tools used in the machining process [
4]. This will allow the machinist to consistently monitor tool wear and replace the cutting tool when necessary. Over the years, researchers have studied and experimented on a multitude of ways to detect tool wear. Online tool condition monitoring is an important technique as it helps in realizing the health and capability levels of a fully automated manufacturing system. A study conducted by Dimla Sr and Lister [
5] shows that tool wear occurs in combination with the predominant wear mode, dependent upon the cutting conditions, workpiece material, and the geometry of the tool itself. The techniques used to monitor tool wear are broadly classified into two types: direct sensing and indirect sensing. Although direct sensing methods provide a visual reference for the condition of the cutting tool, indirect sensing methods have been employed more frequently by researchers, due, in part, to more sensors and cutting tool data being readily available to the user.
Table 1 summarizes the benefits and disadvantages of both types of sensing techniques.
Direct sensing methods have been shown to hold certain advantages over indirect methods. Optical, radioactive, proximity, and touch-trigger probe sensors are some of the sensors used in direct sensing methods. Indirect methods make use of the data obtained when the tool is actively engaged, since they involve recording a variable that can be correlated to tool wear. Methods include sensing cutting forces, acoustic emissions, spindle motor current, and depth of cut. Dimla and Lister [
5], for example, used a Kistler mini accelerometers (type 8703 A) to measure tri-axial acceleration signals and the Kistler tool dynamometer (type 9263 A) to measure the cutting force with respect to the three planes. With the sensor data, the authors were able to obtain a correlation between the spindle speed, feed rate, and forces acting on the axes. This helped in plotting wear-time plots for different configurations of spindle speed and feed rate.
Zhou and Xue [
8] discuss both direct and indirect sensing methods and implemented indirect sensing methods in their experiments. The use of indirect sensing methods initially requires an appropriate sensor configuration, which extracts the necessary features for the user to create a robust model to monitor the state of the cutting tools. Sensor configuration can be achieved with a single sensor or a multi-sensor setup. A single sensor setup includes cutting force, vibration, motor current, and acoustic emission sensors. Other sensors, such as sound and temperature sensors, have not been effectively utilized due to the high degree of noise that results from the ambient conditions. A number of studies have shown that the cutting force is very sensitive to the condition of the tool. Wang and Wang [
9] showed that the tool wear data obtained through force sensors was the most stable and reliable signal. Yin, et al. [
10] used a piezoelectric dynamometer and a rotary dynamometer to determine how reliable these signals were, and they observed similar trends to those observed by Wang and Wang [
9]. However, Koike, et al. [
11], through their study, established that force sensors interfere with the motion control of the spindle and also reduce its rigidity. The authors also mentioned that, in some cases, the expense of using a dynamometer can lead to skyrocketing manufacturing costs, which can ultimately limit its implementation.
Vibration sensors tend to be very inexpensive, are easy to install, and provide a similar data signal to that of the force sensors. It was also determined that sharp tools have lower vibrations in comparison with tools that are blunt. A study conducted by Hsieh, et al. [
12] showed that vibration (acceleration) signals could differentiate between tool conditions during milling, provided they were used in conjunction with appropriate feature extraction techniques. However, the fundamental characteristics of the milling process limit the accuracy of the vibration sensor. Firstly, these sensors cannot always distinguish between vibration signals arising from the milling process and those arising from the air-cutting process (where no material is removed). Secondly, signals obtained by the vibration sensors are difficult to filter, as observed by Gao, et al. [
13].
As tool wear increases, the servo motor tends to draw more current to make up for the increase in cutting force. The study conducted by Madhusudana, et al. [
14] demonstrates that current sensors provide TCM data that are fairly comparable to cutting force sensor data. Also, these sensors are less susceptible to ambient and environmental noise, making the data more accurate. Lee [
15] demonstrated that, at high spindle speeds, the current sensor data were not as usable, as few changes were observed in the cutting force. Thus, current sensors are not consistently suitable for processes utilizing high spindle speeds.
Acoustic emission sensors provide data that are suitable for milling processes, as these signals have not been influenced by other mechanical effects (e.g., spindle or tool holder vibrations). These signals also propagate at a higher frequency than that of the cutting force sensors, thus reducing interference. Zhou and Xue [
3] showed that when acoustic emission sensors are placed on top of the tool holder, they provide accurate crater wear data. However, these sensors are highly susceptible to environmental/ambient noise, making it much more difficult to filter out data signals.
The literature review highlights that using individual sensors in tool condition monitoring systems can present significant challenges. In some cases, such as acoustic emission sensors, the drawbacks outweigh the advantages, making them less suitable for standalone use. As a result, many researchers have increasingly shifted towards a multi-sensor approach. This configuration integrates multiple sensor types, enhancing the depth and diversity of the information collected. While this approach may introduce some redundancy in the signals, it significantly improves the accuracy and reliability of TCM methods. Ghosh, et al. [
16] conducted a study wherein they made use of five sensors (cutting force, vibration, motor current, acoustic emission, and thermal sensors) and discovered that the force sensor contains more useful and accurate information than the other sensors. Therefore, it is critical to identify the ideal type and number of sensors to be used in combination for a specific milling process.
Table 2 shows that vibration sensors are most commonly used out of all the sensor types, making them a reliable option for TCM. Studies conducted by Bahr, et al. [
17] and Wang and Wang [
9] focused on using indirect sensing techniques (vibration and cutting force sensors) to monitor the health of the tool during the milling process. Mannan, et al. [
18] investigated a combination of direct and indirect sensors that could adequately monitor tool wear by inspecting the surface of the workpiece instead of the tool itself. Although several TCM machining experiments have been conducted, the workpiece material largely remained the same throughout the experiments. Okokpujie, et al. [
19] conducted TCM experiments considering only aluminum, whereas Özbek and Saruhan [
20] and Alonso and Salgado [
21] conducted experimental studies involving TCM for steel and titanium, respectively. To broaden the applicability of TCM, alternative materials should be considered, as industries tend to focus on carrying out multiple machining operations each day on a multitude of materials.
For example, a manufacturer from the automotive sector can work with lightweight materials including carbon fiber, magnesium alloys, and fiber-reinforced plastics, whereas the aerospace industry uses materials such as stainless steel, aluminum 6061, and titanium alloys, which are lightweight and have a higher yield strength. Although the use of tool condition monitoring for higher-yield-strength materials has been extensively studied over the years, the literature addressing its application to lower-yield-strength materials is limited. Monitoring tool wear is crucial when machining low-yield-strength materials like polyurethane foam, as even slight tool degradation can lead to a poor surface finish and dimensional inaccuracies.
As manufacturing processes utilize a wide array of materials tailored to specific applications, this research aims to evaluate the effectiveness of tool condition monitoring techniques in assessing the performance of cutting tools when used on materials with varying yield strengths. Cutting tools are essential for maintaining smooth manufacturing workflows and ensuring the quality of machined components. By extending the scope of TCM to include workpiece materials with lower yield strengths, such as polyurethane foam, this study addresses a critical gap in the understanding of how these alternative materials influence tool behavior. This expanded perspective provides valuable insights into tool life and wear, ultimately contributing to improved maintenance strategies, optimized machining operations, and broader applicability across industries and machining processes.
2. Material and Methods
In this research, end milling was carried out on aluminum 6061, polyurethane foam, 17-4 stainless steel and A36 mild steel samples using an LMV-400 3-axis milling machine (Levil Technologies, Oviedo, FL, USA) having a spindle power of 110 VAC 15 A, and a maximum spindle speed of 14,000 RPM. These materials allowed for an investigation of tool wear when milling low- and high-strength materials. As depicted in
Figure 1, five slots (each measuring 101.6 mm in length, 5 mm in depth, and 7.93 mm in width) were milled into each material at three distinct cutting speeds—low, medium, and high—tailored to the specific material type. Each slot was machined incrementally using multiple 1 mm depth-of-cut passes until the full 5 mm depth was achieved, with the total machining volume for each workpiece being 20,142 mm
3.
The process parameters were selected based on the tool material and type, as shown in
Table 3. For each material, two uncoated carbide end mills were used, represented as “new” and “worn” tools. The cutting tools used to mill aluminum 6061 and polyurethane foam are uncoated carbide 2-flute endmills, while the cutting tools used to mill stainless steel and mild steel are uncoated carbide 4-flute end mills. All cutting tools have a diameter of 5/16″ (7.93 mm), overall length of 3″, length of cut of 9/8″, and a helix angle of 30°. The worn tools were naturally worn by subjecting them to a higher depth of cut of 10 mm, a spindle speed of 10,000 RPM, and a feed rate of 60 mm/min. The cutting tools were used on their designated workpiece materials. Each new tool milled 5 slots per cutting configuration (25 passes in all) and the same procedure was followed for the worn tool. Using both new and worn tools allowed for a comparison of changes in the vibration spectra during machining between a new tool and one with significant wear. After machining, the tools underwent microscopic imaging using the OGP SmartScope (Optical Gauging Products, Rochester, NY, USA, distributed by Florida Metrology, Port St. Lucie, FL, USA) and the Keyence VHX-7000 series digital microscope (Keyence Corporation of America, Itasca, IL, USA) to observe the tool wear patterns.
Machining vibrations were recorded using an IFM uniaxial accelerometer with a data acquisition frequency of 10,000 Hz. The high-frequency capability of the IFM accelerometer was essential for capturing the subtle vibration effects caused by tool wear, particularly when machining softer materials such as polyurethane foam. Additionally, a WITMOTION triaxial accelerometer was also used and recorded data at a frequency of 200 Hz, connected to a computer via its dedicated data acquisition dashboard. The sensors were mounted to the CNC spindle to facilitate accurate data collection as shown in
Figure 2. An FFT (Fast Fourier Transform) analysis was conducted on MATLAB R2024b to observe the differences in vibrational amplitudes between the new and the worn tool. A low-pass butterworth filter was included in the analysis to mitigate the effects of high-frequency noise to observe the true behavior of the machining process.
Table 4 shows the properties of the materials used for the milling experiment. FFT has been widely used to observe vibration data and detect anomalies at specific frequencies. Researchers use this tool to analyze vibration signals for a variety of applications, not only TCM.
4. Discussion
In the filtered FFT spectra shown in
Figure 12,
Figure 14, and
Figure 16, the peak vibration was consistently higher for a worn tool compared to a new tool across all cutting configurations and material workpieces. This can be attributed to several factors. First, increased friction arises from the degraded cutting edges of worn tools, which leads to greater friction at the tool–workpiece interface, amplifying vibration levels at characteristic frequencies. Second, tool wear results in uneven cutting forces due to the irregular geometry of the worn edges, causing dynamic instabilities that contribute to higher vibrational amplitudes. Additionally, worn tools are less efficient at cutting, leading to more significant material deformation and resistance, which generates higher forces and translates into elevated vibrational energy. Finally, the rougher cuts produced by worn edges introduce periodic disturbances during machining, further reflected as higher peaks in the FFT spectra.
Furthermore, to understand the statistical significance of the outcomes of this study, an F-test was conducted with a 95% confidence interval. The table below presents the F-values derived from vibration data collected by two sensors, IFM and WITMOTION, comparing new and worn tools. The IFM sensor consistently produced F-values exceeding the critical threshold (F-critical = 3.841), as shown in
Table 6, indicating a strong statistical significance for the differences between the vibration datasets of the new and worn tools. To further support this, the table also includes p-values for each cutting configuration and material workpiece. For the WITMOTION sensor, the F-values for polyurethane foam and Al 6061 were below the F-critical value, indicating no significant difference between new and worn tools. However, for mild steel and stainless steel, the results at the medium and high cutting speeds were similar to those of the IFM sensor, with F-values exceeding the critical threshold, and therefore confirming the statistically significant differences between the new and worn tool datasets. Previous research [
39,
42,
43] shows that any useable data from vibration signals are identified at frequencies between 1 kHz and 2 kHz or higher for tougher materials such as inconel, titanium, and nickel alloys. This highlights the need to use a sensor with high sampling frequency (such as the IFM accelerometer) when machining materials with higher yield strength in order to provide useful machining data to monitor the health and wear of the cutting tools.
Figure 17 provides an overview of the yield strength and Young’s modulus for each cutting speed configuration across the materials that were machined. This chart serves as a visual representation of the diverse mechanical properties exhibited by the materials under investigation. By plotting these properties, it is possible to discern the relationships between material characteristics and machining outcomes, providing valuable insights into material selection and tooling strategies for optimized machining processes. In
Figure 17, the materials highlighted in blue analyzing data from the IFM and WITMOTION sensors had statistically significant differences (above F-critical and
p-values less than 0.05) when comparing the vibrations from machining with a new and worn tool.