See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/228787886

A Combined System for Off-Line Optimization and Adaptive Adjustment of the

Cutting Parameters During a Ball-End Milling Process

Article in Strojniški vestnik – Journal of Mechanical Engineering · January 2005

CITATIONS READS

6 245

4 authors, including:

U. Zuperl Matjaž Milfelner

University of Maribor TECOS
114 PUBLICATIONS 1,564 CITATIONS 33 PUBLICATIONS 598 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Matjaž Milfelner on 07 October 2014.

The user has requested enhancement of the downloaded file.

 COMBINED SYSTEM FOR OFF-LINE OPTIMIZATION AND
ADAPTIVE ADJUSTMENT OF CUTTING PARAMETERS DURING
BALL END MILLING PROCESS

Uroš Župerl, Franci uš, Edi Kiker, Matjaž Milfelner

Faculty of mechanical engineering, Smetanova 17, 2000 Maribor

ABSTRACT
The contribution discusses the use of combining the methods of neural networks, fuzzy
logic and PSO evolutionary strategy in modeling and adaptively controlling the process of
ball-end milling. An overall procedure of hybrid modeling of cutting process (ANfis-
system), used for working out the CNC milling simulator has been prepared. On the basis
of the hybrid process modeling, off-line optimization and feed-forward neural control
scheme (UNKS) the combined system for off-line optimization and adaptive adjustment of
cutting parameters is built. This is an adaptive control system controlling the cutting force
and maintaining constant roughness of the surface being milled by digital adaptation of
cutting parameters. In this way it compensates all disturbances during the cutting process:
tool wear, non-homogeneity of the workpiece material, vibrations, chatter etc. The basic
control principle is based on the control scheme (UNKS) consisting of two neural
identificators of the process dynamics and primary regulator. The CNC milling simulator
tests the system stability and tunes the control scheme parameters. The approach was
successfully applied to CNC milling machine Heller. Experiments have confirmed
efficiency of the adaptive control system, which reflected in improved surface quality and
decreased tool wear.

Key words: machining, force control, adaptive control with optimisation, cutting
conditions, ball-end milling.

1. INTRODUCTION

A remaining drawback of modern CNC systems is that the machining parameters, such
as feedrate, cutting speed and depth of cut, are still programmed off-line. The machining
parameters are usually selected before machining according to programmer’s experience

1
and machining handbooks. To prevent tool damage and to avoid machining failure the
operating conditions are usually set extremely conservative, which means that the values of
feeding and cutting speed are much lower that the recommended values specified by the
tool maker. Thus though the machining constrains are fulfilled, the tool and the machine
are not fully utilized. As a result, many CNC systems are inefficient and run under the
operating conditions that are far from optimal criteria. Even if the machining parameters
are determined off-line by non-deterministic optimisation algorithms based on artificial
intelligence (neural networks) [1], they cannot be adjusted additionally during the
machining process.
To ensure the quality of machining products, to reduce the machining costs and increase
the machining efficiency, it is necessary to adjust the machining parameters during the
machining process (in real-time), to satisfy the optimal machining criteria. For this reason,
adaptive control (AC), which provides on-line adjustment of the operating conditions, is
being studied with interest [2]. In our AC system, the feedrate is adjusted on-line in order
to maintain a constant cutting force in spite of variations in cutting conditions.
The focus of this research is peak force regulation in 4-axis CNC machining through the
use of off-line optimized feedrates and adaptive control. In this paper, neural adaptive
controller is developed and some simulations and experiments with the neural control
strategy are carried out. The results demonstrate the ability of the proposed system to
effectively regulate peak forces for cutting conditions commonly encountered in end
milling operations.
Force control algorithms have been developed and evaluated by numerous researchers.
Among the most common is the fixed gain proportional integral (PI) controller originally
proposed for milling by Tlusty & Elbestawi [3]. Stute & Goetz [4] proposed an adjustable
gain PI controller where the gain of the controller is adjusted in response to variations in
cutting conditions. The purely adaptive model reference adaptive controller (MRAC)
approach was originally investigated by Tomizuka [5]. These controllers were simulated
and evaluated and physically implemented by Liu [6]. Both studies found all three
parameter adaptive controllers to perform better than the fixed gain PI controller.
Unfortunately, adaptive control alone cannot effectively control cutting forces. There is no
controller that can respond quickly enough to sudden changes in the cut geometry to
eliminate large spikes in cutting forces. Therefore, we implement on-line adaptive control

2
in conjunction with off-line optimization. The optimization is performed with algorithm
developed by Župerl [7].
Much work has been done on the adaptive cutting force control for milling [8, 9].
However, most of the previous work has simplified the problem of milling into one-
dimensional motion. In this contribution, we will consider force control for three
dimensional milling.
The paper is organised as follows. The following section briefly describes the overall
cutting force control strategy. Section four covers the CNC milling simulator. Section five
describes the experimental evaluation of combined adaptive control system. Finally,
sections seven and eight present experimental results, conclusions, and recommendations
for future research.

2. COMBINED SYSTEM FOR OFF-LINE OPTIMIZATION AND ADAPTIVE CUTTING

FORCE CONTROL

The basic idea of this design is to merge the off-line cutting condition optimization
algorithm and adaptive force control (Figure 1). Based on this new combined control
system, very complicated processes can be controlled more easily and accurately compared
to standard approaches. The objective of the developed combined control system is
keeping the metal removal rate (MRR) as high as possible and maintaining cutting force as
close as possible to a given reference value. Combined control system is automatically
adjusted to instant cutting conditions by adaptation of feedrate. When spindle loads are
low, the system increases feeds above and beyond pre-programmed values, resulting in
considerable reductions in machining time and production costs. When spindle loads are
high the feed rates are lowered, safeguarding cutting tool from damage and breakage.
When system detects extreme forces, it automatically stops the machine to protect the
cutting tool. It reduces the need for constant operator supervision. Sequence of steps for
on-line optimization of the milling process is presented below.
1. The recommended cutting conditions are determined by ANfis (adaptive neuro-fuzzy
inference system) models, which are basic elements of the software for selecting the
recommended cutting conditions.
2. The pre-programed feed rates determined by off-line optimization algorithm are sent
to CNC controller of the milling machine. Optimization algorithm works on the basis
of feedforward and radial basis networks with the simultaneous use of new advanced

3
 learning algorithm which automatically adapt to current conditions during the
training process. The main objective of the paper is to determine the optimal
machining parameters (cutting speed, feedrate, depth of cut) that maximize the
extend of production, reduce the manufacturing costs and finally improve the product
quality.
3. The measured cutting forces are sent to the neural control scheme,
4. Neural control scheme adjusts the optimal feedrates and sends it back to the machine,
5. Steps 1 to 3 are repeated until termination of machining.
The neural adaptive force controller adjusts the feedrate by assigning a feedrate override
percentage to the CNC controller on a 4-axis Heller, based on a measured peak force (see
Figure 1). The actual feedrate is the product of the feedrate override percentage
(DNCFRO) and the programmed feedrate. If the software for optimization of cutting
conditions was perfect, the optimized feedrate would always be equal to the reference peak
force. In this case the correct override percentage would be 100%. In order for the
controller to regulate peak force, force information must be available to the control
algorithm at every 20ms. Data acquisition software (LabVIEW) and the algorithm for
processing the cutting forces are used to provide this information. The optimization time
by the use of off-line optimization algorithm based on feedforward neural network, is equal
to 0,001s. The combined control system returns the cutting force value to the desired value
level within four or less iteration at the latest.

4
 ANfis-modeli
ANfis models

Priporo eni rez. pogoji

Recommended cutting conditions

Off-line optimizacijski
algoritem
Off-line optimization
algorithm
On-line nastavljanje rezalnih parametrov
DNC f On-line adjustment of cutting parameters

Fref Adaptivni sistem DNCFRO CNC-krmilje Proces frezanja Dynamometer

vodenja CNC-controller
Milling process Kistler 9255
Adaptive controller FAGOR 8040-M

Fm
Simulator CNC procesa odrezavanja
CNC machining process simulator

Fx,Fy,Fz
Algoritem za obdelavo rezalnih sil
Algorithm for processing of cutting forcess

Fig. 1. Structure of combined system for off-line optimization and adaptive adjustment of
cutting parameters

3. FEED-FORWARD NEURAL CONTROL SCHEME (UNKS)

The fundamental control principle is based on the control scheme (UNKS) consisting of
three parts (Figure 2). The first part is the loop known as external feedback (conventional
control loop). The feedback control is based on the error between the measured (Fm) and
desired (Fref) cutting force. The primary feedback controller is a neural network (NM-R)
which imitates the work of division controller.
The second part is the loop connected with neural network 1 (NM-1), which is internal
model of process dynamics. It acts as the process dynamics identifier. This part represents
an internal feedback loop which is much faster than the external feedback loop as the latter
usually has sensory delays.
The third part of the system is neural network 2 (NM-2). The NM-2 learns the process
inverse dynamics.
The UNKS operates according to the following procedure. The sensory feedback is
effective mainly in the learning stage. This loop provides a conventional feedback signal to

5
control the process. During the learning stage, NM-2 learns the inverse dynamics. As
learning proceeds, the internal feedback gradually takes over the role of the external
feedback and primary controller. Then, as learning proceeds further, the inverse dynamics
part will replace the external feedback control. The final result is that the plant is controlled
mainly by NM-1 and NM-2 since the process output error is nearly zero.
This is an adaptive control system controlling the cutting force and maintaining constant
roughness of the surface being milled by digital adaptation of cutting parameters. In this
way it compensates all disturbances during the cutting process: tool wear, non-
homogeneity of the workpiece material, vibrations, chatter etc.

Identifikator dinamike procesa

Process dynamics identifier

f Fm* Fm
Perceptron
NM-1 + -

Fref + - Fm*

+
Fref Primarni regulator f Fm
+ Proces frezanja
Primary controller
Milling process
NM-R +

Fm
Zunanja regulacijska zanka
External feedback loop
Identifikator inverzne dinamike procesa f*
Process inverse dynamics identifier

Fref Perceptron + -
NM-2
f* f

Nastavljanje uteži
Weight ajustment

Fig. 2. Feed-forward neural control scheme (UNKS)

4. CNC MILLING SIMULATOR

A CNC milling simulator is used to evaluate the controller design before conducting
experimental tests. The CNC milling simulator tests the system stability and tunes the
control scheme parameters. The simulator consists of a neural force model, a feed drive
model and model of elasticity (Figure 3). The neural model predicts cutting forces based

6
on cutting conditions and cut geometry as described by Župerl [10]. The feed drive model
simulates the machine response to changes in desired feedrate. The elasticity model
represents the deflection between the tool and the workpiece (Figure 3). Model is adapted
from Muršec [11]. The system elasticity is modelled as static deflection of the cutter. The
elasticity equation for X direction of tool travel is:

X m (t ) = (− Fx (t )) + Fx (t − Ttp )) ⋅ G x (4.1)

Where Xm is the tool elastic deflection affecting the chip thickness, F is the cutting force
Gx is the compliance, t is time and Ttp is tool passing period.

Pozicija Podajanje mm/zob

Željeno podajanje delovne mize Feedrate mm/tooth
Desired feedrate Slide position Fy
1 1 Simulacija odrezavanja
+ +
2
0.001024 s + 0.064 s + 1 s - - Simulation of machining
Fx
Dinamika podajalnega
Rezalne sile
servo-sistema Korekcija po X-osi Cutting forcess
Feed drive dynamics X axis correction
+
asovni zamik -
periode zoba
One tooth period
delay 0.00002
Max
Podajanje Podajanje
+ Korekcija po Y-osi
-
Y axis correction

0.00002
Fx(t)

asovni zamik
periode zoba
Xm(t) Xm(t)
One tooth period
delay
Fig. 3. CNC milling simulator

The feed drive model was determined experimentally by examining responses of the
system to step changes in the desired feed velocity. The best model fit was found to be a
second-order system with a natural frequency of 3 Hz and a settling time of 0.4sec.
Comparison of experimental and simulation results of a velocity step change from
7mm/sec to 22mm/sec is shown on Figure 4.

7
 25

Podajanje / feedrate [mm/s]

20

Model

15 Eksperiment
experiment

10

5
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
as / time [s]

Fig. 4. Comparison of actual and simulated feedrate

Izmerjena rezultirajo a rezalna sila

Measured resultant cutting force
400
Material: Ck45
Frezalo / cutter: R216-16-03M-M GC1025
Podajanje / feedrate: 0.05-02 mm/zob
Rezalna hitrost / cutting speed: 80 m/min
300 R = 3 mm
D
AD= 3 mm
Fmax
F [N]

200

100

0
0 0.1 0.2 0.3 0.4
as / time [s]

Simulirana rezultirajo a rezalna sila

Simulated resultant cutting force
400

300
F [N]

200

100

0
0 0.1 0.2 0.3 0.4
as / time [s]

Fig. 5. Comparison of simulated and measured cutting force

8
 The feed drive model, neural force model and elasticity model are combined to form the
CNC milling simulator. Simulator input is the desired feedrate and the output is the X, Y
resultant cutting force. The cut geometry is defined in the neural force model. The
simulator is verified by comparison of experimental and model simulation results. A
variety of cuts with feedrate changes were made for validation. The measured and
simulation resultant force for a step change in feedrate from 0.05mm/tooth to 2 mm/tooth
is presented in Figure 5.
The experimental results correlate well with model results in terms of average and peak
force. The obvious discrepancy may be due to inaccuracies in the neural force model, and
unmodeled system dynamics.

4.1 SIMULATOR OF CUTTING DYNAMICS

To realise the on-line modelling of cutting process, a standard BP neural network
(UNM) is used based on the popular back propagation learning rule. During preliminary
experiments it proved to be sufficiently capable of extracting the force dynamics model
directly from experimental machining data. It is used to simulate the dynamics of cutting
process. The UNM for modelling needs eight input neurons: for federate (f), cutting speed
(vc), radial and axial depth of cut (AD / RD), type of machined material, hardness of the
machined material, cutting tool diameter (D), and tool geometry. The ANN registers the
input data only in the numerical form therefore the information about the tool, cutting
geometry and material must be transformed into numerical code. The geometry of the
cutter is indicated with an 8-digit systematization code containing the data on the cutting
edge shape, rake angle, free angle, tip radius, base material, cutting coating and length of
the cutting edge. The output from the UNM are cutting force components, therefore three
output neurons are necessary. For simplification of the milling simulator the neural
network is so adapted that during prediction overlooks all input parameters except feeding.
During simulation most input vector parameters do not change (e.g. cutter diameter and
geometry, material etc.). The detailed topology of the used NN with optimal training
parameters is shown in Figure 6. Optimal UNM configuration contains 5, 3 and 7 neurons
in hidden layers.

9
 UNM
α β
U enje UNM / Training UNM 1 0.07 0.35
2 2.1 0.4
Skriti nivoji 3 0.09 0.37
Hidden layers
f
vc

Vhodni vektor / Input vector

Fx
Da/Yes

Dejansko podajanje
AD

Actual feedrate
Fy
U enje Ne RD
Start Geom. orodja Fz
Training? No
Tool geometry

Objective values
Ciljne vrednosti
D

Material

HB

Priprava podatkov za u enje / testiranje

Preparation of data for training / testing

U ni niz-vhodni vektor
(720)
Training set-input
vector
Na rt eksperimentov Proces frezanja Izmerjene rezalne sile
Experiment plan Milling process Measured cutting forcess

Fig. 6. Predictive cutting force model topology

5. DATA ACQUISITION SYSTEM AND EXPERIMENTAL EQUIPMENT

The data acquisition equipment consists of dynamometer, fixture and software module
as shown in Figure 7. The cutting forces were measured with a piezoelectric dynamometer
(Kistler 9255) mounted between the workpiece and the machining table. When the tool is
cutting the workpiece, the force is applied to the dynamometer through the workpiece. The
piezoelectric quartz in the dynamometer will be strained and an electric charge will be
generated. The electric charge is then transmitted to the multi-channel charge amplifier
through the connecting cable. The charge is then amplified using the multi-channel charge
amplifier
(Kistler 5019A). In the charge amplifier, different parameters can be adjusted so that the
required resolution can be achieved. Essentially, at the output of the amplifier, the voltage
will correspond to the force depending on the parameters set in the charge amplifier. The
interface hardware module consists of a connecting plan block, analogue signal
conditioning modules and a 16 channel A/D interface board (PC-MIO-16E-4). In the A/D
board, the analogue signal will be transformed into a digital signal so that the LabVIEW

10
software is able to read and receive the data. The voltages are then converted into forces in
X, Y and Z directions using the LabVIEW program. With this program, the three axis force
components can be obtained simultaneously, and can be displayed on the screen for further
analysis. The ball-end milling cutter with interchangeable cutting inserts of type R216-
16B20-040 with two cutting edges, of 16 mm diameter and 10° helix angle was selected
for machining. The cutting inserts R216-16 03 M-M with 12° rake angle were selected.
The cutting insert material is P10-20 coated with TiC/TiN, designated GC 1025.
Communication between the control system and the CNC machine controller is
accomplished over RS-232 protocol. The feedrate override percentage variable DNCFRO
is available to the control system at a frequency of 1 kHz.

CNC f A/D pretvornik

FAGOR
A/D converter
8040-M
Oja evalec / Amplifier
AMPLIFIER

PCO-MIO-16E-4
Merilnik sil / Dynamometer
Ponastavljeno podajanje
Fx,Fy
Adjusted feedrate Fref

RS-232 LabVIEW

DATA AQUISITION
Data aquisition
PC

Fig. 7. The experimental equipment

6. EXPERIMENTAL TESTING OF COMBINED ADAPTIVE CONTROL SYSTEM

To examine the stability and robustness of the proposed control strategy, the system is
first analysed by simulations using LabVIEW´s simulation package Simulink [12]. Then
the system is verified by two experiments on a CNC milling machine (type HELLER
BEA1) for Ck 45 and 16MnCrSi5 XM steel workpieces with variation of axial cutting
depth (Experiment 1- prismatic workpiece; experiment 2- workpiece with irregular profile,
see Figure 8, 9). Feedrates for each cut are first optimized off-line, and then machining
runs are made with controller action. We conducted two main series of tests, in which two
different shaped workpieces were machined. Details of the experimental conditions and the
dimensions of the workpiece are shown in Fig. 10.

11
a)
Eksperiment 1/Experiment 1:
Prizmati ni obdelovanec/ 40

5.8
Prismatic Workpiece
45
40

11

25
5
16
0

60

Test_A Rezalni pogoji /Cutting conditions:

(Konstantno podajanje / Podajanje / Feedrate: 0.08mm/zob
Constant feedrate) Rezalna hitrost /Cutting speed: v=80m/min
Predprogramirana aksialnma globina reza /
Pre-programmed axial depth of cut AD=2 mm
Radialna globina reza RD =4mm
Fref=280N
Rezultat /Result: Slika / Figure: 11a
Test_B Startno podajanje / Starting feedrate: 0.08mm/zob
(Predlagan kombiniran sistem Dovoljeno nastavljivo obmo je /Allowable adjusting rate
vodenja / Proposed combined 00.8 - 0.20 mm/zob
control system) Rezalna hitrost /Cutting speed: v=80m/min
Aksialna globina reza / Axial depth of cut AD=2-11mm
Radialna globina reza RD =4mm
Fref=280N
Rezultat /Result: Slika / Figure: 11b

b)
Eksperiment 2/Experiment 2:
Nepravilni profil obdelovanca S me
r f re
/ Irregular workpiece profile za n
ja (m
illing
dire
ctio
A D =5 mm n)

A D =3 mm

A D =1.5 mm
A D =3 mm
A D =5 mm

Test_A Rezalni pogoji /Cutting conditions:

(Konstantno podajanje / Podajanje / Feedrate: 3mm/s
Constant feedrate) Vrtljaji vretena / Spindle speed: 2400min-1
Aksialna globina reza / Axial depth of cut AD=2 -5mm
Radialna globina reza RD =16mm
Fref=650N
Test_B Startno podajanje / Starting Feedrate: 2.5mm/s
(Predlagan kombiniran sistem Dovoljeno nastavljivo obmo je podajanja /Allowable adjusting rate of
vodenja / Proposed combined federate: 2.5-11mm/s
control system) Vrtljaji vretena / Spindle speed: 2400min-1
Aksialna globina reza / Axial depth of cut AD=2 -5mm
Radialna globina reza RD =16mm
Fref=650N
Rezultat /Result: Slika / Figure: 12

12
 Fig. 10. Na rt eksperoimentov; a)Rezalni pogoji za prizmati ni obdelovanec. b)Rezalni pogoji za
nepravilni profil obdelovanca / Plan of experiments; a) Cutting conditions for prismatic workpiece.
b) Cutting conditions for irregular workpiece profile.

The first test is conventional cutting with the constant feedrate (Test_A). In the second
test, the proposed combined system was applied in milling to demonstrate its performance
(Test_B).
The ball-end milling cutter (R216-16B20-040) with two cutting edges, of 16 mm
diameter and 10° helix angle was selected for experiments. Cutting conditions are: milling
width RD=4 mm, starting milling depth AD=2 mm and cutting speed vc=80 m/min. The
parameters for adaptive control are the same as for the experiments in the conventional
milling. To use the structure of combined system on Figure 1 and to optimise the feedrate,
the desired cutting force is [Fref] =280 N, pre-programmed feed is 0.08 mm/teeth and its
allowable adjusting rate is [0 - 150%].

Fig. 8. Experiment-1; Prismatic workpiece

Sme
r frez
anja
(mil
ling
dire
ction
AD =5 mm )

AD =3 mm

AD =1.5 mm
AD =3 mm
AD =5 mm

Fig. 9. Experiment-2; Irregular workpiece profile

13
 Figure 11 is the response of the cutting force and the feedrate when the cutting depth is
changed (experiment-1). It shows the experimental result where the feedrate is adjusted on-
line to maintain the maximal cutting force at the desired value.
7.5 7.5

5 5
MRR
[mm3/min]
*104 2.5 2.5
A B C D

0 0
0 25 50 75 100 125 150 0 25
15 50
30 75
45 100
60 125
75 150
90

300 300

200 200

F[N]
100 100

0 0
0 25 50 75 100 125 150 0 15
25 30
50 45
75 60
100 75
125 90
150

0.24 0.24

f 0.16 0.16

[mm/zob]
[mm/tooth]
0.08 0.08
A B C D

0.00 0.00
0 25 50 75 100 125 150 0 25
15 50
30 75
45 100
60 125
75 150
90
a) as / Time [s] b) as / Time [s]

Fig. 11. Experiment-1; Response of MRR, resulting cutting force and feedrate. a)
Conventional milling-Test_A. b) Milling with proposed adaptive control system-Test_B

The second experiment is machining of irregular workpiece (Figure 9) consisting of five

straight cuts with different axial and radial depths of cut. This experiment demonstrates the
ability of adaptive system to maintain the constant cutting force during machining. The
desired cutting force is 650N. A sample time is 20ms and scanning rate is 28.8 kHz. The
results of the second experiment using optimized feedrates and UNKS are presented in
Figure 12.

14
 Rezultirajo a rezalna sila / Resultant cutting force
1.25

F [kN] F ref
0.5

0 15 30 45 60 75 90 115
DNCFRO
120
Corrected feedrate [%]

115
Korigirano podajanje

110

105

100

0 15 30 45 60 75 90 115
Skupno podajanje / Common feedrate
12.5

10

7.5
f [mm/s]

2.5

0 15 30 45 60 75 90 115
as / Time [s]

Fig. 12. Experiment-2; Machining of irregular profile by off-line optimizing of cutting

conditions and adaptive adjusting of feedrate

7. RESULTS AND DISCUSSION

In the first experiment using constant feed rates (Test_A-Figure 11a) the MRR reaches
its proper value only in the last section. However, in second test (Figure 11b), machining
the same piece but using adaptive control, the average MRR achieved is much more close
to the maximal MRR. Comparing the Fig. 11a to Fig. 11b, the cutting force for the neural
control milling system is maintained at about 650N, and the feedrate of the adaptive
milling system is close to that of the conventional milling from point C to point D. From
point A to point C the feedrate of the adaptive milling system is higher than for the

15
classical CNC system, so the milling efficiency of the adaptive milling is improved. The
experimental results show that the MRR can be improved by 27%. The time analysis for
conventional and adaptive control system has been curried out. By adaptive control system
of time saving of 40% with one cut was reached. The complete machining requires 15 cuts;
thus machining of a simple workpiece is shortened for 15 minutes.
The second experiment with small and large step changes is run to test system stability
over a range of cutting conditions. The system remains stable in all experiments, with little
degradation in performance. In the second experiment, the UNKS increases the feedrates to
obtain peak forces close to 650N. The slower response of the neural control scheme is
noticeable at the beginning of cut one and three.
The combined system performance for the irregular workpiece is evaluated by the time
of cut, the maximal cutting force and normalized standard error:
n
( F (k ) − Fref ) 2
k =0

n −1
Nsn = (1)
Fref

Where n is the number of samples (20), and Fref is desired cutting force. The normalized
standard error quantifies the ability of the UNKS to regulate the reference peak force. The
results of the comparison for the Experiment-2 are tabulated in Table 1.
By adaptive control system a time saving of 24% with one cut was reached in comparison
with constant feedrate. By the use of UNKS the peak forces are reduced significantly.
The divisional controller has a quick response and is the simplest to implement. The
adjustable PI controller is slightly more complicated but also demonstrate a short response
time. The MRAC is the most complicated and the response time is considerably longer
(8%). All of the parameter adaptive controllers are able to regulate peak force and remain
stable over a large range of geometric variations.
The results reached are in accordance with the objectives of researches, according to which
the controlled cutting force must not deviate from the desired value for more than 10%.

Table 1. Primerjava performans regulatorjev pri obdelavi obdelovanca z nepravilnim

profilom / Comparison of controller performance for machining of workpiece with
irregular profile

16
 as optimiranja [s] Maksimalna
Normalizirana as
/število iteracij Željena sila rezalna sila
Nepravilni profil standardna napaka obdelave
Optimization time [kN] Desired Maximal
Irregular profile Normalized Standard Time of
[s] /number of Force cutting force
Error– Nsn cut [s]
iterations [kN]
Konvencionalno
frezanje-off-line
optimizacija
0.35 115 0.001/18 0.665 1.4
Conventional
milling-off-line
optimization
Delilni regulator
0.29 98 0.007/23 0.665 1.25
Divisional controller
P-regulator
0.2 95 0.006/22 0.665 1.1
P-controller
Nastavljiv PI-
regulator
0.19 94 0.006/20 0.665 1.0
Adjustable PI-
controller
MRAC 0.22 92 0.012/34 0.665 0.9
Kombiniran sistem
0.18 88 0.001/18 0.665 0.91
Combined system

As compared to most of the existing end milling control systems, the proposed
combined system has the following advantages: 1. the computational complexity of UNKS
does not increase much with the complexity of process; 2. the learning ability of UNKS is
more powerful than that of conventional adaptive controller; 3. UNKS has a generalisation
capability; 4. insensitive to changes in workpiece geometry, cutter geometry, and
workpiece material; 5. cost-efficient and easy to implement; and 6. mathematically
modeling-free.
The experimental results show that the milling process with the designed adaptive
controller has a high robustness, stability, and also higher machining efficiency than
standard controllers.
Current research has shown that neural control scheme has important advantages over
conventional controllers. The first advantage is that it can efficiently utilize a much larger
amount of sensory information in planning and executing a control action than an
industrial controller can. The second advantage is that a neural control scheme responds
quickly to complex sensory inputs while the executing speed of sophisticated control
algorithms in a conventional controller is severely limited.

17
8. CONCLUSION

On the basis of the cutting process modeling, off-line optimization and feed-forward
neural control scheme (UNKS) the combined system for off-line optimization and adaptive
adjustment of cutting parameters is built. This is an adaptive control system controlling the
cutting force and maintaining constant roughness of the surface being milled by digital
adaptation of cutting parameters.
Applicability of methodology of adaptive adjustment of cutting parameters is
experimentally demonstrated and tested on a 4-axis CNC milling machine Heller. The
results of the intelligent milling experiments with adaptive control strategy show that the
developed system has high robustness and global stability. Experiments have confirmed
efficiency of the adaptive control system, which is reflected in improved surface quality
and decreased tool wear. The proposed architecture for on-line determining of optimal
cutting conditions is applied to ball-end milling in this paper, but it is obvious that the
system can be extended to other machines to improve cutting efficiency.

9. REFERENCES

[1] Župerl, U., F. uš (2004) Dolo evanje zna ilnih tehnoloških in gospodarskih
parametrov med postopkom odrezovanja = A determination of the characteristic
technological and economic parameters during metal cutting. Stroj. vestn., 5(2004),
pp. 252-266.
[2] Huang, S.J., C.C. Lin (2002), A self-organising fuzzy logic controller for a
coordinate machine, Int. J. Adv. Manuf. Technol. 19(2002), pp.736-742.
[3] Tlusty, J., M.A. Elbestawi (1977) Analysis of Transients in an Adaptive Control
Servomechanism for Milling with Constant Force, Transactions of the ASME,
Journal of Engineering for Industry 99(1977), pp. 766-772.
[4] Stute, G., F.R. Goetz (1975) Adaptive Control System for Variable Gain in ACC
Systems, Proceedings of the Sixteenth International Machine Tool Design and
Research Conference, Manchester England, pp. 117-121.
[5] Tomizuka, M., J.H. Oh, D.A. Dornfeld (1983) Model Reference Adaptive Control of
the Milling Process, Proceedings of the Symposium on Manufacturing on
Manufacturing Process and Robotic Systems, New York, (1983), pp.55-63.
[6] Liu, Y., L. Zuo, C. Wang (1999) Intelligent adaptive control in milling process,
International Journal of Computer Integrated Manufacturing, 12(1999), pp. 453-
460.
[7] Župerl, U., F. uš (2003), Optimization of cutting conditions during cutting by using
neural networks, Robot. comput. integr. manuf., 19(2003),pp. 189-199.
[8] Bali , J. (2000) A new NC machine tool controller for step-by-step milling, Int. j.
adv. manuf. technol. 8(2000) pp. 399-403.
[9] uš, F., J. Bali (2003) Optimization of cutting process by GA approach. Robot.
comput. integr. manuf. 19(2003), pp. 113-121.

18
 [10] Župerl, U., F. uš (2004) Tool cutting force modeling in ball-end milling using
multilevel perceptron, J. mater. process. technol., Available online 1 June 2004.
[11] Muršec, B., F. uš, J. Bali (2000) Organization of tool supply and determination of
cutting conditions, J. mater. process. technol 100(2000), pp. 241-249.
[12] Kopa , J. (2002) Rezalne sile in njihov vpliv na gospodarnost obdelave = Cutting
forces and their influence on the economics of machining. Stroj. vestn. 3(2002), pp.
121-132.
Authors'address: Dr. Uroš Župerl, B. Sc. Eng., Prof. Dr. Franci uš, B. Sc. Eng., B. Sc. Econ., Prof. Dr.
Edo Kiker, B. Sc. Eng., Dr. Matjaž Milfelner, B. Sc. Eng.
University of Maribor, Faculty of Mechanical Engineering, Smetanova 17, 2000 Maribor
[email protected]

19

View publication stats