Available online at www.sciencedirect.

com

ScienceDirect
Procedia CIRP 130 (2024) 214–219

57th CIRP Conference on Manufacturing Systems 2024 (CMS 2024)

A Comparative Analysis of Axis Drive Current Measurements in CNC

Machine Tools for Machine Learning Assisted Process Monitoring
Tim Reebera,*, Maximilian Berndtb, Peter M. Simonb, Jens Henningerc, Jan C. Aurichb,
Hans - Christian Möhringa
a
Institute for Machine Tools (IfW), University of Stuttgart, Holzgartenstr. 17, 70174 Stuttgart, Germany
b
Institute for Manufacturing Technology and Production Systems (FBK), RPTU in Kaiserslautern, P.O. Box 3049, 67653 Kaiserslautern, Germany
c
Empolis Information Management GmbH, Europaallee 10, 67657 Kaiserslautern, Germany
* Corresponding author. Tel.: +49 711 685-84311; fax: +49 711 685-70040. E-mail address: [email protected]

Abstract

The analysis of axis drive currents in CNC machine tools is widely recognized for its potential in process monitoring during cutting operations,
contributing to the advancement of digital twinning for machining processes. The latest generations of machine tool control systems now provide
the capability to extract drive current data directly from the position control cycle through technologies such as edge computing. This enables a
fleet wide monitoring strategy. Concurrently, an alternative approach involves acquiring three-phase current measurements post the frequency
converter. This proves to be an economical and non-proprietary retrofit solution, particularly beneficial for upgrading older machine tool systems.
In this study, both strategies for current measurement will be thoroughly assessed, and their distinguishing characteristics will be elucidated.
Furthermore, their effectiveness in facilitating process monitoring will be scrutinized through tool wear experiments. This comparative evaluation
aims to provide insights into the strengths and limitations of each method, contributing to the optimization of process monitoring techniques in
machining as well as enabling machine learning methods for tool condition monitoring.
© 2024 The Authors. Published by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0)
Peer-review under responsibility of the scientific committee of the 57th CIRP Conference on Manufacturing Systems 2024 (CMS 2024)

Keywords: condition monitoring, digital twin, machining

1. Motivation can be obstructive and block the way for machining operations,
can be damaged, are often expensive and need to be frequently
Since the development of CNC – machines, autonomous maintained. Furthermore, they require experts to be set up.
machining centres were and are of keen research interest [1]. In brown-field applications, re-equipping milling tools with
Especially in times of labour shortages, which affect machining simple and non-obstructive sensor devices for TCM would be
capacities worldwide, self-optimising machining centres are beneficial for autonomous production and early defect
highly sought after. Tool condition monitoring (TCM) is a detection. An increasingly commercially available retrofit
main component of these systems. It enables machining centres solution poses in-control drive current measurements for TCM
to monitor tool wear to adapt their machining strategies using edge computing. These devices can be retrofitted, are
accordingly. The first step in the chain of these autonomous cost-effective, and do not interfere with normal operating
TCM systems is the setup of sensors, that are capable to conditions. Control internal data sampling has the benefit of
monitor the tool condition. increased semantics through coupling the drive current data
Traditionally, TCM is achieved through implementation of with metadata such as used tool, NC-Code, and PLC-Variables
sensors close to the process, such as force, vibration or acoustic such as cooling liquid pressure or downtimes. However, the
emission sensors. However, extra sensors close to the process

2212-8271 © 2024 The Authors. Published by Elsevier B.V.

This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0)
Peer-review under responsibility of the scientific committee of the 57th CIRP Conference on Manufacturing Systems 2024 (CMS 2024)
10.1016/j.procir.2024.10.078
 Tim Reeber et al. / Procedia CIRP 130 (2024) 214–219 215

sampling frequency is usually limited by typical control cycle sensors on a machine tool feed-drive motor, estimating cutting
times (typically around 1 – 4 ms) and capturing this data power by evaluating induced voltage and multiplying it with
automatically depends on a lot of factors, such as control fluctuating current to derive cutting power [9].
vendor and version. Especially the latter is often damning, Furthermore, the measurement of drive current can be
since most CNC-machines in operation today are considered applied to predict tool degradation in the manufacturing
old. Some vendors offer the ability to extract raw control process. Consequently, the combination of drive current
internal data via drive parameterization software. Using this measurement with an accelerometer can be employed for tool
software for automated TCM poses challenges such as wear monitoring [10]. A novel approach to depict tool wear
automation of measurements and access via the network. They includes converting spindle currents and estimated process
often occur extra load on the control, since they are operating forces from the time domain into a heat map, subsequently
on the control itself. utilized as input data for a convolutional neural network [11].
To include CNC-machines without a control internal data In conjunction with pure data analysis or model-based
interface, current clamps can be easily retrofitted to the drives. resolutions, ML algorithms are increasingly being leveraged
However, the aforementioned semantics are left out and require for efficient process monitoring [12]. Various ML techniques
a robust automatic structuring technique, which can be difficult are employed to reconstruct process forces from diverse drive
when considering industrial relevant scaling. Foremost, current signals such as the motor drive current. It is evident that a Long
clamp measurements can also be noisy and must be converted Short-Term Memory (LSTM) strategy may offer advantages
to a meaningful univariate size, which can be difficult without over model-based approaches; however, the accuracy of force
any rotor shaft position values, which are needed to transform reconstruction heavily relies on the specific process [12]. This
the current values from a stationary system into a rotating one. methodology is also demonstrated for tool flank wear
Both signal sources can be used for machine learning (ML) prediction, utilizing spindle motor current and dynamometer
assisted TCM. To build universal ML-models for TCM that can measurements for the training and validation datasets of the
operate on both kinds of axis drive currents, a thorough prediction algorithm, incorporating LSTM [13].
assessment and comparison of the signals in regard to tool wear Multiple possibilities exist for directly acquiring this data
must be conducted, to assess their common validity for TCM. from the machine tool, with modern NC machine tools capable
A common and simple methodology to convert the signals as of providing such data directly from the NC itself, while an
well as a benchmark against simulated chip thicknesses will be alternative approach involves the use of external measurement
investigated. devices that can be retrofitted, offering a cost-effective
solution.
2. State of the Art
3. Methodology
The investigation and comprehension of the correlation
between drive currents and cutting forces have garnered NC controlled machines use position control to compensate
increased attention within the research community due to its for deviations in their position. This is achieved through a
significance in the industrial sector. The monitoring of drive control cascade with the position at the top level cascade and
currents in machining equipment offers direct insights into the the current at the low level cascade. Current is therefore
dynamic characteristics of the cutting process, such as tool controlled to meet the required compensation of the deviation
wear, cutting force, and overall machining efficiency [2, 3]. in the positioning of the axis. In continuous processes such as
Moreover, the utilization of drive-based data obtained from machining, there is a continuous deviation that corresponds in
manufacturing machinery has the potential to enhance frequency to the tooth engagement frequency and in amplitude
automated methods for diagnosing component failures [4] or to the material that is removed.
aiding in the optimization of energy consumption within a five- Current Measurements at drive units can either be achieved
axis machining centre [5]. It has also been effectively used to through implementing current clamps between the frequency
monitor anomalous behavior in drilling processes using auto inverter and motor drive or through sampling the currents
encoders and control internal data [6]. directly from the NC control system. However, these two
Numerous research endeavors have explored the connection current sources differ in their reference coordinate system. Fig.
between drive currents and cutting forces/power by employing 1 shows the correlations as well as their place of measurement
diverse methodologies. For example, one study proposes a within a current control cascade, the lowest cascade in a NC
technique to estimate and supervise cutting power by utilizing axis control.
an external power measurement device on feed-drive motors, NC control systems refer to the q, d - frame, which is a rotating
where cutting power is evaluated by adjusting motor current coordinate system rotating with the rotor shaft of the drive
system. This is done to simplify the control of the drive system,
based on induced voltage [7]. Another study investigates the
since the q, d – currents provide the nominal and actual values
viability of a sensorless monitoring system, which anticipates
for control. While the 3-phase current is easy to measure, it is
cutting force by utilizing direct feed drive motor current
within the a, b, c – frame that is not very descriptive, since most
commands from the CNC system. Nevertheless, the precision of the process information is superimposed by the sinusoidal
of cutting force prognosis is reliant on factors like friction behaviour of the current. Both dimensions are linked via two
modeling or the kinematics of the five-axis machines [8]. coordinate transformations, as can be seen in Fig. 1. The Clarke
Similarly, an alternative method for monitoring cutting power – Transform provides a first step for simplifying the 3-phase
involves the deployment of external voltage and current values into the α, β – frame.
216 Tim Reeber et al. / Procedia CIRP 130 (2024) 214–219

This is done through: and the material constant 1-mi. The chip thickness in shoulder
𝐼𝐼𝛼𝛼 = 𝐼𝐼𝑈𝑈 (1) milling (under the assumption that the tool setting angle κ = 0):
ℎ(𝜑𝜑) = 𝑓𝑓𝑧𝑧 ∙ sin(𝜑𝜑) (9)
𝐼𝐼𝛽𝛽 = 1⁄√3 𝐼𝐼𝑈𝑈 + 2⁄√3 𝐼𝐼𝑉𝑉 (2)
With fz being the feed rate per tooth. Φ depends upon the
0 = 𝐼𝐼𝑈𝑈 + 𝐼𝐼𝑉𝑉 + 𝐼𝐼𝑊𝑊 (3) cutting velocity and is therefore dependent from the time. In
This shows that two current clamps per drive system are the case of a motor spindle, Fc is directly linked to the torque
sufficient for practical TCM. To transform the α, β – frame into of the drive and corresponds therefore to the torque forming
the d,q – frame, the park transformation can be used: current. When measuring in a stationary system, Fc and Fn can
be converted to Fx and Fy:
𝐼𝐼𝑑𝑑 = 𝐼𝐼𝛼𝛼 cos(𝜃𝜃) + 𝐼𝐼𝛽𝛽 sin(𝜃𝜃) (4)
𝐹𝐹𝑥𝑥 = 𝑔𝑔(𝜑𝜑) ∙ (−𝐹𝐹𝑐𝑐 ∙ cos(𝜑𝜑) − 𝐹𝐹𝑛𝑛 ∙ sin(𝜑𝜑)) (10)
𝐼𝐼𝑞𝑞 = −𝐼𝐼𝛼𝛼 sin(𝜃𝜃) + 𝐼𝐼𝛽𝛽 cos(𝜃𝜃) (5)
With θ being the angular position of the rotor shaft. Iq and Id 𝐹𝐹𝑦𝑦 = 𝑔𝑔(𝜑𝜑) ∙ (𝐹𝐹𝑐𝑐 ∙ sin(𝜑𝜑) − 𝐹𝐹𝑛𝑛 ∙ cos(𝜑𝜑)) (11)
are also referred to as the torque forming current and the flux With g(φ) being a binary function that determines, whether the
current due to their alignment to the corresponding flux axis. cutting edge is in contact with material or not. The use of these
The relationship of Iq to the torque τ is [14]: relationships will help to assess the direct influence of the
𝑑𝑑𝑑𝑑
(6) cutting process on the spindle currents and the measure of
𝜏𝜏𝑀𝑀 = 𝐾𝐾𝑡𝑡 𝐼𝐼𝑞𝑞 = 𝐽𝐽𝑒𝑒 + 𝜏𝜏𝑓𝑓 + 𝜏𝜏𝑐𝑐
𝑑𝑑𝑑𝑑 disturbances in the system.
With Kt [Nm/A] being the torque constant. The total torque of
the drive system is constituted from the equivalent inertia on 3.1. Machine tool and drive current measurement setup
the motor Je per angle velocity Ω (rad/s) increment, the
frictional torque τf, and the cutting torque τc. Kt is only constant For the investigations, a CNC machine tool was used (Fig.
in the case of permanent magnet synchronous motors (PMSM). 2). The machine is outfitted with an edge computing device
In the case of asynchronous motors (ASM) such as the spindle, (Sinumerik Edge), which allows for sampling NC drive signals
Kt is a function of speed and load [14]. in position control cycle (CC) – intervals. The motor-spindle as
In industrial practice, measuring θ is very difficult and not an ASM was equipped? with current clamping devices after the
easily retrofitted in combination with current clamps, since it frequency inverter, to measure the 3-phase currents. To
requires access to the encoders of the drive system, which are
benchmark the drive currents against the cutting force, a force
often digital and use proprietary communication. However, if
measurement platform was integrated on the side of the
Iq could be approximated through Iabc, Torque and therefore
workpiece. Using a NC-Trigger, all three measurement devices
cutting forces could be approximated after subtraction of
frictional and inertial effects. are synchronized in time. The forces as well as the 3-phase

DC - Link Current
Inverse Park Transform
Uq,nom d,q Uα,nom MTCS
Iq,nom
- PI - Element
PWM
Gen.
3-
Phase
Z
Id,nom
(= 0) - PI - Element

Ud,nom
α,β
Uβ,nom
Inv. X Y
Feed
Direction
Iq Iα IU
d,q α,β IV
IW
Id α,β Iβ a,b,c
Measured via: Park Transform Clarke Transform
Current Clamp WCS
Rotor Position Φ
X Y
Sinumerik Edge
Z
Fig. 1 Current control cascade and drive system for CNC machines [15]

Furthermore, cutting forces depend on the removed chip

thickness. For this, an empirical analytical force model for
cutting and normal forces was implemented for the shoulder
milling process using the Kienzle – Equation [16, 17]: DMG Lasertec 65 3D hybrid Axis Drives
X - Axis
𝐹𝐹𝑐𝑐 (𝜑𝜑, 𝑎𝑎𝑝𝑝 ) = 𝑎𝑎𝑝𝑝 ∙ 𝑘𝑘𝑐𝑐1.1 ∙ ℎ(𝜑𝜑)1−𝑚𝑚𝑐𝑐 ( 7)
Y - Axis
Z - Axis
𝐹𝐹𝑛𝑛 (𝜑𝜑, 𝑎𝑎𝑝𝑝 ) = 𝑎𝑎𝑝𝑝 ∙ 𝑘𝑘𝑛𝑛1.1 ∙ ℎ(𝜑𝜑)1−𝑚𝑚𝑛𝑛 ( 8) Spindle
With the angular position of the cutting edge φ, the depth of cut
ap, the specific cutting/thrust force ki1.1, the chip thickness h, Fig. 2 CNC machine tool used in the investigations
 Tim Reeber et al. / Procedia CIRP 130 (2024) 214–219 217

current of the spindle were sampled at 10 kHz, while the control Cu r re n t Me a s u re m e n t S p in d le Iabs,Spindle (Drive Control off)
Analogue Measurement Clamps | Tamura
internal drive currents were sampled at 0.5 kHz, which is the S20S200D15M1 | Sampling Rate 10kHz Iabs,Spindle (Drive Control on)
40
limit of control internal data sampling due to the cycle times of
the CC. dB
Tool wear experiments were conducted using a shoulder 0

Power amplitude
mill with 5 teeth and a diameter of 50 mm in Toolox33, a -20
tempered tool steel. The parameters can be seen in Table 1. A -40
simple variation of the feed per tooth was conducted using all -60
three measurement devices. The experiments were conducted
-80
until wear exceeded 0.2 mm flank wear.
-100
Table 1 Experimental parameters of the tool wear experiments -120
0 500 1000 1500 2000 2500 3000 3500 4000 Hz 5000
Frequency
vc [m/min] fz [mm] ae [mm] ap [mm]
P1 180 0.1 23 4 Fig. 4 In-control and a,b,c-frame current of the spindle during cutting of
P2 180 0.15 23 4 P1; transparent signal: Filtering of the base signal with a moving mean
with window length = 1/2000 Hz

For the 3-phase currents, Clarke transformation was conducted

3.2. Evaluation methodology
for both the spindle and Z-axis measurements in the a, b, c -
frame. Since the drive system rotational angle is missing, the
In Fig. 3, a typical current measurement during the shoulder
quantities of the α, β – frame are summed through a root mean
milling process with parameters P1 can be seen. Both analogue
square, called Iαβ, RMS:
and in-control measurement are shown. The sinusoidal
behaviour of the currents during a constant revolution of the 𝐼𝐼𝛼𝛼 2 + 𝐼𝐼𝛽𝛽 2
spindle is overlaying the effects from the cutting process and 𝐼𝐼𝛼𝛼𝛼𝛼,𝑅𝑅𝑅𝑅𝑅𝑅 = √ (12)
2
therefore needs to be processed. To use the signal for TCM
using machine learning techniques, removing the sinusoid Through summing the α, β – frame using the root mean square
could prove beneficial, since it holds no information about the (RMS), the sinusoidal behaviour can be omitted and the I q –
cutting process itself. In machine learning, using this signal frame can be approximated. This works well because field
source could lead to an overestimation of the sinusoidal oriented controls aim to eliminate the Id – component of the q,
behaviour in comparison to the process. In contrast to force d – frame through the current control, so that only the Iq –
measurements, current measurements of the drive also include component is left.
friction and wear on the drive components. In Fig. 5, an exemplary comparison of the transformation can
Analogue current measurements with current clamps are be seen. As illustrated by the two axes, their absolute values do
susceptible to noise. Fig. 4 illustrates the frequency spectrum not align. This constant offset can be eliminated through the
of the machine spindle in two states: idle (position control investigation of the current – revolution relationship. Fig. 6
deactivated) and ready (position control activated). The shows the relationship between the mean of the calculated
machines frequency inverters operate at 8 kHz. current Iαβ, RMS,Mean and the revolution speed (blue line). After a
If the machines positional control is activated, the operation of short stagnation period, the behaviour of the current falls
the inverter can be discerned at two harmonics of the operating linearly with increased revolutions per minute. Within the
frequency of the current control cycle (CCC = 0.125 ms). To investigated revolutional area, a linear fit can be used to
reduce the impact of the noise in subsequent calculations, approximate the current offset from zero, which can be used to
smoothing was conducted using a moving mean. The / Its get the signal to the same absolute level. However, with data
window length was set to the inverse of the first harmonic of preparation for ML-techniques in mind, which usually involve
the CCC. some kind of scaling technique, this is not necessary for the

Cu r re n t M e a s u re m e n t S p in d l e IU,Spindle IV,Spindle Spindle Current: Compari son

Analogue Measurement & Control internal
Iq,Spindle Analogue Measurement & Control internal Iαβ,RMS,Spindle Iq,Spindle
current | Sampling Rate 10 kHz & 0.5 kHz IW,Spindle current | Sampling Rate 10 kHz & 0.5 kHz
Calculated Spindle Current Iαβ,RMS

80 45 2
A A A
Spindle Current Ia,b,c Iq

In-control measurement Iq

40 41 -2
20 39 -4
0 37 -6
-20 35 -8
-40 33 -10
-60 31 -12
-80 29 -14
0 5 10 15 20 25 s 35 0 5 10 15 20 25 s 35

Fig. 3 d, q-frame and a,b,c-frame current of the spindle (in cut) (P1) Fig. 5 Comparison of Iαβ, RMS, spindle and Iq, spindle for P1
218 Tim Reeber et al. / Procedia CIRP 130 (2024) 214–219

usage this paper has in mind. It should also be noted that this increase in the last 20% of both the experiments, as the flank
relationship is also load dependant. wear increases suddenly with increased flaking at the flank. P1
A further logical step would be to calculate torques and cutting accumulates a service life of 70 m of cutting motion, whereas
forces out of the calculated spindle current, since it now P2 accumulates a service life of 100 m.
approximates the torque forming current. However, since the As a benchmark signal, the force RMS was selected.
spindle is an ASM, the torque-current relationship is dependent Iαβ,RMS,spindle and Iq,spindle were correlated against Fxy,RMS,WP. The
on load and speed. In Fig. 6, this can be seen (orange line). results were an R2 value for Iαβ, RMS,spindle of 0.997 and an R2-
Torque and torque forming currents were obtained from the value of 0.994 for Iq,spindle for P1, and values of 0.989 and 0.987
control at different revolution speeds. The behaviour in the for P2 respectively. This shows that the values of the spindle
examined revolution window is somewhat linear, so torques correspond very well, with high-value and high precision force
could technically also be approximated. measurement devices. Since the RMS was chosen for a
comparison, this shows that even dynamic AC-effects are
Spindle Current Speed Dependancy
Analogue Measurement & Control internal Iαβ,RMS,Mean τSpindle/Iq,Spindle captured well, and not only the static behaviour. Considering
40
current | Sampling Rate 10 kHz & 0.5 kHz
2.4 the frequency distributions of all three signals, the signals agree
Mean of Spindle Current Iαβ,RMS

Torque-Current Relationship
A Nm/A
well within a bandwidth of 5 – 250 Hz, which is the limit for
30
the control internal data due to the limited sampling rate. Since
1.9
current clamps can be sampled at a higher frequency, the signal
25 1.5
could contain higher informational value than in – control
20 1.2
measurement.
15 0.9
To investigate this, a comparison with analytical chip
10 0.6 thicknesses is investigated. An analytical model for the chip
5 0.3 thickness of the cutting process has been programmed to
0 0 calculate the cutting forces during shoulder milling. This
0 500 1000 1500 2000 2500 3000 3500 4000 min-1 5000
Revolution Speed allows for a separation of static forces resulting from the chip
thickness during shoulder milling and the dynamic excitation
Fig. 6 Relationship of current and torque constant for the spindle in of the measurement devices (being the benchmark force
dependence to revolution speed at regular intervals with a line fit in measurement platform and the spindle).
between. The comparison can be seen in Fig. 8. Using eq. (7 – 11), the
measured forces were approximated to test the analytical
model. Specific cutting forces for Toolox 33 were
4. Comparison approximated to kc1.1 = 2447 N/mm2 and kn1.1 = 1444 N/mm2
with mc = 0.968 and mn = 1.101 through fitting the analytical
Fig. 7 shows a comparison between the calculated currents and model to the measured forces. The fit can be observed in the
the measured torque forming currents for both tool wear
experiments. Both signals are benchmarked against the cutting To o l We a r E x p e r i m e n t P 1 Fx Fy Fx,synth,Fy,synth
Analogue Measurement & Control internal
force Fxy,RMS,WP,, which is the RMS of both Fx and Fy measured current | Synthetic and real data Iαβ,RMS,Sp Iq,Sp Fc,synth
750
at the workpiece. Since the cutting force was obtained in a non- N
rotational system, the x- and y-component were also combined 250
using the RMS to mitigate rotational dependencies. For all 0
signals, the RMS was calculated in the stable cutting region and -250
Forces

plotted along the tool service life for easier comparison. Flank -500
wear measurements were conducted in frequent intervals -750
(every 3m) to measure the maximum width of the flank wear -1000
land VBmax. The tool wear rises progressively, with a sudden -1250
-1500
Tool Wear Experiments P1 & P2 FXY,RMS,WP Iq,Spindle -1750
Analogue Measurement & Control internal
current | Sampling Rate 10 kHz & 0.5 kHz Iαβ,RMS,Spindle VBmax 1
1 -
- 0.8
Scaled Quantities

0.8 0.7
P1 P2
Scaled Quantities

0.7 0.6
0.6 0.5
0.5 0.4
0.4 0.3
0.3 0.2
0.2 0.1
0.1 0
0 0 12.5 25 37.5 50 62.5 75 87.5 ms 112.5
0 12.5 25 37.5 50 67.5 75 87.5 m 112.5 Time
Tool Service Life
Fig. 8 Above: Comparison of measured and analytical calculated forces in a
Fig. 7 Comparison of Fxy,rms, Iab,rms, Iq and the maximum of the flank wear stationary system; below: Comparison of scaled currents with scaled
width VBmax for experiments P1 and P2. All quantities have been scaled to analytical cutting forces in the rotary system. The fit was done manually
a range of (0, 1) using min-max scaling. under assumption of a constant revolution speed.
 Tim Reeber et al. / Procedia CIRP 130 (2024) 214–219 219

upper part of Fig. 8. The forces can be approximated well, Agency Karlsruhe (PTKA). The author is responsible for the
reaching values of R2 for Fx,synth and Fx of 0.836 and values of content of this publication.
0.704 for Fy,synth and Fy respectively.
To check the validity of both Iq,spindle and Iαβ, RMS, spindle for TCM, References
we can use the analytical model to check their agreement on
the analytical cutting force through min-max scaling. The [1] Möhring H-C, Wiederkehr P, Erkorkmaz K, Kakinuma Y (2020) Self-
scaled comparison is seen in the lower part of Fig. 8. As can optimizing machining systems. CIRP Annals 69:740–763.
already been determined is the effect of the low sampling rate https://doi.org/10.1016/j.cirp.2020.05.007
of the control internal variable. This may lead to insufficient [2] Bergs T, Biermann D, Erkorkmaz K, M'Saoubi R (2023) Digital twins
for cutting processes. CIRP Annals 72:541–567.
agreement of Iq, spindle with the thickness of the chip when the https://doi.org/10.1016/j.cirp.2023.05.006
cutting edge enters the material due to the low sampling rate. [3] Teti R, Mourtzis D, D'Addona DM, Caggiano A (2022) Process
In the parameter set P1 and P2, an overlap of two cutting edges monitoring of machining. CIRP Annals 71:529–552.
occurs most of the time. Fc,synth bends less steep when this https://doi.org/10.1016/j.cirp.2022.05.009
happens, which can be somewhat approximated using Iαβ, RMS, [4] Bauerdick CJ, Helfert M, Petruschke L, Sossenheimer J, Abele E (2018)
spindle but not Iq, spindle,
An automated procedure for workpiece quality monitoring based on
machine drive-based signals in machine tools. Procedia CIRP 72:357–
The R2 value for Iαβ, RMS, spindle and Fc is 0.432 and for Iq, spindle
362. https://doi.org/10.1016/j.procir.2018.03.245
and Fc it is 0.291 respectively. In this case, the control internal [5] Sato R, Shirase K, Hayashi A (2018) Energy Consumption of Feed
variable also represents the expected influence of the process Drive Systems Based on Workpiece Setting Position in Five-Axis
to a lesser degree then the estimation sampled at a higher rate Machining Center. Journal of Manufacturing Science and Engineering
from the current clamps. 140. https://doi.org/10.1115/1.4037427
In both the currents and the forces, a low-frequency sinusoidal [6] Reeber T, Henninger J, Weingarz N, Simon PM, Berndt M, Glatt M,
movement can be seen. This may be due to a not perfectly Kirsch B, Eisseler R, Aurich JC, Möhring H-C (2024) Tool condition
monitoring in drilling processes using anomaly detection approaches
balanced tool, but this cannot be definitely confirmed.
based on control internal data. Procedia CIRP 121:216–221.
https://doi.org/10.1016/j.procir.2023.08.066
5. Summary and Outlook [7] Miura K, Bergs T (2019) A Method of Cutting Power Monitoring for
Feed Axes in Milling by Power Measurement Device. IFAC-
In this study, we assessed two strategies for current PapersOnLine 52:2471–2476.
measurement in CNC machine tools and evaluated their https://doi.org/10.1016/j.ifacol.2019.11.577
[8] Aslan D, Altintas Y (2018) Prediction of Cutting Forces in Five-Axis
effectiveness in facilitating process monitoring for TCM. Our Milling Using Feed Drive Current Measurements. IEEE/ASME Trans
evaluation aimed to provide insights into the strengths and Mechatron 23:833–844.
limitations of both methods, contributing to the optimization of https://doi.org/10.1109/TMECH.2018.2804859
process monitoring techniques in machining and enabling ML [9] Miura K, Döbbeler B, Klocke F (2018) Cutting Power Estimation via
methods for TCM. Through experimental investigations and External Voltage and Current Sensors on Feed-Drive Axis for the
analytical modelling, we found that both current measurement Straight Milling Process. Procedia CIRP 78:323–328.
https://doi.org/10.1016/j.procir.2018.09.068
techniques can be implemented to obtain very similar results in
[10] Doukas C, Stavropoulos P, Papacharalampopoulos A, Foteinopoulos P,
terms of capturing cutting forces and tool wear. Vasiliadis E, Chryssolouris G (2013) On the Estimation of Tool-wear
The in-control current measurement technique provided for Milling Operations based on Multi- Sensorial Data. Procedia CIRP
enhanced semantics by coupling drive current data with 8:415–420. https://doi.org/10.1016/j.procir.2013.06.126
metadata, offering comprehensive insights into the machining [11] Brecher C, Xi T, Benincá IM, Kehne S, Fey M (2021) Intelligente
process. On the other hand, retrofitting current clamps offered Überwachung von Werkzeugverschleiß/Intelligent monitoring of tool
a cost-effective and non-intrusive solution, particularly wear based on machine internal data. wt 111:309–313.
https://doi.org/10.37544/1436-4980-2021-05-43
beneficial for upgrading older machine tool systems. This
[12] Denkena B, Bergmann B, Stoppel D (2020) Reconstruction of Process
solution may offer the benefit of a higher resolution of Forces in a Five-Axis Milling Center with a LSTM Neural Network in
machining processes that is more sensible to single cutting edge Comparison to a Model-Based Approach. JMMP 4:62.
effects, but this has to be investigated further. https://doi.org/10.3390/jmmp4030062
[13] Bilgili D, Kecibas G, Besirova C, Chehrehzad MR, Burun G, Pehlivan
The study suggests that by ensuring similar results T, Uresin U, Emekli E, Lazoglu I (2023) Tool flank wear prediction
independently from the measurement method, machine using high-frequency machine data from industrial edge device.
Procedia CIRP 118:483–488.
learning-based TCM algorithms can be transferred to different https://doi.org/10.1016/j.procir.2023.06.083
machines, regardless of machine control age or vendor. This [14] Aslan D, Altintas Y (2018) On-line chatter detection in milling using
contributes to the advancement of autonomous machining drive motor current commands extracted from CNC. International
systems and advanced TCM techniques in an industrial context. Journal of Machine Tools and Manufacture 132:64–80.
https://doi.org/10.1016/j.ijmachtools.2018.04.007
Acknowledgements [15] Reinelt W, Lundquist C, Johansson H (2005) On-Line Sensor
Monitoring in an Active Front Steering System Using Extended Kalman
Filtering. In: SAE Technical Paper Series.
This research and development project is/was funded by the https://doi.org/10.4271/2005-01-1271
German Federal Ministry of Education and Research (BMBF) [16] Kienzle O, Victor. H (1954) Einfluss der Wärmebehandlung von stählen
within the “The Future of Value Creation – Research on auf die Hauptschnittkraft beim Drehen. Stahl u. Eisen
Production, Services and Work” programme (funding number [17] Surmann T (2017) Simulation der Dynamik von Dreh- und
02P20A030 ff.) and managed by the Project Management Fräsprozessen, 1. Auflage. Vulkan Verlag, Essen