Int J Adv Manuf Technol (2008) 39:118–128

DOI 10.1007/s00170-007-1186-6

ORIGINAL ARTICLE

Tool condition monitoring in an end-milling operation

based on the vibration signal collected through
a microcontroller-based data acquisition system
Julie Z. Zhang & Joseph C. Chen

Received: 23 April 2007 / Accepted: 20 July 2007 / Published online: 6 September 2007
# Springer-Verlag London Limited 2007

Abstract Machine condition plays an important role in 1 Introduction and literature review
machining performance. A machine condition monitoring
system will provide significant economic benefits when Competition in the global market has transformed modern
applied to machine tools and machining processes. Devel- manufacturing technology from mass production to lean
opment of such a system requires reliable machining data manufacturing—a model originated in Japan that seeks to
that can reflect machining processes. This study demon- avoid wasteful activities to optimize profitability of
strates a tool condition monitoring approach in an end- products and services. The monitoring of manufacturing
milling operation based on the vibration signal collected processes and equipment conditions is an essential part of a
through a low-cost, microcontroller-based data acquisition critical strategy that drives manufacturing industries to-
system. A data acquisition system has been built through wards being leaner and more competitive [1, 2].
interfacing a microcontroller with a signal transducer for Machining is a process that removes unwanted material
collecting cutting vibration. The examination tests of this from a workpiece in the form of chips to obtain the desired
developed system have been carried out on a CNC milling product shape, size, accuracy, and surface quality. US
machine. Experimental studies and data analysis have been industries spend US $100 billion annually to machine
performed to validate the proposed system. The onsite tests metals because the vast majority of manufactured parts
show the developed system can perform properly as require machining at some stage in their production [3].
proposed. Monitoring machine conditions plays a very important role
in producing high-quality, low-cost parts.
Keywords Tool/cutter condition monitoring . Vibration . Using cutter condition monitoring as an example, one
Microcontroller . Data acquisition sees that a dull or broken tool may result in an undesired
surface finish or a part that does not meet specifications. In
terms of the lean manufacturing concept, regularly sched-
All rights reserved. This study, or parts thereof, may not be
reproduced in any form without written permission of the authors. uled machine maintenance and tooling replacement are
This paper has not been published nor has it been submitted for indispensable in keeping the machine running at optimum
publication elsewhere. conditions. However, as tooling is quite expensive, ideally,
J. Z. Zhang (*) cutting tools should be maximally utilized to reduce manu-
Department of Industrial Technology, facturing costs; in practice, cutting tools usually are replaced
University of Northern Iowa, and discarded after a certain period of usage to avoid defects
32 ITC,
Cedar Falls, IA 50614, USA
caused by tool failure, even though the cutters may still be
e-mail: [email protected] functional. Obviously, frequent tool replacement will not only
add machining cost (tool cost and tool setup1 cost), but also
J. C. Chen
Department of Agriculture and Biosystems Engineering,
1
Iowa State University, Tool setup refers to the time for installing the tool to the right place
Ames, IA, USA and recording the tooling information (diameter, tool length difference
e-mail: [email protected] compared to the reference tool, etc) into the CNC controller.
Int J Adv Manuf Technol (2008) 39:118–128 119

Fig. 1 The prototype of the tool

condition monitoring system Data Acquisition Subsystem
Raw signal Featured
Signal
machining signals
processing
(frequency,
algorithm
average, ....)

Yes Pattern recognition

cutter condition good?
(tool wear, part quality)
algorithm (expert
systems, NN, FL, ...)
No
Decision-making Subsystem
Corrective action

impair productivity. As such, the importance of automating and a decision-making subsystem. The data acquisition
machine condition monitoring has been recognized in order subsystem consists of a signal measurement device (sensor)
for manufacturing industries to operate at a low cost while and a signal processing algorithm. When a machining
producing superior product quality and timely delivery [4, 5]. operation is in process, the sensors dynamically collect raw
The goal of automated machine condition monitoring is to machining signals (cutting force, vibration, temperature,
reduce the machining cost and enhance the quality and pro- etc.). Those raw machining signals are then processed to
ductivity of machined products via detection of process and extract machining characteristic features. These features can
machine faults [5]. The prototype of this automated machine be used by the decision-making subsystem to derive the
condition monitoring system, as shown in Fig. 1, has been tool condition and predict part quality to determine if any
created through the comprehension and integration of studies corrective action is necessary.
involving machine tools and machining processes [1, 6]. Accurate and reliable data acquisition (the first subsys-
The machine condition monitoring system in Fig. 1 is tem in Fig. 1) is a key base for machine condition
composed of two components: a data acquisition subsystem monitoring. A number of researchers have studied a variety

Fig. 2 Information flow for the

microcontroller-based data ac- Tool can
quisition system continue to cut Tool
condition
Warning for
tool condition

Vibration
transducer
Raw data
(vibration)
Amplifier &
filter
tool condition
pattern recognition
Multi-channel (Mahalanobis-
Microcontroller (time and Taguchi approach)
frequency domain analysis)

Signal visualization
120 Int J Adv Manuf Technol (2008) 39:118–128

ization and application due to a low performance/cost ratio.

To overcome this drawback, efforts should also be made
towards making reliable data acquisition system affordable
besides searching for more generalized decision-making
algorithms.
The advance of low-cost microcontrollers has made the
development of such a tool condition monitoring system
possible. A low-cost data acquisition system will be built
Fig. 3 Three axes of the accelerometer upon the integration of a USB port interfaced micro-
controller and signal transducers. Since vibration signal was
identified as one of the most significant factors affecting
of machining signals related to machine condition monitor-
tool conditions, surface roughness, and dimensional accu-
ing, such as tool wear and tool breakage detection, surface
racy in such machining processes as turning, milling,
roughness prediction, and dimensional quality optimization.
drilling, and grinding machining operations [7–12], the
However, none of those studies has found its way to real
tool condition monitoring presented in this study will be
industrial use. The main problems facing the researchers are
based on the vibration data collected through the develop-
the lack of a universal mechanism for decision making and the
ment of a microcontroller based data acquisition system.
high cost related to the hardware components. Most of the
By applying a microcontroller, this project proposes to
studies have focused on searching for a more generalized
develop a cost-effective data acquisition system for moni-
decision-making approach [5] and put little attention to the
toring machine condition (the first component in Fig. 1).
high equipment cost. The Kistler dynamometers that were
The objectives of this project include the following:
employed in a few studies for cutting force measurement
cost more than US $10,000 [6, 11]. The accelerometer 1) To develop a microcontroller-based, cost-effective data
sensor and its associated data acquisition system would cost acquisition subsystem, through the hardware interfac-
several thousand dollars. With the sensor fusion technique, ing of the microcontroller and the vibration signal
in which more than one sensor is applied, the cost of a transducer.
machine condition monitoring system could easily exceed 2) To acquire, visualize and analyze sample data collected
US $10,000, which might be a 25–30% of the cost of a in a CNC end-milling operation.
typical CNC machine. Even though such a system could 3) To evaluate and monitor the tool condition through
perform well, the high cost would prevent its commercial- vibration data analysis.

Fig. 4 a Schematic diagram of

components integration. b Inte- X Signal
gration of the components in the Via
data acquisition system Accelerometer Y Conditioning Microcontroller Lap Top
Z Circuit Based PMD Board USB Computer

a
Microcontroller

Accelerometer

b
Int J Adv Manuf Technol (2008) 39:118–128 121

2 Structure of the proposed data acquisition system 3.1 Hardware system development

The functional information flow of the proposed data This data acquisition system employed off-the-shelf compo-
acquisition system is shown in Fig. 2. When machining is nents in a novel integrated approach, described as follows.
in operation, real-time vibration data in three directions will
– Machine vibration measurement sensor
be collected through machining signal transducer (acceler-
Vibration signals are important for monitoring the
ometer). In the stage of raw signal processing, background
machine condition in milling processes. An accelerometer
noise will first be filtered, and then characteristic features
embedded in an evaluation board was used to measure
will be extracted from the filtered machining signals. The
vibration. The evaluation board essentially includes a
dynamic stream of the characteristic features can then be
sensing element and an IC interface able to take
visualized in real time. The extracted cutting information
information from the sensing element and send an analog
will then be sent to the decision-making subsystem to
signal to the data acquisition device. This sensor is a low-
evaluate machine conditions.
cost, low-noise, and low-power consumption tri-axial
accelerometer with two measurement ranges of ±2g/±6g.
The schematic diagram of this accelerometer is displayed
3 Tool condition monitoring in Fig. 3, and the X, Y and Z directions are aligned with
through the microcontroller-based data the conventional milling machine axes. A computer
acquisition system program was written to capture and display the three-
axis vibration signals in the time domain in the manner
Tool condition monitoring was implemented through the of real time. Also, the FFT computation algorithm was
analysis of vibration data that was captured by the data included in the computer program to extract the
acquisition system. The development of this data acquisition vibration aptitude in the frequency domain, which will
system included hardware selection, circuit design and be explained in software development section.
implementation, hardware interface, circuit troubleshooting, – Microcontroller-based data acquisition device
computer software programming, system integration, and A microcontroller-based data acquisition device was
testing in real CNC end-milling processes. The following selected as a key data acquisition hub. This device
three sections describe the development of the hardware serves to interface signal (vibration, cutting force,
system, software system, and integration and testing of the temperature, etc.) transducers with the computer pro-
data acquisition system along with the vibration data analyses. gram installed in a PC via a USB port. Using a

Fig. 5 Data acquisition software interface developed by VBA programming

122 Int J Adv Manuf Technol (2008) 39:118–128

Fig. 6 Data acquisition soft-

ware interface developed by
SoftWIRE programming

standard, high-speed USB port, this device enables Microsoft Excel VBA (Visual Basic Application), and the
low-cost, PC-based analog and digital Input/Output other is based on a graphic programming package Soft-
data acquisition. Multiple analog input channels pro- WIRE (Measurement Computing, Inc). The interfaces of
vided by the device will allow further fusing of more the two software programs are shown in Figs. 5 and 6,
machining signals by adding acoustic, torque, temper- respectively. Both of them were successful in interfacing
ature, or other types of sensors. with the hardware system for real-time data acquisition.
– Electronic components (capacitors, Rresistors and Figure 7 displays one example of vibration data collected in
amplifiers) the X, Y and Z directions when an experimental CNC end-
In order to capture the real machining signal, electronic milling operation was in process.
circuits were built to amplify output voltages and filter Cutting force information is also one of the most
noises. The schematic integration of these components is significant cutting signals representing cutting tool features.
displayed in Fig. 4(a). The construction of these The channel for cutting force data acquisition (shown as a
components is shown in Fig. 4(b). Functions are pink line in Fig. 5) was programmed in the software
described as follows: development. Strain gage may be integrated in this system
– Accelerometer: Converts the physical acceleration into in the future to enhance the performance of the tool
a voltage signal. condition monitoring.
– Signal conditioning circuit: Amplifies the voltage – Vibration data analysis and fast Fourier transform (FFT)
signal and improves the resolution. The operation of any machine will generate some
– PMD Data acquisition board: Carries out A/D conver- vibration. A fundamental role of vibration data analysis
sion and captures the signal. and processing for tool condition monitoring is to
– Laptop computer: Runs the VBA program, stores and quantify signal changes, as changes in vibration often
displays the acquired accelerometer data. relate to changes in machine condition [13]. Measur-
able vibration through instrumentation is actually a
3.2 Software system development composite of the vibrations that generate from the
rotating components of the machine and even other
Two software programs were written separately in different noisy sources. The simplest machine condition moni-
software developmental environments. The first is based on toring is conducted through collecting the “overall”
Int J Adv Manuf Technol (2008) 39:118–128 123

Milling Vibrations
2.5

X
vibration

1.5
Y
1 Z

0.5

0
1 193 385 577 769 961 1153 1345 1537 1729 1921 2113
Fig. 7 Example of acceleration data collected in a CNC milling
operation

vibration and plotting the vibration data in time

domain. The “overall” signal represents the total energy Fig. 9 Installation of the sensor
content of all vibration sources at all frequencies.
Further, the overall vibration can be quantified in
mathematic or statistical approach in terms of the The accelerometer employed for data sensing in this
values such as peak, peak-to-peak, and average. Then, study are analog instruments and they output analog
the value(s) can be compared with a pre-determined signals. The FFT digitizes the input signal by processing
value so as to diagnose the cutter condition. Although the discrete data points and then performs all subsequent
this approach is useful in terms of its simplicity, it operations digitally on numbers. The ideal vibration
ignores the dynamic information contained in the machining signal would fluctuate up and down in a stable
signal. An alternative is to investigate the frequency trend introduced due to the machine spindle’s inherent
distribution of the signal. rotational movements. Tool problems, such as tool tooth
The majority of the past research was performed on the breakage or tool wear, usually distribute unbalanced around
measurement and analysis of the machining signals in the the tool teeth, therefore incurring correspondingly addition-
time domain only. Recent studies have found frequency al vibrations. Meanwhile varied machining parameters or
domain analyses to be more useful in tool condition other conditions may also cause increased vibrations. In
monitoring [11], especially when the tool fault is caused order to diagnose the tool’s true condition, it is crucial to
by unbalanced conditions. Fast Fourier transform (FFT) is associate machine signal changes to defects. This task is not
an efficient signal processing algorithm to conduct frequen- easily accomplished in the time domain alone because
cy domain analysis. For example, the FFT algorithm was measured signal did not distinguish the vibration sources.
applied in a tool wear monitoring system for a CNC turning The analysis in frequency domain can help distinguish the
operation [14]. Their study found that the frequency display machining vibrations generated at different frequency
showed a significant difference for the turning process bands, which may be caused by unbalanced tool breakage
using new tools versus worn tools. or tool wear and will show up as a peak different from the

Fig. 8 Experimental setup for

data acquisition system Tool holder Signal condition
development interface circuit

Accelerometer
Sensor

Microcontroller
Data acquisition
124 Int J Adv Manuf Technol (2008) 39:118–128

In the machining test, the accelerometer was mounted on

the vise (shown in Fig. 9) to collect vibration data generated
in the cutting process. The milling cutter cut through the
entire workpiece (1.5 in). To make the cutter properly
plunge into the workpiece, ramp-on and ramp-off distances
were set in the cutting path (shown in Fig. 11).
– Vibration data analysis in time domain
The experimental data showed that vibration ampli-
tudes varied with the change of cutting conditions. For
example, Fig. 12 displays the comparison of the
vibration data by using tools #3 and #4 under the same
machining parameter setup (feed rate 16.5 in/min,
spindle speed 2750 rpm, depth of cut 0.08 in). Test
runs at this setup lasted about 35 seconds, including
Fig. 10 Implementation of data acquisition system setup ramp-on, machining and ramp-off time. For tool #3,
which is a new cutter, vibration signals in time domain
inherent vibration in the frequency domain. Using this as were evenly spread in the entire cutting process, and
the tool diagnosis principle, one can detect and track tool there was almost no distinguishable difference when
problems , while the machining operation is in process the cutting was performed in ramp-on/off versus when
(Fig. 8). the cutting was performed in the workpiece material; in
other words, the vibration was marginal and not
3.3 Integration and testing of the data acquisition system significant; whereas for tool #4, from the visualized
time domain vibration data, it can be seen that there
– Experimental setup were obvious differences when the cutting was per-
The schematic of the experimental setup is shown in formed in ramp-on/off versus when really cutting
Fig. 8. The installation of the sensor and the implemen- workpiece material. Noticeably, there were some
tation of the data acquisition system are displayed in vibration spikes that occurred in the cutting process
Figs. 10 and 11, respectively. When tested in machining and/or ramp-on/off process. The machining test was
a workpiece, the sensor was protected to prevent any conducted in a full capability machining production
interference caused due to machining chips. lab. At this time, regular machining operations and labs
Two sets of end-mill cutters (listed in Table 1) were were still in place and such kind of spikes could be
tested in the machining experiments. One tool set included caused by any nearby machine turning on/off and
four new end-mill cutters with different specifications. The unknown noise. The spike was taken care of by the
other tool set included six used end-mill cutters in the same designed filter so that its influence was prevented when
specification but with a variety of flaws such as teeth interpreting machining data and making frequency
breakage, tool wear, or having molten material between domain analysis.
cutting teeth. Two feed rates were varied for each tool in As shown in Fig. 12, it appears that machining vibration
order to make experiments in a wider range. The experi- in the Y direction carries more information about the cutter
mental arrangement is listed in Table 2. Data sampling condition. When the workpiece material was cut, clearly
frequencies were set up at least more than two times of larger amplitude vibrations were generated, especially in
spindle frequency to collect vibration data. the Y direction. Figure 12 also showed that the amplitude

Fig. 11 Top view of milling

cutter cutting through a Feed rate
workpiece
End mill

1.5in

Ramp-on Ramp-off
Int J Adv Manuf Technol (2008) 39:118–128 125

Table 1 Cutting tools used in the machining experiment #3 and #4. Tool #3 is a brand new 11/23” 4-tooth high
Tool # Diameter Number Tool condition speed steel end mill without center hole, and tool #4 is
(in) of teeth a used end mill in the same specification with random
flank wear, which is applied as a control group for
1 1 6 New tool comparison with tool #3.
2 1 6 Unequal breakage in a few teeth
The natural frequency brought by spindle rotation for
3 13/32 4 New tool
tool #3 and tool #4 is 46 Hz. No obvious peaks in the
4 13/32 4 Melted material unequally spread
in teeth frequency domain were found around the natural frequency
5 5/8 4 New tool range or the harmonic frequency bands in the X direction
6 5/8 4 Tool with not severe wear, signals when tool #3 (new tool) was used in cutting. On the
7 9/16* 4 Two teeth with breakage contrary, a very obvious frequency peak was found in the
8 1/2 4 New tool first harmonic frequency, around 45 to 50 Hz for tool #4
9 1/2 4 Severe, equal wear in every teeth (old tool, control group for tool #3). A similar pattern was
10 1/2 4 Breakage and melted material
found in the Y direction signals; there was almost no peak
*There is not a brand new tool for comparison with tool #7. But the in the entire frequency band when the new tool was cutting,
data collected by tool #7 can be compared with data collected by tool while cutting by using the old tool generated a huge
#5 and #8 frequency peak around 45 to 50 Hz. Such a pattern was not
found in the Z direction, and it appeared that the Z direction
of vibration signals in the Z direction is generally larger signal distribution is random. Similarly, for other tool pair
than in the X and Y directions. Approximately similar comparisons, obvious frequency peak values appeared in
patterns displayed in Figs. 13 and 14 also have been found the X and Y, but not in the Z, directions. The data analysis
in the comparison of tool #5 and #6, as well as tool #8 and results indicated the feasibility of using cutting vibration
#9. The acceleration signal has a mean value which is amplitudes and the frequency peaks in the X and Y
different from zero, and this can be explained by the directions for the monitoring of tool condition in end-
experimental setup — the Z direction of the accelerometer milling operations. The vibration amplitudes in time
sensor aligns with the direction of free falling acceleration. domain and the frequency peaks at harmonic frequency
The acceleration in the Z direction cannot be zero, and it bands will be the key featured signal for monitoring the tool
should remain consistently around one g (m/s2) (gravity condition. Since the natural frequency is very much
acceleration) in the entire cutting process. However, the
mean value of the Z acceleration did show some variations
for different cutting tools at different cutting conditions.
Table 2 Experimental arrangement table
These variations might be due to the varied experimental
setup or the developed system’s instability and further # of Tool Spindle Feed Depth of Sampling
investigations should be explored accordingly for the next sample # (rpm) (in/min) cut (in) frequency Hz
stage’s research.
1 1 1200 36 0.08 100
In testing runs, vibrations from tool #1 and #2 are 2 1 1200 18 0.08 100
obviously different, but such a pattern has not been found 3 2 1200 36 0.08 100
in their comparison. Vibration from tool #1 showed normal 4 2 1200 18 0.08 100
cutting condition. Breakage in tool #2 occurred under 5 3 2750 33 0.08 300
severe conditions, whereas four out of six teeth were 6 3 2750 16.5 0.08 300
broken in varied extent. The left two teeth could not even 7 4 2750 33 0.08 300
8 4 2750 16.5 0.08 300
complete the testing cutting path. Therefore, the collected
9 5 1800 21.6 0.08 150
vibration data could not be compared compatibly.
10 5 1800 10.8 0.08 150
– Vibration data analysis in frequency domain 11 6 1800 21.6 0.08 150
Machining vibration data in three directions were 12 6 1800 10.8 0.08 150
collected and visualized when the CNC machine is in 13 7 2000 24 0.08 200
operation; meanwhile, FFT analyses were performed to 14 7 2000 12 0.08 200
display the distribution of cutting vibration in the 15 8 2250 27 0.08 200
frequency domain. The FFT analysis indicated that, 16 8 2250 13.5 0.08 200
17 9 2250 27 0.08 200
the frequency peak (FP), at the X and Y directions can
18 9 2250 13.5 0.08 200
be used for monitoring cutting tool conditions. For 19 10 2250 27 0.08 200
example, Fig. 15 displayed the comparison of the FFT 20 10 2250 13.5 0.08 200
analysis result for the vibration data collected by tools
126 Int J Adv Manuf Technol (2008) 39:118–128

3_2, tool #3 4_2, tool #4

4 4
vibration by voltage, in V

3.5 3.5

vibration voltage, in V
3 3
2.5 X 2.5 X
2 Y 2 Y
1.5 Z 1.5 Z
1 1
0.5 0.5
0 0
1 501 1001 1501 2001 2501 3001 3501 1 501 1001 1501 2001 2501 3001 3501
sample points (sample frequency, 100Hz) sample points (sampling frequency, 100Hz)
Fig. 12 Comparisons of the vibration signal amplitudes collected by tools #3 and #4

dependent on the machine spindle movement, the featured that were written in both VBA and SoftWIRE were
signals and the machine’s natural frequency (a pre- able to capture the machining information.
determined value) will be monitored and compared contin- – The vibration signals of the X, Y and Z directions in
uously when the end milling is in cutting process. time domain were captured and displayed in real time.
The quantitative correlation analysis of tool wear and The displayed vibration is helpful in understanding the
cutting vibrations is expressed as the second subsystem in cutting condition.
Fig. 1. The correlation analysis will be the next step – The fast Fourier transform (FFT) function and its
research to identify and quantify machining tool conditions graphic display were integrated into the software
through real-time machining signals, and a detail was not program developed by SoftWIRE. Data were visualized
investigated in this study due to the research scope. in real-time.
– The vibration amplitudes in time domain and the
frequency peaks at harmonic frequency bands of the
4 Conclusion and future research X and Y directions can be used as the key featured
signals for monitoring the tool condition.
This project aimed to monitor the tool condition in a CNC – Although currently only three channel data (vibrations in
end-milling machining processes, based on the vibration X, Y, and Z direction) were captured and displayed, more
signal collected through a microcontroller-based data channels have been programmed in the software pro-
acquisition system. The testing result showed that the grams to capture and display up to eight channels of data
prototype design was successful. The conclusions can be when more sensor devices are applicable in the future.
drawn as follows:
– The data acquisition system, mainly composed of an
accelerometer evaluation board, and microcontroller 5 Future research thrust
(these components cost less than US $200), was an
effective, cost-competitive system that can be applied Cutting force information is one of the most significant
for monitoring machining processes. cutting signals representing cutting tool condition. To keep
– The interface of the hardware system and the software low cost and application flexibility, strain gages were
program was successful. The two software programs originally considered to be integrated in the data acquisition

5_2, tool #8 6_2, tool #6

vibration by voltage, in V

4.5 4.5
vibration by voltage, in V

4 4
3.5 3.5
3 X 3 X
2.5 2.5
Y Y
2 2
1.5 Z 1.5 Z
1 1
0.5 0.5
0 0
1 501 1001 1501 2001 2501 3001 1 501 1001 1501 2001 2501 3001

sample points (sampling frequency at 150Hz) sample points (sampling frequency at 150Hz)
Fig. 13 Comparisons of the vibration signal amplitudes collected by tools #5 and #6
Int J Adv Manuf Technol (2008) 39:118–128 127

8_2, tool #8 9_2, tool #9

3
vibration by voltage, in V

vibration by voltage, in V
2.5 2.5
2 X 2 X
1.5 Y 1.5 Y
1 Z 1 Z
0.5 0.5
0 0
1 501 1001 1501 2001 2501 3001 1 501 1001 1501 2001 2501 3001
sample points (sampling frequency at 200 Hz) sample points (sampling frequency at 200 Hz)
Fig. 14 Comparisons of the vibration signal amplitudes collected by tools #8 and #9

3_1,tool#3 4_1, tool #4

4.5
4.5
3.5
3.5

2.5
2.5
Mag-X

Mag-X
1.5 1.5

0.5 0.5

-0.5 0 -0.5
20 40 60 80 100 120 140 160 0 20 40 60 80 100 120 140 160
Frequency (Hz) Frequency (Hz)

Distribution of vibration acceleration in the X direction in the frequency domain

3_1, tool #3 4_1, tool #4
40 40
35 35
30 30
25 25
Mag-Y

Mag-Y

20 20
15
15
10
10
5
5
0
0
0 20 40 60 80 100 120 140 160
0 20 40 60 80 100 120 140 160
Frequency (Hz)
Frequency (Hz)

Distribution of vibration acceleration in the Y direction in the frequency domain

3_1, tool #3 4_1, tool #4

4 4
3.5 3.5
3 3
2.5 2.5
Mag-Z

Mag-Z

2 2
1.5 1.5
1 1
0.5 0.5
0 0
0 20 40 60 80 100 120 140 160 0 20 40 60 80 100 120 140 160
Frequency (Hz) Frequency (Hz)

Distribution of vibration acceleration in the X direction in the frequency domain

Fig. 15 Comparison of the FFT analysis results on tools #3 and #4
128 Int J Adv Manuf Technol (2008) 39:118–128

system for monitoring tool condition. However, strain References

gages have not been included in real time testing because
of installation problems. A strain gage is usually in a form 1. Al-Habaibeh A, Gindy N (2000) A new approach for systematic
of very thin metal foil (<1 mm) with two terminals for design of condition monitoring systems for milling operations.
J Mater Process Technol 107:243–251
connecting with the electronic circuit, a strain gage must be
2. Frankowiak M, Grosvenor R, Prickett P (2005) A review of the
glued perfectly to the experimental object surface. Once a evolution of microcontroller-based machine and process monitor-
gage has been used and detached, it is almost impossible ing. Int J Mach Tools Manuf 45:578–582
for the gage to be re-used if no special measures are applied 3. DeGarmo EP, Black JT, Kohser RA (1997) Materials and
to protect the gage from the harsh machining environment, processes in manufacturing, 8th ed. Prentice Hall, Upper Saddle
River, NJ
normally full of cutting fluids, chips, and dust. To 4. Chen S, Jen YW (2000) Data fusion neural network for tool
overcome the installation problem of cutting force mea- condition monitoring in CNC milling machining. Int J Mach Tools
surement devices, designing a special fixture that can not Manuf 40(3):381–400
only house the sensor but also hold the workpiece may be a 5. Al-Habaibeh A, Zorriasatine F, Gindy N (2002) Comprehensive
experimental evaluation of a systematic approach for cost
possible solution in the future. effective and rapid design of condition monitoring systems using
Efforts for future work should be made to study the Taguchi’s method. J Mater Process Technol 124:372–383
correlation of frequency peak value in X and/or Y direction 6. Milfelner M, Cus F, Balic J (2005) An overview of data
(s) with tool wear condition. A large volume of experimen- acquisition systems for cutting force measuring and optimization
in milling. J Mater Process Technol 164:1281–1288
tal data should be collected by varying spindle speed, feed 7. Rehorm AG, Jiang J, Peter E (2005) State-of-the-art methods and
rate, and depth of cut with this data acquisition system, and results in tool condition monitoring: a review. Int J Adv Manuf
this data should be analyzed using statistical or artificial Technol 26:693–710
intelligence algorithms. In addition, research effort will also 8. Tzeng Y, Jean M (2005) Dimensional quality of optimization of
high-speed CNC milling process with dynamic quality character-
be engaged to investigate the stability of the developed istic. Robot Comput-Integr Manuf 21:506–517
system that has been mentioned above in the analysis. 9. Ertekin YM, Kwon Y, Tseng T (2003) Identification of common
Alternatively, due to the development of wireless sensor sensory features for the control of CNC milling operations under
network (WSN) technology, future work of embedding varying cutting conditions. Int J Mach Tools Manuf 43(9):897–904
10. Jantunen E (2002) A summary of methods applied to condition
WSN into machine tool condition monitoring to replace
monitoring in drilling. Int J Mach Tools Manuf 42:997–1010
wiring cables provides a promising direction for future 11. Chung KT, Geddam A (2003) A multi-sensor approach to the
research, not only saving cost, but also greatly increasing monitoring of end milling operations. J Mater Process Technol
the flexibility of the machine tool condition monitoring 139:15–20
12. Zhang Z, Chen JC (in press) Development of an in-process
system.
surface roughness adaptive control in end milling operations. Int J
Adv Manuf Technol 31:877–887
Acknowledgements The authors would like to gratefully acknowl- 13. Shearman S (2003) Machine condition monitoring, part 1:
edge financial support for this study from the Grow Iowa Value vibration acquisition and analysis. Sensors. Retrieved August 29,
Applied Research Grant. Acknowledgement and appreciation also is 2007 from http://archives.sensorsmag.com/articles/0503/16/
given to the Department of Industrial Technology at the University of 14. Smith G, Lee S (2005) A method for detecting tool wear on a
Northern Iowa for the usage of equipment, and to Mr. Jeffery Rose for CNC lathe using a doppler radar detector. Int J Adv Manuf
his help with the experimental tests. Technol 25:270–280