International Journal of Machine Tools & Manufacture 38 (1998) 783–798

A new perspective and analysis for regenerative machine

tool chatter
Nejat Olgac*, Martin Hosek
Department of Mechanical Engineering, University of Connecticut, U-139, 191 Auditorium Road, Storrs, CT 0629-
3139, USA
Received 12 October 1996; in final form 18 August 1997

Abstract
The paper contains a practical perspective on regenerative machine tool chatter. Chatter is a well known
phenomenon, occurrence of which is undesired in manufacturing. Aggressive machining conditions, in the
sense of removing more metal rapidly, usually cause chatter. In most cases, these conditions can be determ-
ined a priori to the operation. A chatter stability study and its reasoning based on root locus plot analysis
of time delayed systems is presented as a new and practical perspective in the field. At the junction of
root locus and chatter concepts an area of particular interest to the authors arises: a new method for active
vibration suppression, the Delayed Resonator. It is an active vibration absorber tuning of which is achieved
utilizing a simple time delayed feedback. The cross linking between the Delayed Resonator study and the
subject matter, machine tool chatter, is exciting to share. This is the primary motivation in pursuing this
study. One of the highlights of the work appears at the phenomenon called Dual Frequency Delayed
Resonator. This feature has been conjectured in the literature using the well known “stability lobes”, but
never discussed with detail.  1998 Elsevier Science Ltd. All rights reserved.

Nomenclature
b chip width (m)
C cutting force coefficient (N/m2)
c equivalent damping (kg/s)
F cutting force magnitude (N)

* Corresponding author.

0890-6955/98/$19.00  1998 Elsevier Science Ltd. All rights reserved.

PII: S 0 8 9 0 - 6 9 5 5 ( 9 7 ) 0 0 0 6 2 - X
784 N. Olgac, M. Hosek / International Journal of Machine Tools & Manufacture 38 (1998) 783–798

h instantaneous chip thickness (m)

h0 nominal chip thickness (m)
j imaginary unit ( ⫺ )
k equivalent stiffness (N/m)
m equivalent mass (kg)
N spindle speed (rev/s)
n number of full waves ( ⫺ )
y offset chip thickness (m)
␤ cutting force angle (deg)
␶ period between two cuts (sec)

1. Introduction

Machine tool chatter has long been studied as an interesting phenomenon. Starting with the
early works of [6], [2], [22], and [21], many researchers addressed the issues of modeling the
dynamic progression, structural reasoning and stability limit aspects of this seemingly straightfor-
ward and very common behavior. Further research focused on the particulars of selecting oper-
ational parameters in machining to avoid the build up of these undesired oscillations and analytical
observations on the progression of chatter. Some interesting readings in this group are refs [7],
[4], [20], [24] and [19].
While the fundamentals of the chatter phenomenon were established in the years 2–3 decades
ago, numerous practical issues remained unsolved, even to date. For instance, the timely obser-
vation (possibly prediction) of the arising chatter, remedial control strategies to prevent chatter
are two example topics of active research, [1], [23], [8], [16].
In the present work, the problem is treated from a point of analytical abstraction and generaliz-
ation rather than process specific characteristics. The abstraction is designed to isolate the underly-
ing mechanism yielding the machining chatter. We explain in the following sections the nature
of this generalization and the reasons for authors’ motivation in writing this paper.
As accepted in the community, there are two groups of machine tool chatter: regenerative and
nonregenerative. Regenerative chatter occurs due to the undulations on the earlier cut surface,
and nonregenerative chatter has to do with mode coupling among the existing modal oscillations.
A clear description of the procedure leading to both chatter types is given in [19], for reference.
Since this text evolves primarily on regenerative type, we refer to it simply as “chatter” for the
rest of the document. However, the treatise here is directed to a broader genre of problems than
the chatter alone. More specifically, the study focuses on the fundamental issues of oscillatory
properties of the dynamics which are governed by time delayed differential equations. The connec-
tion of chatter to this general class of differential equations is explained below.
The underlying mechanism for regenerative chatter is quite simple to state. Let us consider the
turning process for the sake of argument. Typically, the tool removes part of the chip from the
surface which is already cut during the previous pass, such as in orthogonal cutting. Since the
cutting force can realistically be assumed proportional to the chip thickness, it carries the signature
of y(t)–y(t ⫺ ␶), where y(t) is the fluctuating part of the depth of cut (so called “offset chip
thickness”) at time t while ␶ is the period of successive passages of the tool. See Fig. 1.
 N. Olgac, M. Hosek / International Journal of Machine Tools & Manufacture 38 (1998) 783–798 785

Fig. 1. Orthogonal cutting.

In this study, we purposely consider the orthogonal cutting which is probably the most chatter
prone cutting mechanism there is. In orthogonal cutting, the chip overlapping factor ␮ is taken
as 1, indicating that the present chip width fully coincides with that of previously cut section.
This simplifies the analysis without sacrificing the generalization of the findings. Otherwise an
overlapping factor 0 ⱕ ␮ ⱕ 1 appears, which only complicates the mathematical expressions, as
shown in [7], and unduly distracts the direction we wish to pursue here.
A classical representation of the chatter dynamics is given by the block diagram in Fig. 2, as
it is widely adopted within the machine tool community. Nominal chip thickness, h0, is disturbed
by the undulating offset chip thickness, y. These undulations create driving forces for the y dynam-

Fig. 2. Block diagram.

786 N. Olgac, M. Hosek / International Journal of Machine Tools & Manufacture 38 (1998) 783–798

ics ␶ sec later during the next passage. G(s) is taken to be the transfer function between the cutting
force, F, and y. For the sake of streamlining the presentation, a single-degree-of-freedom cutting
dynamics is taken into account instead of higher-degree-of-freedom models such as of [19]. The
workpiece is considered to be fixed centered and the only flexibility is in the tool along the radial
(or feed) direction. See Fig. 1.
The dynamic structure represented in Fig. 2 would be at equilibrium if y = 0. That is, if the
cut were of uniform-thickness the undulations should not appear. The cutting force, F, remains
constant and the tool support structure (i.e., k, c) yields a fixed deflection throughout the cutting.
This equilibrium is called “stable” or “asymptotically stable”, if y approaches to zero asymptoti-
cally in response to a small disturbance which may be caused by a bump or dent. In order to assess
the stability question the characteristic equation of the control loop in Fig. 2 should be studied:
1 + (1 ⫺ e⫺␶s)bCG(s) = 0 (1)

Assuming that the force-displacement transfer function G(s) is linear, the entire cutting mech-
anism described by Fig. 1 is also linear. However, its characteristic equation is transcendental
due to the delay element e⫺␶s. Consequently, it has infinitely many finite characteristic roots, all
of which have to be on the left half complex plane in order for the system to be asymptotically
stable ([14,15]). Most controls textbooks treat cases where the system characteristic equation has
a fixed transport delay. It is clear that the variations in ␶ and b could influence the stability nature
of this system considerably. The “distribution of the characteristic poles” vary, meaning for a
particular set of ␶ and b the system can be stable, marginally stable or unstable.
Two important points are worthy of stressing. First, both of the parameters mentioned, ␶, which
results from the spindle speed, and b, the chip width, are under the control of the machinist. The
other elements, i.e. C, G(s), represent the existing machine tool conditions, and they are jointly
considered to be the structure. Note that the open loop transfer function G(s) manifests a stable
dynamic behavior when there is no regenerative feedback or delay effect. This feature can be
observed experimentally by tapping at the tool tip (i.e., a pulse input) and watching it dissipate
the pulse energy and come back to equilibrium.
Secondly, and as a consequence of the first remark, certain selections of ␶ and b can introduce
marginal stability to the system in Eq. (1). At these particular operating points the characteristic
Eq. (1) possesses a dominant pair of imaginary roots, ± ␻i. Then the entire structure mimics a
spring-mass duo (i.e., a conservative system) resonating at the respective frequency of marginal
stability, ␻, provided that all the remaining characteristic roots are stable. That is, the well-known
Q factor becomes very large at these frequencies. This is very similar to the core idea of a recently
introduced technique in active vibration absorption, the Delayed Resonator ([11,12]). A brief
summary of this feedback control technique is given in Appendix A for the interested reader. The
stability issue is revisited from this different angle, and ultimately well known “stability lobes”
in machine tool chatter are reached. They are shown to be the generators of a new phenomenon
called dual frequency resonance ([13]) which is also described in the Appendix B.
It is this exciting linking which motivated the authors to prepare this text, who are hoping to
share their findings with the machine tool community. As explained later, the experimental study
of DR is much easier to control than the machine tool chatter. Additionally, it is immune from
high number of process parameters encountered in machining, such as tool wear, segregated metal
 N. Olgac, M. Hosek / International Journal of Machine Tools & Manufacture 38 (1998) 783–798 787

hardness. Therefore some of the salient features of chatter can readily be demonstrated using the
Delayed Resonator setting instead. The objective here, is to transform the expertise of the authors
on Delayed Resonator vibration absorption methodology into the field of chatter.
The formation of the paper is as follows: Section 2 reviews the differential equations governing
the regenerative chatter dynamics. In Section 3 root locus perspective is presented. Section 4
underlines a novel issue, dual resonance in machine tool chatter, and elaborates on its analytical
implications. Section 5 concludes the work.

2. System assumptions and the characteristic equation

This section is devoted to a brief description of the fundamentals of chatter dynamics just to
establish a basis for further discussions. A simplistic chatter model is considered for an orthogonal
turning process following [19], see Fig. 1. Tool-workpiece interface is taken as a one-degree-of-
freedom structure, in which a planar cutting force induces a relative tool motion with respect to
the workpiece. Tool and its holder are restricted to move only in the horizontal direction. This
is taken as the only non-rigid component in the structure, for simplicity.
For this system the governing dynamic equation is
mÿ(t) + cy·(t) + ky(t) = F(t)cos ␤ (2)
where the offset chip thickness y(t) represents the waviness of the cut surface, m, c, and k are
the equivalent mass, damping and stiffness properties of the tool and tool holder, respectively. F
is the cutting force and ␤ is its angle which is assumed constant. The cutting force is taken to
be proportional to the chip area bh(t)
F(t) = Cbh(t) (3)
with coefficient C as the proportionality constant. The chip thickness conceivably decreases as
the tool elastically deforms during cutting. Furthermore, if there is an undesired lump or dent left
over from the previous pass on the workpiece it influences the present chip thickness. That is,
h(t) = h0(t) ⫺ y(t) + y(t ⫺ ␶) (4)
where h0(t) is the nominal chip thickness resulting from the feed mechanism, and ␶ is the period
of one revolution for turning processes in concern. Combining the Eqs (2)–(4), and applying
Laplace transformation yield
ms2y(s) + csy(s) + ky(s) = Cb[h0(s) ⫺ y(s) + y(s)e⫺␶s]cos ␤ (5)

An alternate form of this equation is written as

y(s) bCG(s)
= (6)
h0(s) 1 + bCG(s)(1 ⫺ e⫺␶s)
cos ␤
where G(s) = . The nominal and the offset chip thickness of the cut surface (ho and
ms + cs + k
2
788 N. Olgac, M. Hosek / International Journal of Machine Tools & Manufacture 38 (1998) 783–798

y) are considered as input and output quantities for Eq. (6), respectively. This equation shows a
regenerative character due to the presence of the terms y(t) and y(t ⫺ ␶) which may eventually
yield a phenomenon called “regenerative chatter”. The formation of this phenomenon is analyzed
in the following section.

3. Root locus plot observations and regenerative chatter

Dynamics given in (6) is linear. Its characteristic equation dictating the stability properties is
cos ␤(1 ⫺ e⫺␶s)
1 + bC =0 (7)
ms2 + cs + k

This is similar to Eq. (1) with a special G(s) defined. It is transcendental and possesses infinitely
many finite characteristic roots. In order to assure a stable cutting operation, they should all be
on the left half of the complex plane.
Conventional root locus analysis is used in this section in order to assess the stability of (7).
The delay element, ␶, is fixed for the time being, which implies a constant turning speed. The
variation of the remaining parameter b 苸 (0, ⬁) is considered next to create the loci.
쐌 For b→0, root loci depart from the starting points, so called “poles”. Two of these poles are
finite at
c √4mk ⫺ c2
⫺ ± j, 4mk ⬎ c2 (8)
2m 2m
and the rest (infinitely many) are at minus infinity
⫺ ⬁ ± (2n␲/␶)j, n = 1,2,3,… (9)
Notice that these infinity roots are “dormant”, i.e. the increased order of the system due to these
roots can be ignored.
쐌 For b→⬁, root loci are at the end points (so called “zeros”). They are all on the imaginary axis
± (2n␲/␶)j, n = 1,2,3,… (10)
implying all the eigen-functions being simple harmonic. These roots represent standing oscil-
lations.
A typical root locus plot is arrived as shown in Fig. 3 for 0 ⬍ b ⬍ ⬁ utilizing a MATLAB
based root finding algorithm. To create this plot the following realistic parameters are used for
the tool holder:
m = 50 kg, c = 2 × 103 kg/s, k = 2 × 107 N/m
C = 2 × 10 N/m ,
9 2
␤ = 70 deg, ␶ = 5 × 10⫺3 s

As stated above, there exist infinitely many characteristic roots, forming a cloud, for any given
b. Notice that this cloud is not dense, i.e. the roots are not everywhere. Consequently all the
 N. Olgac, M. Hosek / International Journal of Machine Tools & Manufacture 38 (1998) 783–798 789

Fig. 3. A typical root locus plot.

branches of root loci are distinct. As b increases starting from zero, the loci first move towards
the unstable right half of the complex plane. Then they all approach to their respective zeros on
the imaginary axis.
The dynamics in concern is expected to be stable by design, when there is no regenerative
effect. In other words, the tool holder forms a lightly damped (dissipative) structure when there
is no cutting. This can be visualized by the root distribution for b→0. Notice that the dominant
poles are very lightly damped. As the chip width is increased such that some roots are positioned
on the imaginary axis (see Fig. 3, critical points), they will enforce marginal stability character-
istics to the system, provided that the remaining roots on the other branches of the root loci are
in the stable half plane. For this value of b the cloud of characteristic roots has two dominant
members on the imaginary axis ± ␻j which represent sustained (indefinite) oscillations as the
prevailing natural behavior. Such a root distribution is descriptive of “resonance” by definition.
That is, the entire machine tool is sensitized to oscillate indefinitely at this frequency, ␻, describing
the phenomenon called “standing chatter”.
This seemingly undesirable behavior can also be called “delayed resonance” because the reson-
ance condition is induced by the presence of the time delay in the dynamics. A recent methodology
of the authors’ group, named “Delayed Resonator (DR)”, is in fact nothing other than utilizing
this undesirable chatter phenomenon for a constructive purpose: to absorb harmonic vibrations at
the frequency of resonance. For details and the salient features of DR we refer the interested
reader to Appendix A.
Revisiting the root locus plot in Fig. 3, if b is increased beyond the critical value, the regenerat-
ive feedback will cause excessive chatter. The system becomes unstable. Because one root locus
790 N. Olgac, M. Hosek / International Journal of Machine Tools & Manufacture 38 (1998) 783–798

branch contains a root within the unstable right half plane. If b is further increased, the order of
instability deteriorates, i.e. more branches bring some additional unstable characteristic roots into
the picture. The cutting process becomes severely unstable, which can cause tool breakage, dam-
age to the machine tool or destruction of the part.
Notice that in all the above analysis a critical parameter, delay ␶ = 1/N, is kept constant. Let
us now review how influential the variation of ␶ can be in regenerative chatter development. First
of all we identify the marginal stability points in Fig. 3 in terms of b and ␶. For this s = ± ␻j is
imposed as the roots to the Eq. (7), i.e.,
[1 + bCG(s)(1 ⫺ e⫺␶s)]s = ␻j = 0 (11)

Solving the two real equations which appear from (11), the following critical parametric values
can be found

(m␻2 ⫺ k)2 + (c␻)2

冪m
k
b* = ,␻⬎ (12)
2C(m␻2 ⫺ k)cos ␤

␶* =
1
␻ 再
(2n + 1)␲ + atan
c␻
m␻ ⫺ k ⫺ b*C cos ␤
2 冎
, n = 0,1,2,…

It is clear that increasing ␶ (i.e., decreasing N) would vertically compress the root locus plot
and vice-versa. Note that, this variation in ␶ not only changes resonance frequency ␻ but also
the corresponding chip width b*. b* is a single valued function of ␻ as given in (12), but ␶* is
multivalued. The multivalued ␶* gives rise to the infinitely many root loci. The parameter n is
the identifier of the root locus branch which carries the root ± ␻j. Clearly, the desirable operating
settings b and ␶ are selected such that
b ⬍ b*(␶) (13)
in order to achieve stable cutting conditions. This proposition has been known for a long time in
the community as the rule: keep the chip width small enough to avoid chatter at a given N.
However, the perspective presented here, coming from root locus analysis and delay induced
resonance, is new. It introduces a very simple tool to secure stable cutting operation. Properly
modeled cutting process can be stabilized away from the chatter boundary.

4. Dual resonance chatter

The analysis and observations presented in the previous section were for a fixed value of n in
Eq. (12). That is, a particular branch marked by n crosses the imaginary axis for the parameters
b*, ␶* of Eq. (12). We wish to discuss the importance of the branch identifier, n, and tie it to an
interesting phenomenon called the “dual-frequency resonance”.
In Fig. 4, b*(N*) and ␻(N*) plots are given for varying values of n. This number is conceived
in the earlier literature as the number of full waves on the cut surface (referring to Fig. 1). For
 N. Olgac, M. Hosek / International Journal of Machine Tools & Manufacture 38 (1998) 783–798 791

Fig. 4. Stability lobes.

instance, n = 3 implies that there are 3 full chatter waves and a fraction before the tool completes
one revolution. In our treatment it is taken as the identifier of the crossing branch which happens
to possess the dominant imaginary root ␻j, for the operating setup.
For n = 0, i.e., the first branch crossing at ␻j, the resonance occurs at a chip width b* and a
corresponding spindle speed N*. Similar curves are also shown for n = 1,2,... branch crossings.
Notice that at a given N* (or the respective ␶*) there is only one b* at which a single branch
crosses over the imaginary axis and the crossing ± ␻j identifies the chatter frequency. As seen
in Fig. 4, for N ⬎ ND (or ␶* ⬍ ␶*D crossing occurs on the first branch (n = 0) first, i.e.,
b*兩n = 1 ⬎ b*兩n = 0. In other words, the root which is tracking the second branch crosses the imagin-
ary axis at much higher b* values (chip widths). Additionally, if this higher b* were selected,
the operation would be unstable. Therefore, for N0 the standing chatter appears on the first branch
at b*兩n = 0 chip width. The corresponding imaginary pole ␻j (not stated in the figure) dictates the
chatter frequency, and it is the dominant pole among infinitely many. For NE ⬍ ND or
␶*D ⬎ ␶* ⬎ ␶*E, however, the order is reversed, b*兩n = 1 ⬍ b*兩n = 0. That is the dominant root is now
on the second branch. Otherwise stated, the second branch crosses over before the first one does.
This argument can be easily extended for higher branch numbers the depiction of which is found
792 N. Olgac, M. Hosek / International Journal of Machine Tools & Manufacture 38 (1998) 783–798

in Fig. 4a. That is, there are certain intervals of N in which a given branch carries the dominant
pole. Ultimately this figure yields the well known stability lobes. For b ⬍ b* the cutting is always
stable, such as F1. For b ⬎ b* the chatter instability occurs, for instance at F2. The operating
point F0 represents the standing chatter, which is nothing other than the delayed resonance (of
Appendix A).
At the operating point D an interesting phenomenon appears: root loci branches 1 and 2 cross
the imaginary axis concurrently. By varying N slightly from ND ⫺ ⑀ to ND + ⑀ the branch that
carries the dominant pole changes for the same b*. For N = ND the cutting dynamics (Eq. (6))
exhibits dual resonance. This is an interesting occurrence where the tool chatter should reflect
both ␻D1 and ␻D2 frequencies jointly, see Fig. 4b. Therefore, the chatter which originates from
this situation would be a “dual frequency standing chatter”. The same phenomenon is encountered
at higher branch numbers, such as 2 and 3, 4 and 5 (see points E1 and E2). The jump in the
chatter frequency at these operating points is in full agreement with the earlier analytical studies,
such as [7], [19]. Dual resonance chatter implies, that the tool marks on the workpiece show two
distinct frequencies. To the best knowledge of the authors, this point has not been observed in
the experimental machine tool literature. It is tested in the authors’ laboratory and reported in [9]
as a practical feature. Both the setup and the test results are explained in Appendix B.
Let us analyze the mathematics behind this development. We first observe that, the b*
expression in (12) is a function of ␻ and it is a non-negative number representing the chip width.
Consequently, m␻2 ⫺ k ⬎ 0 is necessary, which implies that this dynamics can only become
resonant above the natural frequency of the tool holder assembly, or ␻ ⬎ √k/m. Typically, b*
(␻) is a convex shaped function in the operating range of ␻ ⬎ √k/m, examples of which is shown
in Fig. 4a (for various n’s). Next, the following question is posed to investigate dual resonance
chatter: is there an operating point denoted by the delay ␶* (or corresponding N*) and b* which
yields two distinct frequencies of resonance simultaneously? This question can be formulated
analytically as
␶*[b*(␻1),␻1] = ␶*[b*(␻2),␻2] (14)
b*(␻1) = b*(␻2)
From the general trends of the b*(␻) curves in Fig. 4a one can conclude that such points can
arise only at the crossings of the two successive curves corresponding to n and n + 1. If two
non-successive branches are taken for dual resonance (such as n = 0 and n = 2 at point L of Fig.
4a) there is always at least one pair of unstable roots of the characteristic Eq. (11). Therefore the
dual resonance, for instance at L, is an unstable one which cannot qualify for a standing dual
frequency chatter.
Eq. (14) can be rewritten explicitly as
c␻1
(2n + 1)␲ + atan
m␻21 ⫺ k ⫺ b*(␻1)C cos ␤
␻1
c␻2
(2n + 3)␲ + atan
m␻ ⫺ k ⫺ b*(␻2)C cos ␤
2
=
2

␻2
 N. Olgac, M. Hosek / International Journal of Machine Tools & Manufacture 38 (1998) 783–798 793

b*(␻1) = b*(␻2) (15)

Suppressing the mathematical details we can state that this pair of simultaneous nonlinear equa-
tions is numerically solvable by iterative techniques and a solution always exists.
As a side issue, it can be shown that the ␶* (or 1/N*) of Eq. (12) is monotonously decreasing
in ␻. This observation agrees with Fig. 4b and all earlier reports on stability lobes. As an interest-
ing point, it is easy to show that b*(␻) exhibits a minimum value, b*min which remains the same
for all loci branches. The conclusion which arises from this is that: any chip width b ⬍ b*min
creates a stable cutting operation regardless of the spindle speed N. This point, also, fully concurs
with the earlier studies on stability lobes. In more abstract and mathematical research reports on
the stability of time delayed systems this property is referred to as “delay independent stab-
ility” ([17,18,3]).

5. Conclusions

The paper treats the stability question of regenerative chatter dynamics in machine tools from
a new angle. It is based on the root locus analysis. Originality of the study is at the point relating
the chatter dynamics to a recent vibration absorption technique (Delayed Resonator-DR). The
results obtained in the DR vibration study, especially the experimental observations can be readily
extended to chatter. “Dual-frequency resonance” is one such feature which has not been reported
in the experimental literature until now. Commonly known chatter stability lobes and stable
operating conditions for all cutting speeds are arrived via DR analysis, and they are shown to
fully agree with the earlier findings. This work suggests that most of the observations of the DR
vibration study can be directly applied to the machine tool chatter problems, and vice versa, from
which both communities can benefit. Once again, this work offers a complementary methodology
to those which were commonly used in the chatter studies. The practical consequences of the
findings reinforce the earlier results. For instance, it is desirable to operate close to the dual
frequency points, but on the stable side (e.g. right of point F1).

Acknowledgements

This work was sponsored in part by a grant from the NSF, No. CMS-9415428, and an industrial
consortium Variable Frequency Vibration Elimination Consortium with founding members of Pratt
and Whitney and Electric Boat.

References

[1] Y. Altintas, E. Budak, Analytical Prediction of Stability Lobes in Milling, Annals of the CIRP 44 (1995) 00.
[2] S. Doi, S. Kato, Chatter Vibration of Lathe Tools, Trans. ASME 78 (1956) 1127.
794 N. Olgac, M. Hosek / International Journal of Machine Tools & Manufacture 38 (1998) 783–798

[3] E.I. Jury, M. Mansour, Comments on The Routh-Hurwitz Method for Stability Determination of Linear Differen-
tial-Difference Systems’, International Journal of Control 36 (1982) 903.
[4] R.L. Kegg, Cutting Dynamics in Machine Tool Chatter, Journal of Engineering for Industry (November)(1965)
464.
[5] B.C. Kuo, Automatic Control Systems, Prentice-Hall, Englewood Cliffs (1982).
[6] M.E. Merchant, Basic Mechanics of the Metal-Cutting Process, Journal of Applied Mechanics, 11, (Transactions
of the ASME, 66) (1944) A-168.
[7] H.E. Merritt, Theory of Self Excited Machine Tool Chatter, Journal of Engineering for Industry (November)
(1965) 447.
[8] I. Minis, T. Yanushevsky, A New Theoretical Approach for the Prediction of Machine Tool Chatter in Milling,
Trans. ASME, Journal of Engineering for Industry 115 (1993) 00.
[9] N. Olgac, H. Elmali, M. Hosek, M. Renzulli, High Frequency Implementation of Delayed Resonator Concept
Using Piezoelectric Actuators, Proceedings of ACTIVE 95, 1995 International Symposium on Active Control of
Sound and Vibration, Newport Beach, California 1995.
[10] N. Olgac, H. Elmali, S. Vijayan, Introduction to Dual-Frequency Fixed Delayed Resonator, Journal of Sound and
Vibration 189 (1996) 355.
[11] N. Olgac, B. Holm-Hansen, A Novel Active Vibration Absorption Technique: Delayed Resonator, Journal of
Sound and Vibration 176 (1994) 93.
[12] N. Olgac, B. Holm-Hansen, Design Considerations for Delayed-Resonator Vibration Absorbers, ASCE Journal
of Engineering Mechanics 121 (1995) 80.
[13] N. Olgac, Single Mass Dual Frequency Fixed Delayed resonator, U.S. patent no. 5,505,282. Issue date: April
9 1996.
[14] N. Olgac, G. Zhao, A Relative Stability Study on the Dynamics of the Turning Mechanism, Journal of Dynamics
Systems, Measurement and Control 109 (1987) 164.
[15] C.L. Phillips, R.D. Harbor, Feedback Control Systems, Prentice-Hall, Englewood Cliffs, 1988.
[16] S. Smith, J. Tlusty, Efficient Simulation Programs for Chatter in Milling, Annals of the CIRP 42 (1993) 00.
[17] A. Thowsen, The Routh-Hurwitz Method for Stability Determination of Linear Differential-Difference Systems,
International Journal of Control 33 (1981) 991.
[18] A. Thowsen, Delay-Independent Asymptotic Stability of Linear Systems, IEE Proceedings 129 (1982) 73.
[19] J. Tlusty, Machine Dynamics, in: R.I. King (Ed.), Handbook of High Speed Machining Technology, Chapman
and Hall, New York, 1985.
[20] J. Tlusty, F. Ismail, Special Aspects of Chatter in Milling, Trans. ASME, J. Mech. Design, Paper 81-DET-18, 1981.
[21] J. Tlusty, L. Spacek, M. Polacek, O. Danek, Selbsterregte Schwingungen an Werkzeugmaschinen, VEB Verlag
Technik, Berlin, 1962.
[22] S.A. Tobias, Machine Tool Vibration, Wiley, New York 1961.
[23] M. Weck, Y. Altintas, C. Beer, CAD Assisted Chatter Free NC Tool Path Generation in Milling, International
Journal of Machine Tool Design and Research 34 (1994) 00.
[24] D.B. Welbourn, J.D. Smith, Machine Tool Dynamics-an Introduction, Cambridge University Press, London, 1970.

Appendix A

The delayed resonance and its interlinking with regenerative chatter

The Delayed Resonator (DR) is a tunable frequency vibration absorber which is formed using
a simple feedback control law on a passive spring-mass-damper absorber (Fig. A1). In this method,
a Delayed Resonator, is prepared as a tunable vibration absorber to be attached on an oscillating
primary structure. Notice that the feedback utilized for this objective is of P (proportional) type
with a controlled time delay resembling the chatter formation. The governing equation of motion
for this single-degree-of-freedom (SDOF) system is
 N. Olgac, M. Hosek / International Journal of Machine Tools & Manufacture 38 (1998) 783–798 795

mẍ + cx· + kx + gx(t ⫺ ␶) = 0 (A1)

The corresponding characteristic equation of this linear ODE with time delay element is
ms2 + cs + k + ge⫺␶s = 0 (A2)

This equation as well as Eq. (1) can be represented in the general form of A(s) + B(s)e⫺␶s =
0. The coupling between the delayed resonance and regenerative chatter starts at this point. They
are both governed by linear ODEs with time delays (Eq. (1) and (A2)). In machining the delay
appears due to the regenerative process, thus it is under the control of the cutting mechanism.
On the other hand, for the DR study the delay is introduced due to the feedback control law. The
time delayed state is multiplied with bCG(s) in machine tool chatter (See Eq. (6)) where the
factor C is a function of cutting parameters (e.g. tool sharpness, hardness etc.) and it is difficult
to duplicate precisely. The DR dynamics, on the other hand, contains only the feedback gain, g,
as the multiplier, which is accurately generated by the controller. Consequently, repeatability of
the experimental tests on DR is much better than those of machining chatter. The results obtained
from these tests provide insightfull basis for chatter analysis. Instead of experimenting the chatter
phenomenon on a machine tool with a prohibitively large uncertainties and influential parameters
we pursue an analogous experiment on DR with a better handle on the dynamics. This interlinking,
indeed, is the basis of this paper.
Having explained this objective, we can now return to the highlights of the DR. Eq. (A2) is
transcendental, thus it has infinitely many characteristic roots. A focused research activity has
been going on for some time in the authors’ laboratories on this class of systems. We wish to
list some critical observations and salient features associated with the chatter.
(a) Proper selection of the feedback gain, g, and delay, ␶, can impart a pair of imaginary character-
istic roots, ± ␻j, for Eq. (A2) ([11]). These g and ␶ selections are
g = √(c␻)2 + (m␻2 ⫺ k)2 (A3)
atan 2(c␻,m␻2 ⫺ k) + 2␲ᐉ
␶= , ᐉ = 0,1,2,…
␻
as functions of ␻. If for a given set of g(␻), ␶(␻) the characteristic roots (other than s = ±
␻i) are on the stable left half of the complex plane, then the controlled structure of Fig. A1
behaves like a resonator (i.e., mass and spring) with the resonance at ␻. This is called the
Delayed Resonator, and the property is named the Delayed Resonance.
Chatter connection: it is analogous of creating a standing chatter at frequency ␻ for certain
b* and N*.
(b) It is additionally proven that this resonance frequency can (theoretically) be tuned in real time
for a semi-infinite span ␻critical ⬍ ␻ ⬍ ⬁ ([11]).
Chatter connection: the chatter frequency also varies in a large range for varying N*.
(c) ␻critical lower bound constitutes an operational point where the structure of Fig. 4 shows two
resonant modes concurrently, i.e., s1,2 = ± ␻1j, s3,4 = ± ␻2j are the roots to Eq. (A2). This is
called the Dual Frequency Delayed Resonance feature ([10]).
796 N. Olgac, M. Hosek / International Journal of Machine Tools & Manufacture 38 (1998) 783–798

Chatter connection: this issue has been analytically recognized in the past but never experi-
mented. The DR setting enables us to undertake this analogous work. See Appendix B.
(d) Multi-valued solution of ␶(␻) in Eq. (A3) corresponds to the imaginary characteristic roots
which are on different branches of the root locus plot corresponding to Eq. (A2). Indeed,
g(␻) vs. ␶(␻) parametric plots with this multi-valued ␶ give rise to the depiction which is
widely known as the “stability lobes” in the machine tool community.
Chatter connection: obvious generation of stability lobes.
(e) Typically, the intersection of two adjacent stability lobes creates a Dual Frequency Reson-
ance operation.
Chatter connection: same as in (c) above.

Fig. A1. Delayed resonator formation.

Appendix B

The dual-frequency delayed resonance and experimental findings

As explained in Appendix A, the delay induced resonance can appear at two distinct frequencies
concurrently. This property parallels a recent study by the authors on the “single-mass dual-
frequency resonance” ([13,10]). In that work a delayed position feedback control is used on a
single-mass structure in order to force it to resonate at two distinct frequencies. This phenomenon
occurs exactly in the same fashion within the machine tool chatter. As given in Section 4, certain
settings of b* and N* yield this dual resonance (i.e., dual-frequency chatter). We again deploy
an experimental setup for the DR (instead of an actual machining environment) in order to create
this interesting behavior.
The test setup is the same as Fig. A1. The feedback control, however, is on acceleration, instead
of the displacement. The new dynamics then becomes
mẍ + cx· + kx + gẍ(t ⫺ ␶) = 0 (B5)

It is again the same type of governing equation as (1) and (A2). The reason for acceleration
 N. Olgac, M. Hosek / International Journal of Machine Tools & Manufacture 38 (1998) 783–798 797

feedback is to take advantage of accelerometer measurements at high frequencies and low ampli-
tudes.
For the particular test platform the following physical parameters are measured: ma = 0.183 kg,
ca = 70.82 kg/sec, ka = 9,611 kN/m. The control is applied in a piezo-actuator of model PCB
Piezotronics 710A01. For the feedback parameters g = 0.152 kg and ␶ = 0.557 ms the system
exhibits dual-frequency resonance as the time trace of the acceleration describes (Fig. B1). The
respective power spectrum indicates clearly the two frequencies ␻1 = 842 Hz and ␻2 = 1,776 Hz
(Fig. B2). This behavior is tantamount to the one expected at D1/D2 or E1/E2, etc. of Fig. 4b
for chatter.
As explained in Appendix A, this experimental setting is much more immune to structural
uncertainties than a machine tool test-bed. The test results arrived can effectively be transformed
to understand dual-frequency chatter occurrence, which has not been reported earlier in the experi-
mental machining literature.

Fig. B1. Time trace of the dual-frequency resonance of the DR.

798 N. Olgac, M. Hosek / International Journal of Machine Tools & Manufacture 38 (1998) 783–798

Fig. B2. Power spectrum of the dual-frequency resonance of the DR.