Development of a Method and a Smart System for Tool Critical Life Real-Time Monitoring

Abstract

Tool wear management and real-time machining quality monitoring are pivotal components of realizing smart manufacturing objectives, as they directly influence machining precision and productivity. Traditionally, measuring and analyzing cutting force fluctuations in the time domain has been employed to diagnose tool wear effects. This study introduces a novel, indirect approach that leverages spindle-load current variations as a proxy for cutting force analysis. Compared to conventional methods relying on machining vibration or direct cutting force measurement, this technique provides a safer, simpler, and more cost-effective means of data aquisition, with reduced computational demands. Consequently, it is ideally suited for real-time monitoring and long-term analyses such as tool-life prediction and surface-roughness evolution induced by tool wear. An intelligent tool wear monitoring system was developed based on spindle-load current data. The system employs extensive cutting experiments to identify and analyze the correlation between tool wear and spindle-load current signal patterns. By establishing a tool wear near-end-of-life threshold, the system enables intelligent monitoring using C#. Experimental validation under both roughing and finishing conditions demonstrated the system’s exceptional diagnostic accuracy and reliability. The results demonstrate that the current ratio threshold value has good universality in different materials, indicating that monitoring the machining current ratio to estimate the degree of tool wear is a feasible research direction, and that the average error between the experimental surface-roughness measurement value and the predicted value was 10%.

1. Introduction

The machinery industry underpins the progress of the manufacturing sector. As labor scarcity and escalating operational expenses become increasingly acute, the machining industry is compelled to adopt intelligent automation to drive transformation and growth. Fortunately, the real-time monitoring capabilities inherent to smart manufacturing, a cornerstone of Industry 4.0, effectively address the demands of this industrial evolution. Furthermore, the current trend is towards Industry 5.0. It recognizes the power of industry to achieve societal goals beyond jobs and growth, to become a resilient provider of prosperity for a sustainable, human-centric and resilient European industry. This study continues to follow the paradigm and vision of Industry 4.0 and 5.0.

Overlooking machining quality in unmanned factories can readily result in significant product defects or costly rework, compromising production efficiency. Tool wear directly impacts machining precision and overall process integrity. Consequently, real-time tool wear monitoring is indispensable for ensuring machining quality and preventing process anomalies in the pursuit of intelligent manufacturing. In response to evolving industry demands, this study develops an efficient, precise, and economical online intelligent monitoring system for tool life and machined surface quality based on CNC machine tool spindle-load current. The system offers three primary advantages: (1) enhanced workpiece machining quality, (2) improved process efficiency, and (3) reduced operational expenditures.

A plethora of research has been dedicated to tool wear prediction and monitoring, yet existing methodologies necessitate further refinement. Previous studies predominantly concentrate on analyzing and identifying anomalous patterns during the machining process, often requiring costly instrumentation and equipment, thereby impeding industrial implementation. To circumvent these challenges, this study endeavors to develop an online tool wear monitoring system utilizing machine tool spindle-load current, predicated on the analysis of anomalous signal patterns induced by tool wear. The system constructs a diagnostic rule for tool wear and process anomalies based on the evolution of signal characteristics, facilitating real-time tool wear monitoring and machining abnormality detection.

Tool wear exacerbates cutting forces. Traditionally, dynamometers have been employed for direct measurement of cutting forces during milling, but their substantial spatial requirements and high cost render them impractical for sustained production environments. Fontaine et al. [1] utilized a dynamometer to quantify three-axis cutting forces generated by a ball-end mill, investigating the influence of the machining inclination angle—the angle between the workpiece plane and the ball-end mill’s rotational axis—on cutting forces. Through experimental analysis, they established a recommended range of machining inclination angles that minimize cutting force impacts for tool paths traversing from low to high points and vice versa. Albertlli et al. [2] measured concurrently three-axis machining vibration with an accelerometer and cutting force with a dynamometer. Subsequently, they derived a correlation equation between amplitude and cutting force based on the measured three-axis cutting force. Experimental outcomes demonstrated a strong correlation between machining vibration and cutting force trends. Zhu et al. [3] explored the effects of varying tool inclination angles and cutting speeds on cutting force and tool wear. Their experimental findings indicated that high-speed cutting results in lower cutting forces and superior energy efficiency compared to low-speed cutting, attributable to reduced tool-tip-to-workpiece contact time, and consequently diminished frictional wear.

Numerous researchers have investigated the utility of diverse signals for tool wear monitoring and machining process optimization. Altintas [4] explored the potential of inferring cutting force from variations in feed-axis motor input current. His findings suggest that time-domain analysis of feed-axis motor current can effectively predict cutting force magnitude when sampling frequency is sufficiently high. Soliman et al. [5] employed spindle-load current for chatter detection, demonstrating that time-domain analysis of current characteristics can effectively identify chatter even when machine tool current bandwidth is approximately one-third of the chatter frequency. Li et al. [6] utilized feed-axis motor input current for tool wear monitoring. Through extensive cutting experiments, they identified an accelerated wear rate as a diagnostic end-of-life threshold. By monitoring for significant upward trends in tool wear rate during machining, they established a current-based tool-wear diagnosis rule. Yong et al. [7] simultaneously measured milling machine servo-motor current and cutting load. Their experimental results indicated a linear correlation between absolute motor input current and cutting load magnitude. Regression analysis yielded a relationship equation between feed-axis motor current and cutting load. Experimental validation demonstrated that the error between estimated cutting force using motor load current and actual cutting force was within 8%. Herwan et al. [8] measured turning machining vibration, observing a sudden amplification upon tool breakage. They concluded that high-frequency machining vibration monitoring can enable real-time detection of abrupt events such as tool breakage or damage.

Endika Tapia et al. [9] used a statistical method to detect outliers in the manufacturing process. In the study, a software platform that can monitor and detect outliers in an industrial manufacturing process using scalable software tools was built. The platform shows the results visually on a dashboard. The monitoring of a five-axis milling machine and use of simulated tests were performed. The results prove the suitability and scalability of the platform and reveal the issues that arise in a real environment. Iñigo Aldekoa et al. [10] presented a study with evidence that sensorless machine-variable monitoring can achieve tool wear monitoring in broaching in real production environments, with advantages including reducing production errors, enhancing product quality, and facilitating zero-defect manufacturing. The machine data collected facilitated the training of a set of machine learning models, accurately estimating tool wear on the broaches. The test results showed high predictive accuracy, with a coefficient of determination surpassing 0.9. The authors [11] presented the method using the GelSight device for topographic measurements of microworn surfaces of a tool. The instrument uses a gel-backed elastomer tactile membrane providing a different approach from traditionally used optical microscopes and sensors. It provides advantages such as cost, portability, and the ability to measure translucent and highly reflective surfaces over other surface metrology systems. Moreover, industrial communication protocols are essential to interconnect systems, interfaces, and machines in industrial environments for smart manufacturing applications. Endika Tapia et al. [12] evaluated OPC-UA, Modbus, and Ethernet/IP with three machine tools to assess their performance and their complexity of use from a software perspective. The results show that Modbus provides the best latency figures and communication has different complexities depending on the used protocol, from the software perspective.

During machining, elevated temperatures and friction induce oxidative wear on cutting tools. This wear progressively alters tool tip geometry, diminishing sharpness and resulting in increased cutting forces. Consequently, machining vibration and spindle energy consumption escalate with rising cutting forces. Moreover, machining vibration imparts tool marks on the workpiece surface, compromising surface roughness. In large-area material removal processes, severely worn tools experience a substantial reduction in cutting capacity. Insufficient spindle torque to drive the tool under excessive cutting force can lead to catastrophic tool failure, resulting in scrap, spindle damage, and costly production downtime. Hence, real-time tool-wear status monitoring and timely tool replacement are imperative for preserving product quality and process stability.

Several methods can be used for machine status/machining process monitoring, such as force measurement, acoustic emission detection, and vibration detection, etc. A dynamometer can be used for force measurement, but due to its expensive cost, inconvenient installation, and limited machine space for its installment, it is not suitable for the applications requiring 24 h monitoring for many machines in a factory or the machining process with a large workpiece. Much research explored the possibility of using sound/voice signals to diagnose machine health status or abnormalities of the manufacturing process. However, the interferences problem caused by the background noise from other machines/environment is till the major issue needed to be resolved for on-line precision diagnosis. The vibration signals can precisely reflect the instant status of a tool, a machine, or a manufacturing process through time-domain or frequency-domain signal analysis with the advantages of lower cost and relatively easy setup. The motor current of a machine spindle of a CNC machine could provide same the functions with even lower cost compared to the vibration signals, and a digital current meter can be easily installed in the electrical cabinet of a CNC machine to collect the spindle-load current. Therefore, the motor current has become a popular signal for real-time monitoring functions.

To effectively quantify tool wear severity, this study adopts the ISO 8668-2 [13] standard to establish tool wear evaluation criteria. Auto-Optical Inspection (AOI) is employed to measure tool wear, and a predictive model for tool life is developed, incorporating characteristic changes in spindle-load current. During measurement, the end mill under test is laid flat to assess flank wear. Figure 1a illustrates typical end-mill wear patterns, where VB1 represents uniform flank wear, VB2 represents non-uniform flank wear, and VB3 represents localized flank wear. According to the ISO standard, tool replacement criteria based on the tool-edge projection line are as follows:

• Replace the tool when uniform flank wear on each edge exceeds 0.3 mm.
• Replace the tool when non-uniform flank wear on any edge exceeds 0.5 mm.

Beyond ISO standard-based tool wear monitoring, sudden catastrophic tool breakage, as depicted in Figure 1b, can occur during machining, significantly impacting workpiece surface roughness. Despite not reaching the ISO-recommended replacement threshold, increased cutting force induces a sudden spindle-load current spike. In such cases, the operator should replace the tool, based on the required machining accuracy.

To meet the exigencies of real-time computational efficiency and cost reduction, this study employs a digital current meter to continuously monitor spindle-load current characteristic values and their temporal trends. Regression analysis is subsequently applied to establish a model correlating tool wear, spindle-load current, and machining surface roughness (induced by cutting vibration). A spindle-load current-based tool-life-monitoring rule and real-time machining surface-roughness-prediction method are then developed. Based on these rules, a C# program is constructed to create an intelligent tool status and life monitoring system. Finally, experimental cutting is conducted to validate the performance, accuracy, and reliability of the developed prediction method and monitoring system.

2. Research Methodology

2.1. Signal Analysis Methodology

This study examines the time-domain signal characteristics of spindle-load current and cutting vibration during machining operations. Continuous material removal induces tool wear, resulting in increased cutting resistance. Consequently, corresponding elevations in current and vibration signal characteristics occur. These time-domain signal variations accurately reflect the tool’s real-time wear status, enabling the utilization of associated signal features for monitoring tool wear deterioration.

2.1.1. Spindle-Load Current Sampling and Analysis

Machining-induced tool wear is a continuous, gradual process whose rate varies based on machining conditions (e.g., roughing or finishing). As illustrated in Figure 2, the initial cutting phase with a new tool exhibits a slow wear rate. However, with increasing cutting time and accumulated wear, the wear rate accelerates markedly as the tool approaches its critical wear threshold. This accelerated wear rate is mirrored in a corresponding increase in spindle-load current. Consequently, the rate of change in spindle-load current can serve as an indicator of the tool’s proximity to its critical wear threshold.

The fastest current sampling frequency—1 Hz of the digital current meter—was used in the study. Because tool wear is continuously accumulating during the machining process (slow wear rate at the early and middle stage, and fast wear rate at the final stage), the cutting vibration and spindle-motor load current vary, following a certain trend. 1 Hz sampling rate is quick enough to collect sufficient data to show the signal variation trend to differentiate the varying tool wear status. In this study, the relatively current increase ratio was calculated based on the sampled current data, and compared with the threshold value.

Experimental findings suggest that a current sampling frequency of 1 Hz suffices for analyzing these signal characteristic changes.

Figure 3 depicts the temporal variation of the experimentally measured spindle-load current. As illustrated in Figure 3, the load current approaches zero during spindle quiescence (non-rotation). A pronounced inrush current is observed during spindle startup. During idle spindle rotation, a minor current, often termed motor idle current, persists, and exhibits slight fluctuations across varying set speeds. Even at a constant speed, minor current variations occur. Upon tool engagement with the workpiece and commencement of machining, the current rises and stabilizes. Conversely, upon tool disengagement from the workpiece, the spindle-load current synchronously decreases to idle current levels. Tool wear amplifies cutting resistance, necessitating CNC-controller current adjustments to maintain the requisite motor torque. Consequently, the relative increase in spindle-load current, compared to the initial cutting current, increases with a new tool, and serves as a valuable signal analysis feature for assessing tool wear severity through tool wear monitoring.

To enhance analysis accuracy, this study excludes current data acquired during spindle startup and idle operation. For each complete cutting pass, the root mean square (RMS) value of the top 20% current values is calculated. Subsequently, RMS values are averaged across multiple cutting passes (post tool wear) and compared to the new tool-cutting RMS value. The relative increase in the RMS value and its rate of change (time-domain slope) are defined as characteristic indicators of the approaching critical wear threshold. These features are compared against measured tool wear data and ISO standards to determine tool proximity to the critical wear state.

2.1.2. Machining Vibration Sampling and Analysis

This study concurrently collects cutting vibration data and motor load current to evaluate the feasibility of motor load current-based analysis. Cutting vibration primarily originates from momentary impact and friction as the blade engages the workpiece surface. Vibration amplitude is predominantly correlated with cutting resistance, which escalates with tool wear. Prior to cutting-vibration signal acquisition, the theoretical blade passing frequency is calculated to determine the minimum vibration sampling frequency according to Equation (1):

$$F_q={\displaystyle \frac{fS}{60}},$$

(1)

where $F_q$ is the blade passing frequency (Hz), $f$ is the number of blades, and $S$ is the spindle speed (rpm). Given the high-frequency, periodic nature of machining vibration, the Nyquist Theorem mandates a sampling frequency at least five times the vibration signal’s maximum primary frequency, to prevent aliasing distortion.

To accentuate vibration signal features, this study analyzes the initial 20% of the vibration signal, selected based on experimental data stability. Given the periodic nature of vibration data, with both positive and negative components, direct averaging can obscure characteristic features. Consequently, the root mean square (RMS) algorithm is employed to analyze vibration signal trends. The RMS value neutralizes the impact of positive and negative signs on the average, while assigning greater weight to larger amplitude-signal squares, effectively capturing time-domain vibration-signal trend characteristics resembling a normally distributed periodic function. The RMS calculation formula is as follows:

$$RMS=\sqrt{{\displaystyle \frac{\sum _{i=1}^n{x}_i^2}{n}}},$$

(2)

where $n$ represents the total number of data points and $x_i$ represents the i-th data point.

2.1.3. Outlier Detection Using Quartiles

This study employs the quartile method, a statistical outlier detection technique, to eliminate anomalies within the spindle-load current data. The method establishes fixed data intervals and identifies data points exceeding dynamic upper- and lower-quartile-based outlier detection thresholds. These data points are classified as outliers and excluded from the subsequent analysis. Figure 4 presents a schematic of the quartile method. The definitions of terms IQR (InterQuartile Range) = Q3 - Q1, Q1, Q2, and Q3 in Figure 4 are as follows.

Q1: 

First quartile, representing the 25th percentile of the data arranged in ascending order;

Q2: 

Median, representing the middle value of the data arranged in ascending order;

Q3: 

Third quartile, representing the 75th percentile of the data arranged in ascending order.

2.2. Modeling Method

To effectively analyze tool wear signal characteristics, this study leverages regression analysis and machine learning techniques to develop predictive models. Changes in machining current under constant cutting volume signify tool wear, subsequently leading to increased surface roughness. Regression analysis is predominantly employed for real-time surface-roughness estimation and current-increase slope calculation.

2.2.1. Linear Regression Analysis

As previous tool wear studies indicate, machining tool wear is characterized by a gradual increase in cutting resistance. Consequently, this study employs linear regression analysis to estimate the slope of the increasing machining tool wear signal, thereby predicting the tool’s remaining life. The simple linear regression equation is expressed as follows:

$$y={\beta }_0+{\beta }_{1}x,$$

(3)

where y represents the predicted output value, $x$ represents the input value, ${\beta }_0$ represents the intercept, and ${\beta }_1$ represents the slope.

2.2.2. Multivariate Linear Regression

Multivariate linear regression is a statistical technique that models the relationship between multiple independent variables ($X$) and a dependent variable ($Y$). By fixing the number of independent variables and the model structure, it generates predictions ($\widehat{Y}$) and compares them to actual values ($Y$) in training data, calculating the error ($\epsilon$) between them. The objective is to estimate regression coefficients ($\beta$) and the intercept ($c$), which minimize overall error. The core equation for multivariate linear regression is the following:

$$\widehat{y}={\beta }_1{x}_1+{\beta }_2{x}_2+\dots {\beta }_n{x}_n+c,$$

(4)

where $n$ is the number of independent variables, ${\beta }_n$ is the regression coefficient for each independent variable, and $c$ is the intercept.

To assess the discrepancy between predicted and actual values, two primary error metrics are commonly employed: Mean Squared Error (MSE) and Mean Absolute Error (MAE). These metrics are defined by the following equations:

$$MSE={\displaystyle \frac{1}{N}}\sum _{i=1}^N{(y_i-\widehat{y_i})}^2$$

(5)

$$MAE={\displaystyle \frac{1}{N}}\sum _{i=1}^N|y_i-\widehat{y_i}|$$

(6)

where $i$ denotes the data index, $N$ denotes the total number of data points, $y_i$ denotes the actual value for the ith data point, and $\widehat{y_i}$ denotes the predicted value for the i-th data point, using the regression model.

2.3. Experimental Setup

The experimental setup, depicted in Figure 5, simultaneously collects spindle-load current, machining vibration, and machining parameters. Spindle-load current is directly acquired from the digital current meter using a C# program. Machining vibration and parameters are obtained from both the DAQ system and the controller. The integrated data are transmitted to an edge computer via Servbox proxy. Tool wear is assessed using an Automated Optical Inspection (AOI) image system (Figure 6), while workpiece surface accuracy is measured with a surface-roughness gauge (Figure 7).

3. Experimental Design and Data Analysis

To satisfy the efficiency and cost-control mandates of intelligent manufacturing, anomaly-monitoring systems require real-time communication, monitoring, and adaptive anomaly threshold adjustment to accommodate diverse machining conditions. Compared to costly dynamometers and vibration accelerometers, we employ low-cost digital current meters, primarily focusing on spindle current measurement, while concurrently measuring machining vibration to validate current signal accuracy. To analyze current signal characteristics indicative of wear and machining anomalies, this study conducted two distinct experiments. First, a simple linear cutting-tool path experiment measured spindle-load current and machine tool vibration simultaneously, to verify current signal characteristics under critical tool-life conditions. Second, experiments on severe tool wear were conducted to examine current characteristics under varying material removal rates. Additionally, surface-roughness correlation experiments established the relationship between tool wear and machining surface roughness.

Building upon the established experimental design, this study develops a tool-life monitoring rule and a surface-roughness estimation rule. To validate the accuracy and reliability of this system under complex machining conditions, two distinct verification experiments were conducted. The tool-life verification experiment employed two different materials to assess the system’s capability to diagnose tool life across varying material conditions. The surface-roughness-estimation verification experiment comprised two phases: initial validation within the rule’s parameter range, followed by extended testing beyond this range to evaluate the system’s surface-roughness prediction accuracy under complex machining scenarios.

To investigate spindle-load current characteristics related to tool life and the correlation between tool wear and surface roughness, two experiments were conducted: a tool-life experiment and a tool-wear–surface-roughness correlation experiment. Two types of end mills, procured from Precision Technology Co., Ltd., Taichung, Taiwan, were employed for cutting experiments. These end mills varied in terms of tooth count, coating, and helix angle. Tool specifications are detailed in

Table 1. Specification table of end mills.

Name	Mode	Flutes	Helix Angle	Coating	Diameter (mm)
T1	PPE0603	3	45°	TiSiN	6
T2	PSE0604	4	35°	AlTiBin	6

.

3.1. The Tool-Life Verification Experiment

To investigate tool life under varying machining parameters, a tool-life experiment was conducted using SK2 high-hardness quenched steel (HRC 55 ± 2) as the workpiece material. Two tool specifications and three cutting parameters were evaluated. Each experimental condition was replicated three times to ensure data reliability.

Given machine rigidity constraints, the experiment focused on characterizing tool critical wear under varying cutting widths while maintaining constant feed for the tooth and cutting depth. A tool-life monitoring rule was established based on these experimental findings. The experiments primarily involved continuous straight-line cutting. Tool wear was assessed when the relative increase in spindle-load current and machining vibration demonstrated significant time-domain variations. This process was iterated until the tool approached ISO standards or experienced tool tip fracture.

Development of Tool Critical-Life Prediction Models Based on Experimental Data

As outlined in Section 2.1, a tool flank wear of 0.3 mm served as the criterion for tool-life determination. Multivariate regression analysis was applied to the experimental data. For the T01 tool experiment, data exhibiting wear between 0.2 mm and 0.4 mm were utilized for model development, while data within the same wear range were employed for validation in the T02 tool experiment. The regression equation derived from T01 tool data is

$$\widehat{y}=0.0668x_1+0.9x_2+1$$

(7)

where $\widehat{y}$ represents the predicted current ratio, $x_1$ represents the cutting width, and $x_2$ represents the maximum flank wear.

As detailed in

Table 2. Experimental data table of T01 end-mill cutting experiment.

Experiment Number	Cutting Width (mm)	Measured Wear (mm)	Measured Current Ratio	Predicted Current Ratio	Error (%)
01	2	0.23	1.44	1.34	6.6
02	2	0.28	1.46	1.39	4.7
03	2	0.33	1.46	1.44	1.5
04	2	0.32	1.33	1.42	7.3
05	2	0.33	1.41	1.43	1.5
06	2	0.36	1.44	1.46	1.6
07	2	0.21	1.32	1.32	0.3
08	2	0.28	1.38	1.39	1.1
09	3	0.3	1.54	1.48	4.0
10	3	0.36	1.55	1.53	1.1
11	3	0.3	1.46	1.47	1.1
12	3	0.31	1.51	1.48	1.7
13	3	0.23	1.45	1.41	1.0
14	3	0.33	1.45	1.50	1.5
15	3	0.35	1.52	1.52	0.2
16	3	0.37	1.52	1.54	1.2
17	4	0.2	1.29	1.45	12.6
18	4	0.25	1.57	1.50	4.5
19	4	0.33	1.64	1.57	4.2
20	4	0.24	1.42	1.49	5.0
21	4	0.32	1.49	1.56	4.7
22	4	0.38	1.54	1.61	4.5
23	4	0.23	1.56	1.48	5.3
24	4	0.33	1.58	1.57	0.9
25	4	0.37	1.71	1.61	6.0

The mean error is 3.5% and the mean absolute error (MAE) is 0.0496.

and

Table 3. Experimental data table of T02 end-mill cutting experiment.

Experiment Number	Cutting Width (mm)	Measured Wear (mm)	Measured Current Ratio	Predicted Current Ratio	Error (%)
01	2	0.23	1.43	1.34	6.1
02	2	0.39	1.61	1.49	8.0
03	2	0.24	1.35	1.35	0.6
04	2	0.43	1.63	1.43	6.1
05	2	0.25	1.46	1.36	6.9
06	2	0.33	1.47	1.43	2.5
07	3	0.23	1.41	1.41	0.0
08	3	0.32	1.51	1.49	1.4
09	3	0.34	1.77	1.51	15.0
10	3	0.28	1.41	1.46	3.2
11	3	0.35	1.42	1.52	6.9
12	3	0.25	1.53	1.43	6.4
13	3	0.35	1.77	1.52	14.0
14	4	0.33	1.84	1.57	14.8
15	4	0.37	1.84	1.60	12.9
16	4	0.2	1.59	1.45	8.7

The mean error is 7.1% and the mean absolute error (MAE) is 0.1177.

, the established model can predict the corresponding tool-life current ratio threshold by substituting the cutting width and predetermined critical wear of 0.3 mm into the equation. Under equivalent wear conditions, varying cutting widths influence proportionally the tool-life current ratio. This phenomenon may be attributed to the relatively small spindle size of the experimental machine tool. When the tool is initially sharp and cutting resistance is low, the motor input-current range across different cutting widths is comparatively narrow. Conversely, with severe tool wear and increased cutting resistance, the spindle requires a larger current input to overcome the cutting resistance of wider cuts while maintaining speed, resulting in a broader motor input-current range across different cutting widths.

Experimental results indicated a consistent increase in tool-life current ratio, exceeding 1.4 times across varying cutting widths. To ensure diagnostic rule generalizability, a current ratio threshold of 1.4 times was established. If the measured spindle-load current during machining surpasses the initial value by a factor of 1.4, the rule categorizes the tool as severely worn, and recommends replacement. Figure 8 summarizes the tool-life diagnosis steps and criteria.

Different machining parameters, such as feed rate, spindle speed, depth of cut, etc., could cause different chip load and cutting forces. The diagnosis model was built based on different experiments with different machining parameters, and the tool wear monitoring rules is developed based on the result of whether the relative current increasing ratio is close to or exceeds the threshold value. Because the relative current increasing ratio is defined as the ratio between the instant current and the current of the first cut of a new cutter, the increasing ratio could reflect the current status of the accumulated tool wear.

3.2. Experiment on the Correlation between Tool Wear and Surface Roughness

Surface roughness is primarily influenced by cutting volume per tooth, material properties, and machining vibration. As tool wear progresses, spindle-load current and machining vibration tend to increase. To investigate the relationship between spindle-load current, surface roughness, and wear during finish machining, medium carbon steel (HB167-229, HRC20) was selected as the workpiece material. Two distinct end mills and three finish machining parameters were employed for the cutting experiments.

To simulate finish-machining conditions, feed per tooth was varied while maintaining constant cutting width and depth. Spindle-load current, tool wear, and surface roughness were measured concurrently. Prior to each experiment, a calibration process using a test piece compared actual measurements to the test piece measurement report. Multiple measurements were acquired from the same surface, and the average of the five most concentrated data points determined the surface center-line average roughness (Ra). Each experiment was replicated three times, to assess result reproducibility.

Establish Prediction Rules for Surface Roughness during Processing

This study utilized T01 tool experiment data to develop a surface-roughness prediction model, as depicted in Figure 9. To mitigate overfitting, the model was constrained to a second-order polynomial. Subsequently, 67 measured data points from the T02 tool experiment were employed for validation. The average absolute error in predicted surface-roughness ratio was 0.164, corresponding to a 9.6% error rate. These findings suggest that the regression model accurately predicts surface roughness for different tools under consistent material and cutting-volume for tooth conditions. Consequently, the relationship between current and surface-roughness ratio is modeled by the following equation:

$$y=-2.3695x^2+8.0588x-4.6501$$

(8)

where $y$ represents the ratio of increase in surface roughness and $x$ represents the ratio of increase in spindle-load current.

This study presents a novel method for predicting surface roughness during machining. Initially, the user inputs the surface roughness of a workpiece produced by a new end mill. Subsequently, the established current–surface-roughness relationship is employed to calculate the spindle-load–current ratio during the machining process, enabling estimation of the wear increase ratio. By multiplying this increase ratio by the tool’s initial surface roughness, a predicted current surface-roughness value is obtained. Figure 10 illustrates the flowchart of this surface-roughness prediction methodology.

For the finishing process, surface roughness is usually a machining tolerance which needs to be followed. Thus, the tool wear should be controlled by the requirement of surface roughness, rather than the standard of ISO 8668-2. It is why the correlation model between tool wear and surface roughness was developed in this study. The correlation model could provide two functions: (1) predict the surface roughness, based on the relative ratio of current increase; and (2) define the threshold value, based on the tolerable surface roughness defined. With the second function, a user can define a threshold value based on the correlation model. When the current increasing ratio is close to or exceeds the threshold value, the system will display an alarm to remind the user that the cutter has worn and will cause unqualified surface roughness.

4. Human–Computer Interface (HCI) System

Leveraging the established tool remaining-life prediction model, this study developed a real-time tool-status-monitoring human–computer interface (HCI) system using C#. As illustrated in Figure 11, the system incorporates function modules for data acquisition, analysis, rule-based decision making, and abnormal condition alerts, to facilitate real-time monitoring and diagnostics. Meanwhile, a human–machine friendly interface was built for easy use for users. As shown in Figure 11a,b, users can easily interact with the system, set up the initial conditions, and observe the real-time variations in the extracted diagnosis signals and the monitoring results at HCI.

The spindle-load current fluctuates when the tool cuts in and out. Because the current signals collected at the cut-in and cut-out area cannot reflect the true tool-wear status, they should be ignored for the signal feature calculation. For this consideration, the human–computer-interface system was designed to help users to choose the locations of the cutting path and where to start and end, collecting current data from the NC program.

5. Experimental Verification

To assess the accuracy, reproducibility, and robustness of the machining anomaly-monitoring system, a rocker arm component was selected as the test workpiece. Machining paths for roughing and finishing were designed and simulated using NXCAM (Siemens, ver. 2306). The roughing validation experiment employed straight-line cutting for material removal, while the finishing validation experiment utilized small-width contour cutting. These validation experiments evaluated the accuracy of both the tool-life diagnosis and surface-roughness prediction modules. The roughing experiment verified real-time monitoring of the spindle-load current ratio. A current ratio exceeding the tool-life threshold triggered a system prompt for tool replacement, indicating potential tool flank wear. The finishing experiment involved selecting three fixed measurement positions on the machined surface to assess the surface-roughness prediction module’s accuracy based on actual measurements. The tool-life prediction module’s predictive accuracy was tested under varying machining materials. Figure 12 illustrates the workpiece and simulated machining path.

5.1. Cutting Experimental Verification of Tool Critical Life

To validate system accuracy, experiments were conducted using SK2 quenched steel (HRC 55 ± 2) under the parameters outlined in

Table 4. Machining parameters for cutting experimental verification of tool critical life.

Parameters	Value
Cutting width (mm)	3
Cutting depth (mm)	0.5
Feed per flute (mm/flute)	0.005
Hardness of material	HRC 55 ± 2

. Prior to experimentation, a smooth cutting area on the workpiece was designated as the system diagnosis surface, based on the simulated cutting path. Preselecting a stable cutting area effectively minimized interference. The monitoring system’s tool–critical-life ratio threshold was established at 1.4 times. As illustrated in Figure 13, time-domain analysis of average load current reveals an increasing trend, attributed to wear. Upon exceeding the current–increase ratio threshold, processing is halted, and the tool is inspected. As depicted in Figure 14b, the tool tip exhibits partial chipping, indicative of severe wear. Consequently, tool wear can be estimated based on the current–increase ratio, enabling tool-life monitoring through current characteristics.

5.2. Cutting Experimental Verification of Surface Roughness

To validate surface-roughness prediction accuracy, five replicate measurements were conducted to assess measurement reproducibility. Simultaneously, three points, as indicated in Figure 15, were measured for verification in each experiment. Prior to experimentation, an initial cut using machining parameters outlined in

Table 5. Machining parameters for cutting for experimental verification of surface roughness.

Parameters	Value
Cutting width (mm)	0.5
Cutting depth (mm)	0.5
Feed per flute (mm/flute)	0.02
Hardness of material	HRC 20

yielded a measured surface roughness of 1.27 μm, which was input into the system. A surface-roughness threshold of 2 μm was established. During the cutting experiment, surface roughness within the fixed processing area was measured after each machining cycle.

Table 6. Cutting for experimental verification results of surface roughness.

Surface Roughness (μm)
Measuring Point—Machining Cycle	First Machined Surface	P1-1	P2-1	P3-1	P1-2	P2-2	P3-2
First measurement	1.28	1.39	1.22	1.22	1.24	1.46	2.60
Second measurement	1.29	1.44	1.21	1.28	1.32	1.36	2.58
Third measurement	1.26	1.47	1.27	1.23	1.30	1.35	2.45
Fourth measurement	1.26	1.50	1.27	1.27	1.25	1.30	2.41
Fifth measurement	1.24	1.48	1.21	1.21	1.30	1.41	2.46
Average measurement value	1.27	1.46	1.24	1.24	1.28	1.39	2.50
Predictive value	1.27	1.26	1.45	1.27	1.36	1.62	2.60
Error (%)	0.3	13.5	17.3	2.3	6.1	16.9	4.0

The average error is 10.0%.

presents the experimental results. The predicted surface roughness exceeded the threshold value only after the second machining cycle. For the same machining area, P1-1 measurements consistently exceeded P1-2 values across different cycle numbers. This discrepancy may be attributed to either measurement error in P1-1 or incorrect probe pressure. The error between predicted and average values across the three verification points ranged from 2.3% to 17.3%, with an average experimental error of 10%. These findings demonstrate the system’s ability to predict surface roughness increases corresponding to actual tool wear. Moreover, the outlier detection mechanism effectively eliminates abnormal cutting-load current values caused by corner cutting or tool path design errors, reducing the likelihood of misjudgment.

6. Conclusions

To enable low-cost machining status and anomaly monitoring, this study established an intelligent tool anomaly-monitoring system based on the measurement of spindle machining load current. Preliminary experiments were conducted in the roughing scenario to investigate the current characteristics of tool life, and then experiments were conducted in the finishing scenario to analyze the current characteristics of surface-roughness changes. Validation experiments were conducted to verify the function of the intelligent tool anomaly-monitoring system. Conclusions and prospects are as follows:

• In different materials, when the critical-life–current ratio exceeds the threshold value of 1.4 times, the flank wear is observed to be greater than the ISO-defined tool life wear of 0.3 mm. The results demonstrate that the current ratio threshold value has good universality in different materials, indicating that monitoring the machining current ratio to estimate the degree of tool wear is a feasible research direction.
• Based on a large number of experiments, the surface-roughness prediction rule established by this study was verified within the cutting parameter range of the preliminary experiment establishment. The average error between the experimental roughness measurement value and the predicted value was 10%. The average error in the verification outside the cutting parameter range of the preliminary experiment was 7.9%. The experimental results demonstrate that surface roughness can be predicted in real time by analyzing the changes in current characteristics.