INTERNATIONAL JOURNAL OF MATHEMATICS AND COMPUTERS IN SIMULATION Volume 12, 2018

Time-delay systems with application to

mechanical engineering process
dynamics and control
André A. Keller

be used such as in physics, engineering, chemistry, biology,

Abstract— Time-delay systems with constant or variable delays economics for modeling complex real-life systems.
can take the form of delay differential equations (DDEs) from a
mathematical point of view. DDEs combine the continuous aspect of A. Formulation of Delay Differential Systems
differential equations and sample features of difference equations. We proposed here a general mathematical formulation.
Such “mixed difference equations” go back to the astronomer’s three- Delay differential equations belong to a class of functional
body problem of Condorcet in 1767. R.E. Bellman & K.L. Cooke differential equations between ordinary differential equations
published a seminal study on DDEs in the 1960’s. Russian
(ODEs) and partial differential equations (PDEs). More
mathematician Komanovskii, Myshlis & Nosov (1999) developed the
study of their stability and applications to the industrial problems. specifically, DDEs combine the continuous aspects of ODEs
Mathematical software packages like Mathematica® and MATLAB® with sample features of difference equations with advanced or
introduced solvers recently for DDEs with constant delays. DDEs are retarded delays. In the simple case of one state variable with
essential for modeling, forecasting, and simulation of complex real- constant delays, we can write
life systems for which delays cannot be neglected. Time-delay .
( () ( ))
systems are applied extensively in various domains of science and
industry such as mechanical systems, electrical systems, industrial ()
x t= f t , x t , x t − τ 1 , , x t − τ m) ( (1)

() ()
processes, and biological systems. Some delays are due to inherent
properties of a system where the propagation of effects takes time. where x t = dx t / dt . Variable delays in (1) should be
The necessity of feedback controls to stabilize a system also
( ( )) {
such as τ i φi t , x t , i ∈ 1, 2, , m . In matrix form,
introduce some inevitable delays. This study introduces this modeling = }
process and analysis using some real-life deterministic applications we may have
from mechanical engineering dynamics and control such as with the
metal rolling system and machine tool chatter.
= () ( ()
x t f t , x t , xτ t , x ∈  n ( )) (2)

Keywords—Block diagram, delay differential equation,

where f :  ×  × C , C =
n 1
C 0,τ , 
1 1
([ ] n
) . The delayed
mechanical application, transcendental characteristic equation. xτ ( t )
state vector in (2) is such that= {x ( t ) : t ≥ τ } .
I. INTRODUCTION
B. Transcendental Characteristic Equation
T HIS introduction is devoted to an initial general
formulation of delay differential equations. A standard
example illustrates the differences accounted between an
A simple comparison between ODEs and DDEs is
proposed [1] by using a simple standard example. A one-
ordinary differential equation (ODE) and a corresponding dimensional scalar linear ODE is used. The corresponding
hybrid difference-differential equation (DDE). After that, we DDE delays merely the state variable by a positive fixed time
present a short history of DDEs in the literature. Finally, we delay. Let the one-dimensional scalar linear ODE
mention the numerous scientific domains in which DDEs can
x ( t ) =
−µ x (t ) , µ ≠ 0 (3)

This work comes from a Plenary Lecture invited by AIP (American

Using (3) with a fixed delay τ >0 yields the corresponding
Institute of Physics) and presented in the 2nd Int. Conf. on Mathematical DDE
Methods & Computational Techniques in Science & Engineering (MMCTSE
2018) in Murray Edwards College, University of Cambridge, UK, February
16-18, 2018. x ( t ) =
−µ x (t − τ ) , t ≥ 0 (4)
André A. Keller is with the Center for Research in Computer Science, Solutions of ODEs are finite-dimensional and do not
Signal and Automatic Control of Lille, University of Lille – CNRS France.
Address correspondence to Prof. André A. Keller: Associate Researcher,
depend on the history. Solutions of DDEs are infinite-
Equipe Systèmes Multi-Agents et Comportement (SMAC), Centre de dimensional and strongly depend on the history. More
Recherche en Informatique, Signal et Automatique de Lille (CRISTAL), specifically, the ODE in (3) has the algebraic characteristic
Université de Lille, Bâtiment M3 Extension, 59655 Villeneuve d’Ascq, equation
France. E-mail :[email protected].

ISSN: 1998-0159 64
 INTERNATIONAL JOURNAL OF MATHEMATICS AND COMPUTERS IN SIMULATION Volume 12, 2018

D. Applications of Time-Delay Systems

∆ ODE ≡ z + µ= 0, z ∈ (5) DDEs are used in numerous scientific domains in physics,
and requires an initial value. Eq. (5) is obtained by assuming engineering, chemistry, biology, economics for modeling,
the candidate function x ( t ) = Ce zt and has one eigenvalue forecasting and simulating. Backer et al. (1999) [14],
Bocharov and Rihan (2000) [15] investigated DDEs in
z = − µ . DDE in (4) has the transcendental characteristic biosciences, that is in population dynamics. Kolmanovski and
equation Myshkis (1999) report applications in medicine [12] pp. 71-80
(e.g., arterial blood pressure regulation, cancer chemotherapy,
∆ DDE ≡ s + µ e −τ z =0 (6) a model of survival of red blood cells, regulation of glucose-

[ −τ , 0]
insulin system, human respiratory system epidemiology,
and requires an initial continuous function on . Eq. epidemiology, immunology, cell kinetics notably. This study is
(6) has an infinite number of eigenvalues. The eigenvalues are focused on mechanical engineering with applications in cutting
ProductLog ( − µτ ) and milling processes. In such models, we distinguish two
defined by z= where essential pieces, the workpiece, and the cutting inserts. In
τ cutting machines, the workpiece rotates, and the cutting inserts
ProductLog ( z ) gives the principal solution for w in are fixed. On the contrary in the milling machines, the
workpiece is fixed, and the cutting inserts are rotating on an
z = we w . axis. The time-delay corresponds to the time of one complete
.
revolution of the workpiece or cutting inserts [16]. This
C. History of Time-Delay Systems substantial diffusion of applications with DDEs is also due to
A short history of delay differential equations was the inherent properties of a system and necessary feedback
proposed by [2], pp. i-iv. Such mixed (or hybrid) equations control to stabilize it. An inherent property of a system is that
were early studied by S.-D. Poisson (1781-1840) in his propagation of effects takes time to process. Feedback controls
Memorandum “Essay on the mixed difference equations” take into account delays between the observation of
(transl.) in 1806 [3]. We can also refer to J.B. Biot in 1806 discrepancies and the introduction of effective controls.
who presented a talk “About the mixed difference equations”
(transl.), in the French Institut des Sciences, Lettres et Arts par II. TIME-DELAY SYSTEM MODELING
divers Savants [4]. S.-D. Poisson also published on the We provide some basic types of delays such as internal
particular solutions of the differential and difference equations
delays and control or loop delays. Different types of time-
[5]. We could also quote the book S.F. Lacroix on “mixed
delay systems result from the comparison between the order of
difference equations” in 1819 [6]. However, the origin of such
the derivatives and the maximal order at shifted arguments.
mixed equations was attributed to Condorcet in 1767 in the
context of the astronomer’s three-body problem. This problem We deduce retarded types, neutral types, and leading types. A
is to determine the motions of three point masses which attract TD system can also be characterized by time-varying
each other according to Newton’s law (see online, the article coefficients and delays. There is also a distinction between
by Chenciner [7] for a detailed numerical introduction). time-continuous TD systems and time-discrete or digital TD
The equations may not be autonomous and contain a vector systems. A system can be described as a set of inputs or
of control variables. Delays may be placed on the state control variables, a set of outputs, and a collection of
variables and their derivatives in the case of neutral forms. information containing the history. The state-space
Moreover, the scalar integro-differential-difference equations representation consists of two equations, a state equation, and
(IDDEs) of the population models involve distributed time- an observation equation.
delays. In such equations, the motion is not only explained by A. Types of Delays
the position and velocity of a given physical particle but also
Delays can be neglected if they have no critical effects on
all the history of the prior states. The motion equation will be
the analysis or design of a system. Otherwise, delays have to
determined not only by the initial point condition but on an be taken in the modeling process such as in metallurgical
integrable function defined on the pre-interval just before the processing system, chemical systems, power system,
motion. In the 1960’s, Bellman and Cooke published a transportation and communication, and environmental system.
seminal study on the DDEs [8], [9]. The Russian Delays intervene between the application of input or
mathematicians Kolmanovskii, Myshkis and Nosov [10]-[12] control and the resulting output. There are two main categories
also contributed to the study of DDEs’ stability and of delays, internal delays and control delays. Internal delays
applications to industrial problems. More recently, the are inherent to the system. Control or loop delays are
presence of algebraic constraints on the state or boundary introduced for control and correction purposes. Thus, in the
conditions results in delay differential algebraic equations steel rolling example in mechanical engineering, the thickness
(DDAEs). Deshmukh (2010) [13] studied the stability of of metal sheets is measured at some distance from the rolls
solutions of nonlinear DDAE with time periodic coefficients in resulting in a measurement delay.
mechanical engineering.

ISSN: 1998-0159 65
 INTERNATIONAL JOURNAL OF MATHEMATICS AND COMPUTERS IN SIMULATION Volume 12, 2018

Delays may be fixed-time (or discrete time or stationary) or A, B :    n×n are 2π - periodic functions and that
time-varying (or nonstationary). Delay can be certain or
uncertain. In the stochastic approach, parameters may fluctuate f :   [ −1,1] is also a 2π - periodic function ns.
due to the internal reasons and to a noisy environment. We
recall that in cutting and milling machines, the time-delay
Furthermore, we have δ ,τ 0 , ω , Ω ∈  ++ and δ ≤τ0 . The
represents the time taken for one complete revolution of the two examples from mechanical engineering proposed by the
workpiece (cutting machines) or cutting inserts (milling). authors are a model of a variable speed rotating cutting tool,
Parameters determine the amplitude and frequency of the delay and a model of an elastic column subjected to a periodic force
variation. [16].
We must also distinguish time-continuous (or analog) TD
B Types of Time-Delay Systems systems and time-discrete (or digital) TD system. Dynamic
Let a DDE in the general form be written in matrix form as equations are the same except that the first-order derivative in
the analog version is replaced by an advance by one sampling
x( n ) ( t ) = f (t , x ( t ) , x(1) ( t ) , , x( n −1) ( t ) , period, that is x is replaced by x ( k + 1) .
(7)
x ( t − τ 0 ) , x( ) ( t − τ 1 ) , , x m ( t − τ m ))
1
C State-Space Representation
In a system, we find a set of inputs (or control variables)
where x ∈  N denotes the state vector of finite dimension.
u = ( u1u2  ur ) , a set of outputs y = ( y1 y2  ym )
T T
The expression x ( t ) = d x / dt
(k ) k k
in (7) is the kth
and a collection of information containing the history. The
derivative evaluated at t. The order of the derivative on the
inputs and the outputs are related by the dynamic equations of
LHS is n . The maximal order at shifted arguments is m .
the system. If r= m= 1 the system is called SISO (single-
A retarded type is obtained if 0 ≤ m < n . A neutral type is
represented for m = n . A leading type supposes that m > n . input single-output). If r > 1 and/or m > 1 , the system is
An example can be the scalar DDE called MIMO (i.e., multi-input multi-output).

x f ( x ( t ) , x ( t − τ ) ) for which N= n= 1 and m = 0

The equivalent state-space representation of an invariant
= analog TD system consists of two matrix-equations, that is a
. In the Hutchinson’s population model (1948), the population state-equation and an observation-equation, so that
loop is controlled by a delayed retroaction loop. Denote the x ( t ) = Ax ( t ) + Bu ( t − τ )
intrinsic growth rate by r , the carrying capacity by K , and  (11)
the time-delay by the constant τ . By Hutchinson, we have the y ( t ) = Cx ( t )
delayed logistic equation
where A denotes the state matrix, B the input matrix, and
 x (t −τ ) 
x ( t ) r x ( t ) 1 −
C the output matrix of (11). A more general analog time-
=  (8)
invariant TD system should be
 K 
This model is also known as a Verhulst-Pearl retarded logistic
 n
equation (see Keller [2] pp. 30-33 and Hutchinson (1948)
[17]). The transcendental characteristic equation for (8) is  ( )

x t = Ax ( ) ∑ Ai x ( t − τ i ) + Bu ( t )
t +
 i =1
 Cet −τ  
D ( z ) ≡ z + 1 − B ju (t − θ j )
r

 K
r =

0 (9)
 + ∑
 j =1
and the solutions of (9) are  n
(12)
 Ce r (t −τ ) r ( t − τ )  y ( t ) = Cx ( t ) + ∑ Ci x ( t − τ i ) + Du ( t )
ProductLog    i =1
 K 
 
+ ∑ D ju (t − θ j )
r
z= r − 
t −τ  j =1

Michiels et al. (2007) [16] proposed a generalized linear where matrices of coefficients express direct transmissions in
time-varying delay of the form (12) .

x ( t ) = A (ωt ) x ( t ) + B (ωt ) x ( t − τ ( t ) ) ,
III. DYNAMICS, STABILITY AND CONTROL
(10) Problems studied by using deterministic and stochastic TD
τ (t ) =τ 0 + δ f ( Ωt ) systems are common, dynamics, stability and control.
Dynamics uses time and frequency representations. Stability
In (10), we have x ∈ n and suppose that analysis uses stability checking techniques such as the

ISSN: 1998-0159 66
 INTERNATIONAL JOURNAL OF MATHEMATICS AND COMPUTERS IN SIMULATION Volume 12, 2018

characteristic equation and usual stability criteria (e.g., root ,

W (s) e W (s)
locus criterion, Nyquist criterion). Control analysis) (i.e., = s , we may also write
controllability and observability) refers to problems like H ∞
control and filtering. W ( −bτ e aτ ) e
(
W − bτ eaτ )=
−bτ e aτ (19)
A Transfer Function Comparing (18) and (19), we deduce the roots

z = W ( −bτ e aτ ) − a
Laplace Transform (LT) converts time-domains into 1
frequency-domain functions. The consequence of this τ
transformation is that differential equations are transformed The roots of the Frisch-Holme characteristic equation are
into algebraic equations that make solving easier. The shown in Fig.1 in the complex plane.
solutions of the algebraic approach are then transferred into
the time domain. LT of function f (t ) is
∞
L  f ( t ) 
= ∫e
− st
f ( t ) dt ≡ F ( s ) (13)
−∞
where s denotes the Laplace operator. Let the state space
representation of a delayed SISO system be

x (=
t ) Ax ( t ) + A1x ( t − τ ) + Bu ( t )
 (14)
y ( t ) Cx ( t ) + Du ( t )
=
Taking LT of (14) by using (13) for initial states yields

 sX ( s ) =
AX ( s ) + A1 X ( s ) e − st + BU ( s )
 (15)
Y ( s ) CX ( s ) + DU ( s )
=
From (15) we deduce the transfer function

) C ( sI − A − A1e− st ) B + D
H ( s=
where H (s) = Y (s) / U (s) .
B Lambert Function-Based Method
This method is used to solve analytically the transcendental
characteristic equation of DDEs as in sections I-B and II-B.
Let the boundary-value problem of the Frisch-Holme
prototype [18]
Fig.1 roots of the transcendental characteristic equation of
 y ( t ) =
−ay ( t ) − by ( t − τ ) , τ > 0 Frisch-Holme prototype for coefficients a = 1 , b = 2 and
 (16)
τ =1
y ( t ) φ ( t ) , t ∈ [ 0,τ ]
=
where a and b are positive coefficients. This model C Asymptotic Behavior of Time-Delay Systems
illustrates a cell population growth [15]. It can be shown that The asymptotic behavior of DDEs may involve oscillatory
the lag value produces instability. solutions that do not appear without delay. The delay may be
Assuming the candidate function y ( t ) = Ce zt , we deduce destabilizing with possible stability regimes. We study the
dynamics of the logistic delay prototype (8) for different
from (16) the transcendental characteristic equation [19],[2]
values of the intrinsic growth rate (or Malthus’ coefficient r
and of the delay τ .By using the variable changing
pp. 47-49.

 
D ( z ) ≡ ( a + z ) e zτ = ( t ) K 1 + y 
−b t
(17) x=   , we obtain the equivalent Wright
Multiplying both sides of (17) by τ e
aτ
, we get
 τ 
form [20] by means of some simple algebraic manipulations
τ (a + z) e τ (a+ z)
−bτ e
= aτ
−rτ (1 + x ( t ) ) x ( t − 1)
x ( t ) =
(18)
(20)
According to the definition of a Lambert function for which

ISSN: 1998-0159 67
 INTERNATIONAL JOURNAL OF MATHEMATICS AND COMPUTERS IN SIMULATION Volume 12, 2018

Using (20), Fig. 2 and Fig. 3 illustrate the dynamics in

interactive objects from Mathematica ® software. These
objects contain controls (sliders) for different parameter
values. These interactive applications let explore ranges of
coefficient values, delays, and initial conditions. The
consequences on the results of such changes are observed
immediately. Fig. 2 shows an oscillatory convergence to the
origin in the phase space ( xτ , x ) for an intrinsic growth rate
of r = 1 and a unit delay.

Fig. 3 logistics DDE dynamics for r = 2 and τ = 1

D Stability Checking Techniques

Criteria like Ruth-Hurwitz for non-delay systems will not
operate. Using the Neimar D-partition can be exhibited in the
parameter space. Frequency response tools like Bode plots,
Nyquist and Nichols's plots contribute to LTI system analysis.
The Neymar D-partition is retained here for this
presentation [12] pp. 236-240 and [2] pp. 199-202. This
Fig. 2 logistics DDE dynamics for r= τ= 1 . technique is adequate to the study of the stability of the DDE
Fig. 3 shows an oscillatory convergence to a limit cycle in Frisch-Holme prototype (16).
the phase space ( xτ , x ) for r = 1 and τ = 2 The transcendental characteristic equation for this model in
(17) can be rewritten as
D ( z ) ≡ a + z + be zτ =0 , (21)

where z = σ + jω , j = −1 . Putting the expression of z

into (21) , separating the real and imaginary parts and solving
in a and b for a constant delay τ , we get the parametric
for equations defined by
a =−σ − ω cot (ωτ )
 (22)
b = ω e sin (ωτ )
στ −1

We are searching for ( a, b ) - values for which D ( z ) = 0 .

Taking the pure imaginary z = jω , the boundaries of regions
are formed of the straight line a + b = 0 and curves defined
parametrically by (22) with σ = 0 , that is
a = −ω cot (ωτ ) and b = ω sin −1 (ωτ ) . Fig. 4 shows a
stability region Γ 0 in which the characteristic roots of the
transcendental function are negative.

ISSN: 1998-0159 68
 INTERNATIONAL JOURNAL OF MATHEMATICS AND COMPUTERS IN SIMULATION Volume 12, 2018

Q is the concatenation
Matrix
Q = Q B | Q12 B | Q 22 B | Q13B | Q32 B | Q33B 
1
1

for which the calculation rules are:

(i) Q1 = I ,
1

(ii) Q kj = 0 for j = 0 or j > k ,

r +1
(iii) =
Q j AQ rj + A1Q rj −1 .
Example (Malek-Zavarei and Jamshidi, 1987 [21], p.139). We
suppose the coefficient matrices

 0 0 −1   0 0 −2  0
     
=A =1 0 2  , A1 0 0 1 , B = 0 .
 0 1 −3   0 0 −4  1
     
(i) Q1 = I , (ii), and (iii)
1
We use the rules:

( A ) , Q32 =
Q13 = ( A1 ) . We deduce
AA1 + A1A , Q13 =
2 2

 0 −1 −2 3 10 8
Fig. 4 D-partition of the parameter space ( a, b ) for
= a  
Q  0 2 1 −7 −13 −4  . In this example, the
standard scalar Frisch-Holme type DDE [adapted from Keller  1 −3 −4 11 25 16 
(2011), p.220 (©LAP Lambert Academic Press)]  
E Controllability and observability system is controllable to the origin since rank ( Q ) = 3 .
2) Observability Criterion
The attributes controllability and observability characterize
the control system. Controllability attribute characterizes the Let a system with time-varying coefficients
relationship between the input and the state. The problem of
controllability is to know if a control u ( t ) always exists
x ( t ) = A ( t ) x ( t ) + A1 ( t ) x ( t − 1)
from an initial state to any other state in a finite time.
 (24)
Observability attribute characterizes the relationship between  y ( t ) = Cx ( t )
the state and the output. The problem is to know if the initial
state can always be identified by observing the output and the
 0 1 0 0
where A ( t ) =  , A (t ) 
input of the system.
=  , and
 −2 −t   −t −1
Let the general linear MIMO TD system
 N

x ( t ) = Ax ( t ) + ∑ A i x ( t − τ i ) + Bu ( t )
 C = (1 0 ) . The observability matrix of (24) is expressed
 i =1
 by P ( t ) ( I A ( t ) A1 ( t ) ) ∗ C  . For this
(23)
N = 
T T T

y = Cx ( t ) + C x ( t − τ ) + Du ( t )
 ∑
i =1
i i example (Malek-Zavarei and Jamshidi, 1987 [21], p.148), the
1 0 0
The invariant matrices of coefficients A, A1 , A 2 , , A N , B observability matrix takes the form P ( t ) =   .
will have to characterize the controllability whereas the 0 1 0
invariant matrices of coefficients C, C1 , C2 , , C N , D will We conclude that the TD system is observable for t ≥ 0 .
have to characterize the observability.
IV. APPLICATION TO MECHANICAL ENGINEERING
1) Controllability Criterion
DDE systems have been developed in industrial areas. We
will give some details about the diversity of these applications
Theorem. Let a state equation and will focus the detailed presentation on two standard well-
x (=
t ) Ax ( t ) + A1x ( t − τ ) + Bu ( t ) where x ∈  N and known mechanical systems namely the metal rolling system
and the machine tool chatter problems.
u ∈  m . The system is controllable to the origin if a matrix
Q has rank N .

ISSN: 1998-0159 69
 INTERNATIONAL JOURNAL OF MATHEMATICS AND COMPUTERS IN SIMULATION Volume 12, 2018

A. Industrial Applications of Time-Delay Systems where k p and km are gains of proportional controller in (25)
Industrial applications of time-delay systems are multiple . The state-space representation of the metal rolling system is
and diverse. DDE systems can be found in physics,
engineering (e.g., aerospace, robotics, mechanical systems,  x (=
t ) Ad x ( t − τ ) + Bu ( t ) , x ( 0=
) x0
 (26)
 y ( t ) = x ( t )
electrical systems, manufacturing, metallurgical systems),
chemistry, bioscience (e.g., population dynamics), medicine
(e.g., epidemiology, immunology, cell kinetics), water where k p and km are gains of proportional controllers.
resources (e.g., hydraulic control), communications (e.g.,
The block diagram of this metal rolling system in (26) is
traffic control, transport process), computerized systems.
represented by Fig. 6. Suppose that the transfer function
This presentation is devoted to the stability and control of
between the upper roller position and the input voltage of the
mechanical systems. Mechanical engineering can be
represented by fixed or time-varying nonlinear TD systems. driver motor is G ( s ) = km / s . The transfer function of the
C ( s ) = k p / s . The transfer function of
We may also have algebraic boundary conditions as with time
periodic DDAEs (i.e., delay-differential-algebraic equations). used controller is
Boukas and Liu (2002) [22] studied three applications: (1) A the censor is H ( s ) = e −τ s . The effect of the external
rolling system, 2) a heating system, and (3) a controlled
feeding of the production system. Michiels et al. [16] disturbances on the system is denoted by P ( s ) .Suppose that
proposed two mechanical applications: (1) a variable speed
rotating cutting tool and (2) an elastic column subjected to a G ( s ) is given by
periodic force. km
This presentation focuses on the description and analysis of G (s) =
two mechanical engineering applications with milling and s (1 + Ts )
cutting machines, i.e., (1) Metal rolling system and (2)
Machine tool chatter. We describe the model for each We get the following second-order DDE [22] p.6
application, show the dynamic equation, the equivalent
representations. Elements of stability analysis are shown in
application 2.
y ( t ) + y ( t ) + k p km y ( t − τ ) =
 k p k m yd ( t ) .
B. Mechanical Application 1: Metal Rolling System
In metal rolling system, ribbons of hot metal sheets run
though rollers as in Fig. 5. The goal is to produce metal sheets
of a given thickness by controlling the upper roller, whereas
the lower one is fixed. A drive motor is used to operate this
adjustment. The sensor thickness is placed downstream at a
distance d of the rollers. The ribbon is fed at speed v into
the system [22] pp. 4-6, [23].
Fig. 6 block diagram of the metel rolling system [adapted
from Boukas and Liu (2002) Figure 1.2, p. 6]

C. Mechanical Application 2: Machine Tool Chatter

Machining processes like turning, milling, and drilling can
be disturbed by vibrations produced by coupling between a
machine tool and the cutting forces. Vibrations result in waves
on the machine surface. In this study, a machine tool removing
a chip is deviated from its steady-state cutting by a small
vibration x ( t ) between the machine tool and a workpiece as
Fig. 5 metal rolling system [adapted from Mirkin [23]]
in Fig. 7. Dynamic forces P ( t ) act between the machine tool
The delay depends on both distance and speed, such that
and the workpiece, as in Fig. 7. The goal is to analyze the
τ =d /v . The measured thickness is y ( t − τ ) , the real machine tool stability by using a method based on the
y ( t ) , and desired thickness be denoted by yd ( t ) .
feedback control theory [24]. Fig. 7(a) shows the positions of
thickness
the workpiece, the tool,, and chip. It also indicates the fee
We obtain the DDE direction. Fig. 7(b) shows how forces are acting in the system.

y ( t ) + k p km y ( t − τ ) =
k p k m yd ( t ) (25)

ISSN: 1998-0159 70
 INTERNATIONAL JOURNAL OF MATHEMATICS AND COMPUTERS IN SIMULATION Volume 12, 2018

transfer function is G ( s ) or the corresponding frequency

function G ( jω ) = Me jφ in complex polar form with usual
notations. The chip thickness modulation is expressed by
δ (t ) =x (t ) − µ x (t − T ) . The response function is

H1 ( jω ) = 1 − µ e − jωT
. The frequency response function of

the cutting process is H 2 ( jω ) = Ke jω . The block diagram

of the metal cutting system is shown in Fig. 8.

Fig. 7 Regenerative chatter in metal cutting [taken fron Mirkin

(2012) [23]]
Fig. 8 block diagram of the metal cutting system [adapted
The machining process model shows a closed loop from Nigm [24] Fig.2, p. 253]
interaction between two blocks, a cutting process sub-model,
and a structural dynamics sub-model. Deflections x ( t ) result The open-loop frequency response function of the system is

from the action of cutting forces F ( t ) on the structural T ( jω ) =G ( jω ) × H1 ( jω ) × H 2 ( jω ) (29)

dynamics submodel. These deflections, in turn, modulate the
cutting forces [16][24]-[25]. . Eq.(29) is equivalent to T ( jω ) = Ae jα with gain
The dynamic equation results from the combination of the A = MCK and phase α =φ +ψ + θ .
cutting process and the structural dynamics. A cutting force
Finally, we apply a stability criterion for the frequency at
equation can be expressed by
which the open-loop phase is such as α = π . The system is
stable if the gain is A < 1 . The system is unstable if A > 1 ,
(
F (t ) = k f + x (t − φ (t )) − x (t ) d ) (27)
and on the threshold of stability if. A = 1 If A = 1 , then
where k denotes a cutting force coefficient, f a nominal K = 1/ MC . The system is stable if K < 1/ MC and
feed rate, d a nominal depth of cut, and where the colored unstable if K>1/M.
expression is a regeneration term which modulates the cutting
V. CONCLUSION
force. The model of structural dynamics computes the
deflection of the machine tool structure resulting from the The challenge of this article was to be an introduction to the
cutting forces. Using the transfer function approach, we have vast domain of the DDE modeling. Interest for these hybrid
systems goes back to an ancient times as we showed it in the
X (s) 1/ m introduction. Block diagram techniques have contributed
=
G (s) = 2 (28)
F ( s ) s + 2ξω s + ω 2
significantly to these developments and the industrial
applications. The approach is mostly introductory on the whole
The dynamic equation results from (27)-(28). We have the subject without addressing particular points of concern. This
typical second-order DDE article is limited to some illustrative techniques without
x ( t ) + 2ξω x ( t ) + ω 2 x ( t )=

k
m
(
x (t − φ (t ) − x (t )) , ) addressing them all of course. The field of mechanics is
privileged, but the presentation is limited to two significant
where ξ denotes the damping ratio, ω the frequency, and
examples such as rolling sheet steel and the vibrations of a
machine cutting tool.
m the modal mass.
The system consists of three blocks: the machine tool Other examples relating to mechanics are presented in the
structure, the chip thickness modulation, and the cutting book by Boukas [26] pp. 14-28 on “Control systems
process. Each of them represents a transfer function with engineering” (transl.). These examples are the speed control of
inputs and outputs. The input of the machine tool block is the a conveyor, the flushing system, the positioning servo of a
force F and the output is the relative displacement x . The parabolic antenna, the control of a ship's rudder, the control of
the angular position of an airplane rudder. Other specifications

ISSN: 1998-0159 71
 INTERNATIONAL JOURNAL OF MATHEMATICS AND COMPUTERS IN SIMULATION Volume 12, 2018

on circuit analysis are proposed in different studies of the [5] S.D. Poisson, “Mémoire sur les solutions particulières des équations
différentielles et des équations aux différences [in French], “ J. Ecole
author such as other system representations (e.g., phasor Polytech., JEP, vol 13, no. 6, pp. 60-125, 1806.
representation [27]) , solving methods (Method of Step [6] S.F. Lacroix, “ Des équations aux differences mêlées [in French],” in
Algorithm MSA, Differential Transform Methods DTM [28] Traité du Calcul Différentiel et du Calcul Intégral , vol. 3, Ch.8, Paris,
with application to the retarded Kalecki’s business cycle France, 1819, pp.575-600.
[7] A. Chenciner (2007). Availablet
model, Solow vintage capital growth model with variable http://www.scholarpedia.org/article/Three body problem .
delay [29]), stability criteria besides the root locus criterion [8] R. E. Bellman and K.L. Cooke, “Asymptotic behavior of solutions of
(e.g., BIBO stability, Lyapunov stability criterion, Nyquist differential-difference equations,” Tech. Report 35, American
Mathematical Society, New York, 1959.
stability criterion) with applications to environmental pollution
[9] R. E. Bellman and K.L. Cooke, “Differential-difference equations,”
systems and fishery environmental systems [27]. Tech. Report R-374-PR , The Rand Corporation, Santa Monica, CA,
Other important technical aspects of dynamics and USA, 1963.
control can be addressed such as the assumption of a stochastic [10] V.B. Kolmanovskii .and V.R. Nosov, “Stability of functional
differential equations,” in Mathematics and Its Applications, vol. 463,
environment, PID control and fuzzy control. A noisy Academic Press, London New York, 1986.
environment can be adapted to describe real-life modeling [11] V.B. Kolmanovskii .and A. Myshkis, “Applied theory of functional
differential equation,” in Mathematics and Its Applications vol. 85,
processes of populations. In the papers [29]-[30], populations Kluwer Academic Press, Dordrecht New York, 1992.
interact, fight or cooperate, in real life conditions with constant [12] V.B. Kolmanovskii .and A. Myshkis, “Introduction to the theory and
applications of functional differential equations,” in Mathematics and
time-delays in a noisy environment. A typical small delay Its Applications vol. 463, Kluwer Academic Press, Dordrecht New
stochastic process can be York, 1999.

( t ) f ( x ( t ) , x ( t − τ ) ) dt + σ g ( x ( t ) ) dW ( t ) ,
[13] V. Deshmukh, “Approximate stability analysis and computation of
=
dx solutions of nonlinear delay differential algebraic equations with time
periodic coefficients,” J. Vib. Control, vol. 16, no. 7-8, pp. 1235-1260,
where f , g are known functions, τ is a constant time-delay, 2010.
[14] C.T.H. Backer, G.A. Bocharov and F.A. Rihan, “A report on the use of
σ scales the noise amplitude, and W ( t ) is a Wiener delay differential equations in numerical modelling in the biosciences,”
Tech. Report no 343, University of Manchester, UK, 1999.
process. PID controllers involve three possible coefficients: a [15] G.A. Bocharov and F.A. Rihan, “Numerical modelling in biosciences
proportional gain (P) in reaction to the current error, an using delay differential equation,” J. Comput. Appl. Math., vol. 125, no.
1-2, pp. 183-199, 2000.
integral gain (I) reacting on the sum of past errors, and a [16] W. Michiels, K. Verheyden and S.-I., Niculescu, “Mathematical and
derivative gain (D) in reaction to the rate of change of the computational tools for the stability analysis of time-varying delay
systems and applications in Mechanical Engineering”, in J. Chiasson
error. In this paper PID controls are applied to a continuous
and J.J. Loiseau (Eds.), Appl. of Time-Delay Systems, NCIS, vol. 352,
multiplier-accelerator model (the Phillip’s macroeconomic 2000, pp. 199-216.
business cycle model) and to nonlinear Goodwin ‘s [17] G.E. Hutchinson, “Circular causal systems in ecology,” Ann. N.Y. Acad.
Sci., vol. 50, pp. 221-246, 1948.
macroeconomic growth model [31]. A fuzzy logic controller [18] R. Frisch and H. Holme, “The characteristic solutions of mixed
operates is similar to an artificial decision-maker that operates difference and differential equations occurring in economic dynamics,”
in a closed-loop system in real time. Description and Econometrica, vol. 3, no. 2, pp. 225-239, 1935.
[19] F.M. Asl and A.G. Ulsoy, “Analysis of a system of linear delay
applications to dynamic macroeconomic models can be found differential equations by using differential transform method,” Appl.
in [31]-[32]. Compared to a PID control, the simulation results Math. Comput., vol. 181, pp. 215-223, 2003.
[20] E.M. Wright, “A non-linear difference-differential equation,” J. Reine
with fuzzy controllers show more efficient stabilization of an Angew. Math., vol. 1955, no. 194, 2009.
economy. [21] M. Malek-Zavarei and M. Jamshidi, Time-Delay Systems: Analysis,
Optimization and Applications, Amsterdam, NL, North-Holland, 1987.
In our studies, specialized MATHEMATICA® packages [22] E.-K. Boukas and Z.-K. Liu, Deterministic and Stochastic Time Delay
(e.g., Control System Professional, Polynomial Control Systems, Boston, MA, USA, Birkhäuser, 2002.
System, and SchematicSolver) are used for analyzing and [23] L. Mirkin, Introduction to Time-Delay Systems, Lectures of Automatic
solving the systems symbolically and numerically [33]-[35]. Control–Doctorate Program, Lund Univ.(2012).See
http://www.control.lth.se/media/Education/DoctorateProgram/2012/Lect
These studies also used MATLAB® specialized packages ures
(e.g., Simullink and ControlSystem, Fuzzy Logic Tool 2) [36]. [24] M.M. Nigm, “A method for the analysis of machine tool chatter,” Int. J.
Mach. Tool Des. Res. 21 (3) 251-261, 1981.
[25] S. Jayaram, S.G. Kapoor and R.E. DeVor,”Estimation of the specific
REFERENCES cutting pressure for mechanistic cutting force models,” ASME J.
[1] M.S. Fofana, “Delay dynamical systems and applications to nonlinear Manuf. Sci. Eng. 122 (3), 391-397 (2000).
machine-tool chatter,” Chaos Solitons Fractals, vol 17, no 4, pp. 731- [26] E-K. Boukas, Systèmes asservis [in French], Montréal, CN, Editions
747, 2003. Ecole Polytech. Montréal, 1995.
[2] A.A. Keller, Time-Delay Systems with Applications to Economic [27] A.A. Keller, “Circuit analysis to environmental economic dynamics and
Dynamics and Control. Saarbrücken, DE, Lambert Academic Press control,” in Proc. Int. Conf. Development, Energy, Environment,
LAP, 2011. Economics (DEEE), Puerto De La Cruz, Tenerife, ES, WSEAS Press,
[3] S.D. Poisson, “Mémoire sur les équations aux différences mêlées [in 2010, pp. 133-141, Available
French], “ J. Ecole Polytech., JEP, vol 13, no. 6, pp. 126-147, 1806. http://www.wseas.org/multimedia/books/2010/Tenerife/DEEE.pdf
[4] J.B. Biot, “Sur les équations aux différences mêlées [in French], “ Essay [28] A.A. Keller, “Generalized delay differential equations to economic
of the Institut des Sciences, Lettres et Arts par divers Savants, Paris, dynamics and control,” in Recent Advances in Applied Mathematics,
France, vol., pp. 296-327, 1806. Proc. American Conf. Applied Mathematics (AMERICAN-MATH’10),

ISSN: 1998-0159 72
 INTERNATIONAL JOURNAL OF MATHEMATICS AND COMPUTERS IN SIMULATION Volume 12, 2018

University of Harvard, Cambridge, MA, USA, 2010, pp. 278-286.

Available
http://www.wseas.org/multimedia/books/2010/Harvard/MATH.pdf
[29] A.A. Keller, “Stochastic delay Lotka-Volterra system to interactive
population dynamics, “, in Recent Research in Applied Mathematics,
Simulation and Modelling, Proc. 5th Int. Conf. Applied Mathematics,
Simulation, Modelling (ASM’11), Corfu Island, GR, 2011, pp. 191-196
[30] A.A. Keller, “Population biology models with time-delay in a noisy
environment,” WSEAS Trans. Biology Biomedicine, vol. 8, no. 4,
October 2011, pp. 113-134. Available http://wseas.us/e-
library/transactions/biology/2011/54-076.pdf
[31] A.A. Keller, “Optimal economic stabilization policy under uncertainty,”
in Advanced Technologies, K. Jayanthakumaran, Vukavar, HR: In-The,
2009, pp. 441-470.
[32] A.A. Keller,”Fuzzy control of macroeconomic models,” Proc. World
Academy of Science, Engineering and Technology, Prague, CZ, vol. 31,
July 2008, pp. 145-154.
[33] M.D. Ludovac and D.V. Tošić, SchematicSolver version 2.2,
MATHEMATICA® Application: Symbolic Signal Processing, 2009.
Available kondor.etf.bg.ac.rs/lutovac
[34] M.D. Ludovac, D.V. Tošić and B.L. Evans, Filter Design for Signal
Processing Using MATLAB® and MATHEMATICA®, Upper Saddle
River, NJ, USA: Prentice Hall, 2001
[35] M.S. Stachowicz and L. Beal, MATHEMAICA Fuzzy Logic, Wolfram
Research. Available http://www.wolfram.com
[36] The MathWorks, User’s Guide: MATLAB 7, Simulink 7, Simulink
Control Design, Control System Design, Fuzzy Logic Toolbox 2”,
Natick, MA, USA, 2008

André A. Keller is at present an Associated Researcher at the Center for

Research in Computer Science, Signal, and Automatic Control of Lille, by the
University de Lille in France. He received a ‘Doctorat d’Etat’ (PhD.) in
Economics/ Operations Research from the University of Paris 1 Panthéon-
Sorbonne, and a post-doctorate from the University Paris X Nanterre. He is a
Reviewer for the international journals major publishers such as Elsevier,
Hindawi, Springer, World Academic Publishing, World Scientific, WSEAS
Press.
He taught applied mathematics (optimization techniques) and
econometric modeling, microeconomics, the theory of games and dynamic
macroeconomic analysis. His experience centers are in building and analyzing
large-scale macroeconomic systems, as well as forecasting. His publications
consist of writing articles, book chapters, and books. One book is “Time-
Delay Systems with Applications to Economic Dynamics & Control”
(Saarbrücken, DE: Lambert Academic Press LAP, 2011). One another recent
book is “Mathematical Optimization Terminology: A Comprehensive
Glossary of Terms” (Cambridge, MA, USA: Elsevier/Academic Press, 2017).
Other recent books are on “Multi-Objective Optimization in Theory and
Practice I: Classical Methods” (Sharja, UAE: Bentham Science, 2017) and
“II: Evolutionary Algorithm” (in Press). His research interests include high-
frequency time-series modeling with application to the foreign exchange
market, discrete mathematics (graph theory, combinatorial optimization),
stochastic differential games and tournaments, circuit analysis, optimal
control under uncertainties. (fuzzy control).
Prof. André A. Keller is a member of the American Association for
Science and Technology (AASCIT). He reviewed a scientific project for Israel
Science Foundation (IFS), and numerous scientific papers in international
Journals major publishers such as Elsevier, Hindawi, Springer, World
Academic Publishing, World Scientific, WSEAS Press. He received the Best
papers award, notably for American Math 10’ in Harvard University,
Cambridge, MA, USA. He published a paper in NAUN Journal International
Journal of Fuzzy System and Advanced Applications, vol. 1, 2014, pp. 36-54.

ISSN: 1998-0159 73