I

\
.. , .. ,.. . , rvt r

80-WA/PROD-12
n.E AMERICAN SOCJETY OF MECHANICAL ENGINEERS
345 E 47 St., New York, N.Y. 10017
The Socoety shall no1 be respOnsible lor statement$ or opinions advanced in papers or
n a.scusson at meeti"9S of the Society or ol Its Oivrsions or SeciiOr>$. Ot pointed .n
its OUDitcations. o;scussJOn is printed only il ,,.. peper is publiShed in ASME
.J<Umal or lOt gene<al puoticat10n upon presentation.Foll
C/Dt should be given to ASME. the TecM1cal Division. and the auth0f1SI.
f\SME:
1
0 cf E-.j f.,, lt.<rJ ....
\/c\ 10'-.J rio VV l'1f - l?b
Control Optimization of

Graduate StuclenL
S. Mal kin
Auoe. PrOiesso<,
Mem.ASME.
(irinding
Y. Koren
Senior lecturr.
Material Processl119 and Mact..lne Tool
Faculty of Mechanical Eng-1119.
Technion-lscaellnslltute of Techf101o9y,
Haifa. Israel
.4 computerized adaptive control grinding system has been developed. The system is de-
signed to optimize both the grinding and dressing conditions for maximum removal rate
su.bjec: to constraints on ;.::orkpiece burn and surface finish. The control strategy of the
system is based upon online convergence along a predet ermined optimal trajectory de
ri,;ed from grinding theory. A pilot system uas developed and illustrative re.<ults are pre
sented u:hich demonstratf its performance and the practical feasibility of the optimiza
tion concept.
Introduction
Adaptive control (AC) as applied tO machine tool systems refers
to control of the operating parameters in reference to measurements
of the process characteristics in order to operate the machining process
in a desired condition. These AC systems can be classified as either
Adaptive Control Constraint (ACC) or Adaptive Control Optimization
(ACO), depending on the nature of the "desired condition -ll-3]. With
ACC, the "desired condition" is simply specified by fixed constraints
on the control system or machining process. With ACO. the "desired
condition" is to maximize an index of performance subject to con-
straints of the control system or machining process. In principle, an
ACO system should provide better performance than an ACC
system.
Although there has been considerable research on ACO systems,
few. if any, of these systems are used in practice. The main difficulties
encountered with such systems have been mainly in specifying an
index of performance together with an appropriate C'Ontrol policy
which is not too complex. and in developing sensors which can reliably
measure the necessary process parameters in a prooduction en\'iron
ment. Practically all the AC machining systems used today in pro-
duction are of the simpler ACC type and seldom invohe the control
of more than one operating parameter. This genetal situation with
AC machine tool systems appears to be especially true for grinding
12]. A number of ACO grinding systems have been de,eloped 12-8),
but have not been in!plemented industrially. at least to any significant
extent. Controlled-force grinding systems 19) in use are of the ACC
type.
In the present paper, a computerized ACO system is described for
plunge grinding of steels. The objective of the system is to maximize
removal rate subject tO constraints on surface finish and workpiece
bum. Both the grinding and dressing parameters are: controlled based
upon on-line sensing of the grinding power and off-l.ine measurement
of surface roughness. On-line convergence to the optimal conditions
proceeds along a predetermined optimization traje<:t:Ory derived from
grinding theory. This same optimization strategy has been used for
Contributed by the Production Ensineering Division for pr_,t.ation at the
Winter Annual Meeting, ChicaJ(o. Ill., November 16- 21, 1980 o! the American
Society of Mechanical Engineers. Manuscript received at ASME headquarters
July 1980. Paper No. 80-WA/PROD-12.
ofi-line grinding and dressing optimization 110]. The development
of a pilot computerized ACO grinding system is described and results
presented which illustrate its operation.
Grinding Model
Tne optimizatiun for the ACO system is based mainly upon
tht- grinding modt- l of Malkin lll-13] for plunge grinding such as il-
lustrated in Fig. i for the particular case of external cylindrical
grinding. The aspects of the grinding model which are
summarized in ,;ection include the partition of the grinding power
among its fundamental components, the prediction of the critical
grioding power for the burning constraint, the dependence of surface
on process parameters, and the influence cf dressing on grinding
performance.
Grinding Power. The total grinding power can be considered to
coruist of chip formation. plowing. and sliding components:
(1 )
Each power component (\.\'atts) can be expressed in terms of the
operating parameters as follows:
0
WHEEL
Fig. 1 lllustratlon of elrtemal cylindrical plu119e grlndlno
1
and
Prh = 13.8 (-;rd,.t:tbl
Ppt = 0.96 bu,,
(2)
(;'!)
P.,/ = !Ct + C27rnudu.ft,d.)d,
112
ut
112
nu.-t/'!bA ( 4)
where d.,. is t he wor kpiece diameter lmml. b is the grindinl( width
(mm). t; is the radial infeed velocity (mm/sl. t, is the peripheral wheel
"elocity (m/s). ""' is the rotational speed of the workpiece ( rev/sl, and
d,. is the equivalent diameter (mm) which for external cylindrical
grinding is given by:
d _ d.,du
- d$ +d.,.
where d . is the wheel diameter. The expression for P,.h in eq. :l is based
on a constant specitic chip formation energy of 1:1 .. J/mm'
1
which is
generally valid for grinding of both plain carbon and alloy steels of
various compositions and heat treatment. The quantity within the
parentheses is the vol umetric removal rate. The plowing component
Ppt in equation (3) is based on a constant tangential plowing force per
unit width which is also relatively insensitive to the particular steel
being ground. The sliding component in equatinn ( -I I is due to rubbing
of the wear flats on the abrasive grains against the workpiece and. as
such. is proportional to the fraction A of the wheel surface consisting
of dulled flat area. The quantity within the parentheses in equation
( -I ) is the frictional shear stress between the wear tlats and the work-
piece which includes a factor C 1 plus an additicnal term proportional
to C 2 arising from the curvature difference between the wheel and the
workpiece. For a particular case of grinding on AISI 1090 hot rolled
steel. the constants are C
1
= 7.55 x tQ-
3
and C
2
= 2.10 x tQ-
3
[lll:
these constants will scale up or down accordingly depending on the
particular wheel-workpiece combination. With these particular values
of C
1
and C
2
, the wheel can be considered to have an "effective wear
flat area" which may be larger or smaller than the actual wear flat area.
Therefore. the expression for P,, with these C
1
and C2 values can be
taken as generally valid for grinding of different steels when A is in-
terpreted as an "effective" rather than an "actual" wear flat area.
Grinding Burn Constraint. One main limitation to the removal
rate for grinding of steels is workpiece burn. On the basis of a heat
transfer analysis and experimental measurements. it has been shown
that burning occurs when a critical grinding zone temperature is
reached [121. The corresponding threshold grinding power (Watts)
for workpiece burn can be written in terms of the grinding parameters
as:
Pb = 19.5 d..,utb + 12.8 bd,
1
'
4
ut
114
n.w "
4
d,..
1
12 (5)
with the grinding parameters in the same units as in equations (2-4).
This burning prediction has been shown to be for both straight
surface and external cylindrical grinding is in the form of a power
function relationship between the ratio of voiUI!netrical removal rate
per unit width Zl b to wheel velocity u, :
(6)
or since Z = -;rdu.Ufb
R. = Ro ( (7)
with the constants R
0
and x evaluated to the surface finish
for a given amount of material removal after dressing. The exponent 1
x typically ranges from 0.4 to 0.6, although both Ro and x are reduced
by extended sparkout at the end of the grinding cycle before disen-
gaging the wheel from the workpiece.
Dressing. The manner in which the wheel is dressed can greatly
influence grinding performance [ 15. 161. With finer dressi ng (e.g .. liner
dressing lead or dressing depth with a single point dresser I, the wheel
is duller Oarger A l thereby raising the grinding power in equation (2),
but the surface finish is better. This tradeoff between grinding power
and surface finish indicates that there should be an optimum dressing
2
condition for a maximum removal rate subject to burning and surface
finish constraints. For si ngle point dressing, whi ch is used in t he
present study, the surface finish has been found to have a power
function dependence on the dres.<>ing lead !d and dressing depth ad
[161. which when combined with equation (6) gives:
Ra = R,, adY fd : - . ( z )'
bv,
(8)
where R;, is a constant. Typical values for the dressing parameter
exponents are .v = 0.25 and z = 0.60. which indicates that the dressing
lead has a much bigger int1uence than the dressi ng depth on $Urface
finish. A direct relationship has also been found between the effecti,e
wear tlat area and these dressing parameters [ l il. althou!(h the need
to use such a predictive equat ion is avoided with the present grinding
optimization syst.em.
Optimization Strategy
The performance index of the grinding process is t he volumet ric
removal rate which for external cyclindrical grinding can he
written:
(9 )
The objective of the oprimization is to maximize Z subject to the
constraints of workpiece burn and surface finish requirements. For
any particular opti mization. the workpiece diameter d., is nearly
constant and the grinding width b is fixed. so the performance index
can be simply taken as the radial infeed velocity Uf. The optimization
problem therefore can be written:
Maximize Uf
Subject to: P S Pb
Ra S Rax
! !OJ
In the present ca>< hoth the wheel velocity and the wheel diameter
are also considered ' " be constant. The variable grinding parameters
are l!f and n., .. whi,h are the parameters directly controlled on the
machine tool, although a different pair of parameters (e.g .. l., and a
in Fig. ll could al"" be selected. Accordingly, eqs. (1-51 have been
written in terms of l'f and nw. The variable dressi ng parameters are
{d and ad.
Consider fi rst a simplified optimization problem with fixed dressing
where the objective is to maximize Uf subject only to the burning
constraint without any surface finish constraint. In this case. the
equality condition on t he power constrai nt applies and the opti mi-
zation problem can he written:
Maximize u1
Subject to: P = Pb
( 111
From equations (I .'>I. it can be shown [181 that for a specified wear
flat area A there is a particular combination of Uf and n". which will
satisfy equation (ll l, and this is the optimal working point (Uf*, nu.l
for a given equivalent diameter. wheel diameter, and wheelspeed. The
collection of all the optimal points for various wear flat areas defines
an optimal locus in the L'f- n . plane [til. Several examples of optimal
loci are shown in Fig. :2 for a number of equivalent diameters with l'.
= 30m/sand d, = -140 mm. which correspond to wheel velocity and
wheel diameter of the present grinding system. Any point on an op
timallocus is the optimal operating point for a particular wear flat
area. optimal points further out along the locus at larger removal rates
correspond in to sharper wheels (smaller effective wear flat areal. The
comp ete erivation of t he optimal locus relationship is presented in
reference [1 il.
The optimal locus provides a convergence path to the optimal
working point provided that the gri nding power can be measured on
line [ 1 i l. At the opti mal working point. the equality P = Pb prevails.
Starting from some initial point on the optimal locus where P <Ph.
the optimal working point can be reached by proceeding out along the
optimal locus to faster infeed velocities until the measured power is
equal to the corresponding burning power given by eq. !51. Likewise.
Transactions of the ASME
(
(
0
> 0
"f
0
wm
s 30 tft/s
ds 440 mm
2 3 4
" (rev/s)
Fig. 2 Examples ol opt lmalloei in U.. r - n,. pi-
Opti ma l
locus
5 6
Fi;. 3 llluS1ralion of optimal loeus wHh machW. tool limitations irl ., - n.
pla ne
if P > Pb the optimal working point can be rt-ached by backing off
alon!( the optimal locus until P = Pb.
The surface finish constraint can now also be induded in the op-
timiz.ation problem together with the burning constraint according
t.O the optimization problem in equation ( 10). As the removal rate is
:ncreased by proceeding out along the optimal locus as described
above. the surface finish becomes rougher (!'f:e eq1uation (7)). The
surface fi nish constraint may be reached prior 1o t he burning limit
= Rax and P < Pb) or vice versa (P = Pb and Ra < Razl. For fixed
dressing these would be the two possible type$ of solution t.o the op-
timiz.ation problem in equation {10). It can be shown that further
improvements in grinding performance can be obtained by altering
the dressing parameters [ 1 il. In the fi rst case with a tight surface
finish constraint. a faster removal rate can be obunined further out
along the optimal locus without violating the constraints with finer
wheel dressing. Likewise, if the burning constraint is tight, the
dressing should be coarser. An optimal dressing condition (/d . ad )
is obtained when both the surface finish and burni ng constrai nts are
simultaneously active at some point on the optimatllocus, which in
turn corresponds t.o the optimal grinding point I of , ).
In addition t.O these surface finish and burning constraints, other
constraints can a rise due to machine tool limitations as illustrated
schemat ically in Fig. 3 t.ogether wit h an optimal locus. These con-
straints include an upper limit on the infeed veiocity (l'f ). upper and
Journal of Engineering f or Industry
p
vf
GRINDING
COMPUTER
n.,., MACHINE
Ro Raa
r
;
Ro
OPERATOR
Dressing Dress in
Fig. 4 Optimal aG.ptive control concepl
lower limits on the workpiece spindle speed and n,.., ), and
maximum available grinding power CPx ). Taking these additional
factors into account, the same optimization strategy described above
stiU applies provided that a "modified" optimal locus ABCD in Fig.
3 is used. In principle. the section DE should also be included as a
continuation of this modified optimal locus. since proceeding from
D towards E results in increased r;i, but this additional performance
is practically negligible. Another situation that can arise is that point
D falls below point C on then..,, constraint in Fig. 3 in which case r;
1
cannot exceed the point where the P, line intersects the cune BC.
T he point D might also intersect the nu:x constraint above Ufx in which
the P, constraint i,: never reached. The point C might also lie above
Ufz in which case t-it her the L'fx or P, constraint would limit t he per-
formance. the pre,ailing constraint being the one which intersects
BC at a lower radial in feed velocity.
Design of Adapti ve Control System
The grinding and dressing optimization strategy described in the
previous section provides a practical basis for an ACO system. The
overall concept for such a system is shown in F!g. 4 which includes a
grinding machine, computer. and operator interconnected. Mea-
!'urements of the j!rinding power Pare fed from the 11rinding machine
to the computer. and the computer in turn controls the pa-
rameters l 'f and n, to operate along the optimal locus (or modified
uptimallocusl and converge towards the optimal operating point. The
required surface fini$h limit Rex and periodic surface finish mea-
surements Ra are input by the operator to the computer. and the
computer in turn sug{!ests new dressing conditions to the operator
which he sets on the j!rinding machine. Control of the dressing could
also be provided directly from the computer to the grinding machine
without passi ng through the operator.
A pilot ACO system was developed t.o demonstrate this concept.
This pilot system consisted of a Kellenberger GOOR cylindrical grinder
interfaced to a computer. This grinding machine is
equipped with a 3k\\' main d rive system giving a fixed wheelspeed
t. = 30 m/s with wheels of diameter d, .. 440 mm. The two controlled
grinding parameters are workpiece spindle speed nu and the radial
in feed velocity Uf- Thf' workpiece spindle is driven by a servo-activated
DC motor in the continuous range 0.4- 63 rev/s (n.,., = 0.4 rev/sand
nu-z = 63 rev/s). For controlling the radial infeed velocity L'f a stepping
motor drive was attached to the infeed control handwheel of the
machine. A Hall element sensor (f. W. Bell PX220ZBl was connected
to the main drive motor to measure the machine power. the net
j!rinding power P heing obtained by substracting the idling power
from the measured power. The net power could also have
been obtained by directly measuring the power force component with
strain gages, but this method might not be acceptable in an industrial
environment.
3
,---- ------,
I COMPUTER I
I
I
I
I
I
I
I
I
I
I
L _ _________ _j
T
Fig. 5 adaptive control sySiem: mote< inteed drive: 21nteed control
handwh .. l: 3-grinding wn-.1 mote<: 4power sensor: SwOfkpiece spindle OC motor.
6-tacllo-generator. 7 voltage-to-trequ...cy converter
A computer program was developed for controlling the l!rindinJZ
machine. Before grinding, the constraints IR
0
x,nk.,..,nk'X Vfx. and Px)
and the fixed parameters (dw. d,, and o,) aroe input by the operator
to the computer. During the grinding operation. the on-line measured
power directly fed to the computer. and intermittent surface finish
measurements are input by the operator. (The surface finish could
also have been input on-line if a suitable t ransducer were available
to measure the surface finish while grinding.) Based upon these in-
puts. the computer provides two types of outputs: on-line control
signals proportional to the desired grinding parameters nu and I.: f. and
a message to the operator on the computer screen suggesting new
dressing parameters. Thus the system functions in a mixture of on-line
and off-line modes.
The on-line control system is illustrated in Fig. 5. The heart of the
control system is an algorithm incorporating the optimal locus opti-
mization strategy described in the previous. section. The grinding
operation might start at an arbitrary point in the Vf - nU' plane. but
is immediately transferred by the computer to a point on the optimal
locus by changing nw. Thereafter, the trajectory of convergence
towards the optimal operating point is along the opti mal locus. For
controlling the rate of convergence. instead of providing an external
reference as is typically done, the reference to the control loop Ph is
calculated according to equation 5 in the block Gb to which the control
variables Uf and n u are fed. As a consequenc<e, the convergence rate
depends on the error e = Ph - P. which converges to zero when pro-
ceeding along the locus.
From preliminary grinding experiments, the grinding power P was
found to be much more sensitive to infeed velocity Vf than to the
spindle speed nu .. Therefore. only Uf is directly determined by the
controller in Fig. 5. and the corresponding n ... is calculated on the
optimal locus. Such a single-input-single-out.put controller is much
simpler to design than a multi-output controller. The controller al-
gorithm was selected according to the equatinn:
ur(i) = L'f(i- 1) + K ,e(i) + Kz[eUJ - e(i- l)j (121
where the index i is the number of the sampling event. The first two
terms in equation ( 12) constitute an integral controller of gain K 1
which ensures a zero error te = 0) at the optimal point. The last term
is essentially a derivative controller of gain H2 which was added to
decrease the tendency for overshooting while converging towards the
optimal point. It should be noted that if the available grinding power
Px is less than the allowed power Pb. the value of P, is used in place
of P11 when calculating the error e. This situation occurred with the
pilot system due to the relatively small power Of the grinding machine
which was available.
4
The adaptive control system is of the sampled-data tvpe. Every
time period T. the measured power f> is sampled. the hurnin!( puwer
P11 is calculated using the previous vr and n.,. and the new
control parameters L'f and nu are assigned. Between samplin!( events.
the values of the control parameters are kept constant b, storing their
values in computer registers assigned to the digital-tuanalo!( cun
verters I DACJ. As in other sampled-data systems. the ;;election of the
sampling period T depends on the dynamic response of the process.
From numerous experiments of the open-loop response of the power
P to the input !'i , , .\'aS found that the grinding system behavior can
be approximateu a second order linear system with dead time
having a transfer !mction:
PCsl e-"
L'f(S) s
2
+ +
tl:l)
The range of numerical values of the constants for various wheels.
dressing conditions. workpiece materials and and wnrk
speeds were found w he:
damping fa('tor
natural frequency
dead ti me
de-gain
= 0.:35 - tl.80
Wn = 0.1 - 0.:1 rad/ s
H = 0.1 - 0.8 s
K = 10-:15
It is necessary that t he sampling frequencv { = 1/ T be much l(reater
than Wn [ 121. For the present system r = :1Hz was found to provide
satisfactory performance. This sampling frequency was used except
during rapid before the wheel engages the workpiece. Durin!(
rapid traverse, a much higher sampling frequency of 100Hz was used.
a sudden rise in the power being taken as an indicat ion of initial
wheel-workpiece contact. After sensing initial contact. a command
to slow down the infeed ,elocity is subsequently produced by the
computer within 0.1Jl second. and the main control system in Fig .. :;
becomes operative.
Having selected the type of controller and sampling frequency. the
problem then remains to specify the controller gains K 1 and K If the
gains are too small. the system will respond smoothly without over
shoot. but the convergence will be wo sluggish. If the are too
large. t he system will have a fast response. hut the damping will he
insufficient. Therefore. selection of the appropriate gains
a balance between the desirability for both fast and smooth re-
sponse.
A prelimi nary set of experiments was carried out with the pilot
system operating on the optimum locus to choose appropriate values
of K 1 and K 2. and ;orne of the results are shown in Fig. 6. In these
Transactions of the ASME
I
n
f' ,,
' I
J I
! .:
l
l I ;
; IV I
il ; l
1..
(\ ! T
, \. I.S .._
! I a..
>12
!
.Jea4.0
0 () 2D 3040200 () 20300 () 2D 0 I() 2() 0 I() 2D
To- ltl
(a)
0 I() 2D 01020 01020
r ..... {)
(b)
Fig. 6 Dynamic response of system tor .. rlous gains K, and K
2
: 32AJ8VBE whHI,
SAE 1045 stHI worl<pieee ( 1110 Btlnell), d 440 mm, d. = 165 rnm. b = 15 mm, 0. 3
mm r..noed !Tom worl<piece radius- cycle, single polnl dressing belore each cycle
whh w 10 I'm and ' = 5
experiments. the grinding performance was limited by the machine
power limit, the net a"ailable grinding power being P, = 1.5 kW. ln
Fig. 6(o ), it can be seen for Kz = 3.3Jlm/s/kW that increasing K 1 gi"es
a fa.<ter response but with undesirable overshoots at the two largest
gains. Likewise forK
1
= 1.3 Jim/s!k W in Fig. 6(b ), it can be seen that
undesirable fluctuations in L'f are obtained with the two largest values
of Kz. Based on these and numerous other experiments. the controller
gains of the pilot system were selected as K 1 = 3.3 Ji m/s/k W and K 2
5.3 JJm/s/kW. These controller gains were used except when the op-
eration approached close to the optimal operating point. as defined
in the present case as the error being within 5% of the allowable power.
At this point, a fast response is no longer needed to maintain optimal
grinding conditions, so the gains K 1 and K 2 are subst;antially reduced
and a software low-pass filter is added to the power line.
For repetitive grinding of identical parts, the adaptive control
system makes use of a learning sta.rting point. After the first part is
ground, the system selects from the optimal grinding conditions ob-
tained with that pa.rt a starting point for the next pa.rt. This starting
point could be the same as the optimum operating point reached at
the end of the pre\ious part, or it might be preferable to choose a
slightly lower infeed rate if there is significant part-to-part grinding
\'ariability. The learning starting point can be reached after rapid
tra\'erse by a constant in feed acceleration dvrldt. which the pilot
system was found to be limited by system stability to The
Journal of Engineering for Industry
="'Ct.[ ...o: I 4
1&
!,: /"'/ [
I I
. ! : ,..._.,... 1;-.--
' : ' / i
:t I I
I
C 02DXIO 10 200 10 20 0
T..,. {s)
Fig. 7 Grinding of repeiHI cycles wHh power llmH constraint: 32A46KIVBE whHI,
SAE Llo&O st .. l workpiece (240 Brlnell ), d 440 mm, d. = 165 mm, b = 32 mm. 0.3
mm remoed from workpiece radlua per cycle, single point dresatn.g only t..tore Cycle
No. 1 with .o11 c 10 I'm and ' a 5 p.m
opti=.al control system is engaged after reaching the initi21 starting
point.
Illustrative Results
l\t:merous grinding experiments were run in order to demonstrate
the of the adaptive control grinding system. and some
ilhw.:ative results are presented in this section. As mentioned abo\'e,
one serious limitation with the ACO pilot system was the relati\'ely
low power of the available grinding machine, which severely limited
the performance. For plunge grinding with parts as narrow
as b = 15 mm. the pilot system did not reach the burning limit while
operaing on the opumallocus. However, in a number of experiments.
spindle spt>ed values not on the optimal locus were deli b.
eratc!y input to the system which caused burning e\'en when
operating at less than full power. This underscores the ad\antage of
operating on the optimal locus.
So:ne results illustrating operation of the grinding system for re
petiti"e grinding cycles with only the grinding power constraint are
in Fig. 'i. For Cycle No. l. it can be seen when starting from
an initial infeed velocity = liim/s that it takes 28 seconds for the
proress to converge near to the optimal point where the power line
filte: is activated and t he gains reduced. as seen from the smoothing
of tho: t 'f curve. In the subsequent cycles. the system acceierat ed to
the l:arning starting point. and the convergence time was reduced to
abot. 4 seconds. For this series of experiments. the grinding wheel
was cressed only before the first cycle. The increase in optimum in feed
velocity which can be seen from one cycle to the next indicates self-
of the wheel (wear flat area decreases) which is accom-
pani<d by a deterioration of surface finish. For redressing of the wheel.
say a.."ier Cycle No. 10. the learning starting point for the subsequent
cycle could be selected on the basis of the optimal operating condition
reaciled in Cycle No. 1.
. .l.n example illustrating optimization of the grinding parameters
l 'f and noc subject to both power and surface finish constraints is
pre;.-.-oted in Table l. For this operation. the surface finish limit was
spec::iied as Raz = 0.3ii JJm (CLA). The same dressing was performE'd
befo:-! each optimization cycle using a single point diamond. At the
end C!i the first optimization cycle. a radial in feed velocity of 4.6Jlm/ s
at the constraint P = P,. but the surface finish Ro = 1.1
J.tm exceeded the allowable value. For the second cycle. the infeed
veloc:ty was reduced to 0.50 ills, which is the value predicted according
to equation (7) with x = 0.5 to obtain R
0
= Roz = The re-
sulti.=g surface finish of Ra "" 0.22Jlm was better t han required. so the
remo;aJ rate was thereafter increased to 0.71 ).lro/s and the re!:ultinj!
surface finish was Considering the scatter in sur:"ace finish
the optimizat ion could ha\e been stopped at this
5
Table I Grinding Optimization
Illustration
32A46K8VBE wheel, SAE 4340 steel workpiece (240 Sri nell)
d
5
440 ""' d,; 110 nn, b 32 ""' 0.2 om radial re<raval per cycl e
s;ngle point dressing before each cycle: ad IOum, fd l .Oum
Maximum allowed surface roughness Ru 0. 35ulll
-.
-
Cycle Ho. s (um/ s Ra (um)
1 4.6 1.1
2 0 . 50 0.22
3 0.71 0 .30
4 0.86 0.55
5 0. 42 0 . 32
I
6 0. 46 0.42
'
!
i
-
Table 2 Grinding and Dressing
Optimization Illustration
32A461C8V8E wheel, SAE 4340 stee 1 worknh!Ce ( 240 8r1 ne 11)
d
5
440 ""' d,; 110 ""' b 32 ""' 0.2 ""'radial r emo val per cycle
Single point dressing before each cycle
'1uloum al lowed surface roughness Ru 0.6utft
-
Cycle No.
i
, (...,/ s )
I
R
1
(ldll)
ad
(um) f d (ulll)
1
I
5.8 1.5 10 1.2
2 4. 6 0.55 10 0. 40
3 4.1 0.65 11) 0.34
4 4. 9 0.62 10 0.37
- '---
point. but three additional optimization cych!S were carried out during
which the scatter in the surface finish mea,surements became even
more apparent. These results clearly demonstrate how t he surface
finish requirement can impose a severe co:nstraint on the grinding
performance, in the present case leading to an order of magnitude
reduction in radial infeed velocity from the initial operating point at
the power limit.
For this illustrative exampl e. a faster removal rate could have been
obtained by selecting a finer dressing cond1ition. An illustrative ex
ample including optimization of both the grinding and dressing
conditions is given in Table 2. for which the surface finish limit was
specified as Rax = For this optimi2:ation, the dressing con-
dition was varied by changing the dressing: lead fd. and the radial
dressing depth remained llxed. The initial dressing condition resulted
in a rough surface finish R
0
= at the power limit. A finer
dressing condition for the next optimization' cycle was selected with
t he aid of equation (!!).and this resulted in a surface finish R
0
= 0.5n
at the power limit. slightly better than required. with only a rather
small decrease in radial infeed velocity from 5.8 to 4.6
These results clearly demonstrate the importance of optimizing the
dressing. especially when compared with optimization results in
Table I. Considerinj! the typical scatter in :surface finish measure-
ments, the opti mization could be considered to be complete at this
point. Two additional optimization cycles were run. and the results
imply that relativelv small chanf{es in the dressing are less significant
than the scatter in surface finish measurements.
In these and numerous other experiment!\ with the pilot
regenerative gri nding chat ter was never found to occur. even though
chatter is generally considered to he one of the main !(rindin!(
constraints (20(. The build-up in re)!enerative chatter is reinforced
by having both a fixed wheelspeed and fixed workspeed. \Vith the
adaptive control system. the workspeed c'mtinually changes while
moving along the optimal locus. thus avoiding a fixed relationship
6
between wheelspeed and workspeed. This should suppress the ten
clency for regenerative chatter thus an addit ional
benefit from the adaptive controll{rinding system.
Conclusion
A computerized ACO grinding was developed which opti-
mizes both :!rinding and parameters for plunge grinding of
steels. The system based upon a pract ical optimization
desi:med to maximize remo,al rate tn constraint..<, on wmkpiece
burn and surface !In ish. althou!(h other machine tool limitations are
also considered. The design and operation of a pilot ACO !( rindin)(
system was described and ,orne illustrative results were presented
which demonstrate how the con,er)!es towards the optimal
conditions. This ACO concept offers the advantage of heinl( relativelv
straightforward to implement. requiring the addition ol a small
computer. power sensnr. and interface to the basic grindi ng machine.
An exist ing CNC' grinder can he upgraded wan AC'O system usinj!
the present concept with only the addition ol a power sensor and some
additional software incorporating the optimal control st rate!(y.
Refer ences
1 Kuren. Y .. and llPnl'ri. .J. . Sumt'mal Cllntr'>l "/ Jfadl illP Ttml.<.
Khanna Puhl ihers. Ddhi. India. 1!1:..;. Ch. 9.
:! Wada. R.. and Kuctama. H .. :\ctaptl\1' Control in Hulf -Japan
S,cil't , . ,; PrP<. En c .. \ 'ul. I L :-iu. I. pp. 1-10.
:1 . Ku<'nil. \\' .. and \\"frnr. t: .. .\dapti,e Control Opumizatiun uf Hi!!h
Effkienc, External (;rindm>!-<' un<'ept. Technulul(ocal Hasi, and :\pphcati.n:
.-\nnaf.<,;{thPCI RI'.
4 lnuue. H .. Tamakuhri. K .. Sutuh. T .. :-io)luchi. H .. \\"aida. T . nnct Sata.
T .... An Adaptive Cuntwl System ul c:rondinl! Proo<c.>ss: l'ro" lntttrnau .. nu/
Cun{erqntr nn PrrJdU(' IHII'I Pt . I, -Japan Society uf Precasiun En
Toky(). pp.
:, Salje. F. .... Strate)lien Zur ProzessnptimierunJ( twim Aussenrundein-
,;techscleifen:tndu.<tru Anzeurr. \'ul. 19. pp. IU.;>i -Hl91.
6 Salje. E: .. :1>1ushardt. H .. and Scherr. E .... Rauheitsmessunj! ziir Stcuerunu
und Opt.imierunl! der -Jahr.<huth dN Sch/Pi/ . Ht.rl.
f.app- .und

'Thnith. :\USjfabe. \'ulkan.\'erlaJ(. r:.;sen. J9:il. J>ll

: Smith. H. 1.. . "Encrl(y Adapti,e GrindinK ... .-\mPriran Marh1111.,1. -lui\
pp. l t.;-ll!o
..; Oczus. K .. H ot,ch. A .. and Purzyrki. .J. ... Adaptive Cuntrol Systtm a-
the Optimizalitn :nl hc.d of c,lindrical 1
1
rtJC: Jntt'rnattul1nl
{irPrH:f l lfl Prun. :on lmnitutiun uf En:,:int-tr:-4
llndial. New Delho. 1:1;:. pp.
9 Hahn. .... l-..ntrolled Fone :\ :>:ew Technique fur P rl'dsiun
Internal Trans . ..ISM. J. ui Ind .. \'ul. !31\. pp.
-:!9:1.
111 :--lalkin. S. anct Koren Y .... Oif lint> Optimization w11h .1
:".1icro-Cumpuler: .-lnno/,, uJth ( ' /f(f'. \ 'ol. :!!). 1\)1)11. pp.
II :l>lalkin. S .. :-;p<l'i fic and in Ahmsi'"
I,,.,K Third Surth Anwrwan ,\/ttalu urh1nc Rc:otanh <'onf'rnn. Pius.hur!!h.
I' a .. May 1 pp. V:l-41iC..
12 Malkin.:'-. .. T hPrmal Part:l. Surfoce TemperaliJrl
and \\"nrkpiece Hurn: AS.\11:.' .. J .. ; t-:nMtllt'<'Tinu ;,, lndu.,tn. \' ul. \11\. W7 4
pp. 1181- 1191.
1:1 :--1alkin. S .. "llurninJ( Li mu f.,r and Cylindrical "'
Steel." Annal.< of tl..- ('f RP. \"ul. 19ill. pp.
\4 Snc>e,s. R .. l'Ptetl< .. J.. and Oecncut. .-\ .... The of Chip
(;rind .-\nnaJ., ui CfRI'. \'ol. 2:1. pp.
1;) Mal kin. S .. and tl.lurray. T .. "EITerts of R.l!ary Dressinl( on c:rindinu
Wheel l'erfurmann ... Tran>. A.'i.\tt-; . . J "' t-:nu. Jor Ind .. \'ul. ltJU. 1\171!. pp.

In Malkin. S .. :u>ci Murra_, .. T .. Compari""' uf Point and
Oressinj! uf Wheels." l'mc f tjth s .. rtlt .-\m('Tf(' (lll .\(otafw.,rkiiiC
Re.worch Cunjctrl'trc ,., .\mherst. Hf';;, pp. :!";):S-:!:):t
17 :--talkin. :> .. and Koren. Y .. "Strate)(ICS to Optimize c;rindinl! and
Parameter. fur Maximum Rmoval Hme: lu apprar.
111 :VIavne. R. \\" .. tnd :".1alkin. =-' .. .. Op;::nizaruul =>fOperating
fnr c:rinding nl A.'.Hf: .. J. ui j"r fndustn. \'ul. 9S. 1!17ti.
pp. 11111'- 10:;2.
19 Knren. Y .. ntTounputer Cuntrollir :--tanutacrurin)l Systems."
Tran.<. A.'.\11:.' . J .,j t-:nl/. jur Ind .. \' ul. 1111. ... :1. 19:9. pp .. 1:!6- :\:12.
:!U :-o;nuevs. H .. ..1nd Rrown. 0 .. Paramet('ts in Crindin;.:
Chatter." /'me-. lllth Itt! . . \/TOR Con( ..
Sept . 1969. pp.
:!1 lnsaki. 1.. C. . tnd S .. "Suppressum uf Chan r i n
Gr ind in:.:: Ru/1 . . Japan .'im. u/l
1
rvc. t:m: . \"ul.ll. :-<o. :). 197n. pp. l:l:l - 1:11!.
Pnnl4d in !;. S. A.
Transactions of the ASME