SAND2006-1938C

Exergy and Irreversible Entropy Production

Thermodynamic Concepts for Control Design:
Nonlinear Regulator Systems
Rush D. Robinett, III David G. Wilson
Energy, Infrastructure, and Knowledge Systems Intelligent Systems and Robotics Center
Sandia National Laboratories, P.O. Box 5800 Sandia National Laboratories, P.O. Box 5800
Albuquerque, NM 87185–0741 Albuquerque, NM 87185–1003
Email: [email protected] Email: [email protected]

Abstract— This paper1 develops a novel control system design and losslessness as understood in nonlinear circuit theory, and
methodology that uniquely combines: concepts from thermody- their counterparts in classical physics. Most recently, Ortega,
namic exergy and entropy; Hamiltonian systems; Lyapunov’s Jiang, and Hill [5] reviewed recent results on the stabilization
direct method and Lyapunov optimal analysis; electric AC power
concepts; and power flow analysis. Relationships are derived of nonlinear systems using a passivity approach. Passivity
between exergy/entropy and Lyapunov optimal functions for properties play a vital role in designing asymptotically sta-
Hamiltonian systems. The methodology is demonstrated with bilizing controllers for nonlinear systems where the nonlinear
two fundamental numerical simulation examples: 1) a Duffing versions of the Kalman-Yacubovitch-Popov lemma are used as
oscillator/Coulomb friction nonlinear model that employs PID key testing tools. The dissipative characteristics of dynamical
regulator control and 2) a van der Pol nonlinear oscillator
system. The control system performances and/or appropriately systems has its origins in work by Willems [6] with further
identified terms are partitioned and evaluated based on exergy specifics given by Hill and Moylan [7]. In [7], a technique
generation and exergy dissipation terms. This novel nonlinear is introduced for generating Lyapunov functions for a broad
control methodology results in both necessary and sufficient class of nonlinear systems represented by state equations.
conditions for stability of nonlinear systems. The system, for which a Lyapunov function is required, is
assumed to have a property called dissipativeness. In other
I. I NTRODUCTION
words, the system absorbs more energy from the external
Today’s engineering systems sustain desirable performance world than it supplies. Different types of dissipativeness can
by using well-designed control systems based on fundamental be considered depending on how the “power input” is selected.
principles and mathematics. Many engineering breakthroughs Dissipativeness is shown to be characterized by the existence
and improvements in sensing and computation have helped of a computable function which can be interpreted as the
to advance the field. Control systems currently play critical “stored energy” of the system. Under certain conditions, this
roles in many areas, including automation, manufacturing, energy function is a Lyapunov function which establishes
electronics, communications, transportation, computers, and stability, and in some cases asymptotic stability, of the isolated
networks, as well as many commercial and military sys- system. It was shown that for a certain class of nonlinear
tems [1]. Traditionally, almost all modern control design is systems, that an “energy” approach was useful in analyzing
based on forcing the nonlinear systems to perform and behave stability. Kokotovic and Arcak [8] provide a recent discus-
like linear systems, thus limiting its maximum potential. In sion about the historical perspective of constructive nonlinear
this paper a novel nonlinear control design methodology is control theories. Structural properties of nonlinear systems and
introduced that overcomes this limitation. passivation-based designs exploit the connections between pas-
Several of the popular advanced nonlinear control system sivity and inverse optimality, and between Lyapunov functions
approaches are based in passivity and dissipative control and optimal value functions. Recursive design procedures,
theories. Initially, Moylan [2] discussed the implications of such as backstepping and forwarding, achieve certain optimal
passivity for a broad class of nonlinear systems, a connection properties for important classes of nonlinear systems. Some
is established between the input-output property of passivity of the more popular nonlinear control system designs [9],
and a set of constraints on the state equations for the system. [10], [11] have their fundamental foundations built upon these
Later, Wyatt, et.al. [3], [4] clarified the meaning of passivity concepts.
In other engineering disciplines, Alonso and Ydstie [12]
1 Sandia National Laboratories is a multiprogram laboratory operated by
connect thermodynamics and the passivity theory of nonlinear
Sandia Corporation, a Lockheed Martin Company, for the U.S. Department
of Energy’s National Nuclear Security Administration under contract DE- control. The storage function is derived from the convexity
AC04-94AL85000. of the entropy and is closely related to thermodynamic avail-
ability. Dissipation is related to positive entropy production. The term on the left represents the rate at which energy is
In this form the supply function is a product of force and changing within the system. The heat entering or leaving the
flow variation variables. Results are discussed in relationship system is given by Q̇i and the work entering or leaving the
to heat conduction and reaction diffusion equation problems. system is given by Ẇj . Next, material can enter or leave the
Anthony [13] suggests that non-equilibrium thermodynamics system by ṁk that includes enthalpy, h, kinetic and potential
of irreversible processes may be included into the framework energies, ke, pe, etc. In addition, each term is “summed” over
of a Lagrangian formalism. This formalism presents a unified an arbitrary number of entry and exit locations i, j, k.
method for reversible and irreversible processes. A straight- The second law or entropy rate equation for a system [14]
forward procedure allows for the incorporation of both the is given as
first and second laws of thermodynamics into the Lagrangian.
The theory is illustrated in three representative examples which X Q̇i X
Ṡ = + ṁk sk + Ṡi = Ṡe + Ṡi . (2)
include; material flow, heat conduction, diffusion and chemical Ti
i k
reactions.
The main contribution of this paper is to present a novel Where the left hand term is the rate entropy changes within
nonlinear control design methodology that is based on ther- the system and the right hand terms represent, in order, the
modynamic exergy and irreversible entropy production con- rate heat conducts entropy to and from the system and the
cepts. Relationships are developed between exergy, irreversible rate material carries it in or out. These two terms can be
entropy production, Hamiltonian systems, Lyapunov optimal combined into one term Ṡe , the entropy exchanged (either
functions, electric AC power concepts, and power flow, for positive or negative) with the environment and Ṡi is the
control system design. Both necessary and sufficient condi- irreversible entropy production rate within the system. Figure 1
tions for stability are determined for nonlinear systems. By (middle) shows the entropy exchanges and production within
combining the first and second laws of thermodynamics, an the system [15].
exergy analysis approach is developed to construct Lyapunov The irreversible entropy production rate can be written as
optimal functions for Hamiltonian systems. The first time the sum of the thermodynamic forces and the thermodynamic
derivative of the Lyapunov functions, based on exergy, irre- flows [15], [16]
versible entropy production rate, and power flow is partitioned S˙i =
X
Fk X˙k ≥ 0 (3)
into either exergy dissipative or exergy generative terms. k
This paper is divided into eight sections. Sections II and
III provide the preliminary thermodynamics and Hamiltonian where the entropy change is the sum of all the changes due to
mechanics definitions. Section IV develops the relationships the irreversible flows Ẋk with respect to each corresponding
and connections between thermodynamics and Hamiltonian thermodynamic force Fk .
mechanics. Section V defines the necessary and sufficient Next, for systems with a constant environmental tempera-
conditions for stability of nonlinear systems. Section VI shows ture, a thermodynamic quantity called the availability function
how, with simplifications to this novel control theory, conven- which has the same form as the Helmholtz free energy function
tional Lyapunov optimal and passivity control design method- is defined as [15]
ologies are recovered. Section VII presents regulator control Ξ = E − To S (4)
design examples that include; 1) a PID control regulator for a
nonlinear Duffing oscillator/Coulomb friction dynamic system where To is the reference environmental temperature. The
and 2) a van der Pol nonlinear oscillator system. Numerical availability function is described as the maximum theoretically
simulations resulted in the demonstration of both performance available energy that can do work which we call exergy.
and stability criteria. Finally, section VIII summarizes the Exergy is also known as negative-entropy [14], [17]. By
results with concluding remarks. taking the time derivative of the availability function (4) and
substituting in the expressions for (1) and (2) results in the
II. T HERMODYNAMIC C ONCEPTS exergy rate equation

In this section the first and second laws of thermodynamics P  To


Ξ̇ = i 1 − Ti Q̇i
are used to define exergy. One interpretation of the first law of P   P
thermodynamics states energy is conserved (see Fig. 1-left). + j Ẇj − po ddtV̄ + k ṁk ζkf low − To Ṡi .
The second law of thermodynamics implies that the entropy of (5)
the universe always increases. The first law is a conservation Where Ξ̇ is the rate at which exergy stored within the system
equation while the second law is an inequality. Mathematically, is changing. The terms on the right, in order, define the rate
a result of the first law can be written in terms of it’s time exergy is carried in/out by; i) heat, ii) work (less any work
derivatives or energy rate for a system [14] as the system does on the environment at constant environmental
X X X pressure po if the system volume V̄ changes), and iii) by the
Ė = Q̇i + Ẇj + ṁk (hk + kek + pek + . . .) . (1) material (or quantity known as flow exergy). The final term,
i j k To Ṡi , is the rate exergy is destroyed within the system.
 III. H AMILTONIAN M ECHANICS B. Reversible Thermodynamic Systems
The derivation of the Hamiltonian [18] begins with the A thermodynamic system is reversible if
Lagrangian for a system defined as dQ
dS
H = HT dQ
L = T (q, q̇, t) − V(q, t) (6) dS = T =0
H H Hh i
t = time explicitly dS = [dSi + dSe ] = Ṡi + Ṡe dt = 0
q = N-dimensional generalized coordinate vector
where q̇ = N-dimensional generalized velocity vector which implies that S˙e = Q̇/T since by definition the second
T = Kinetic energy, and law gives S˙i = 0.
V = Potential energy. C. Irreversible Thermodynamic Systems
The Hamiltonian is defined in terms of the Lagrangian as
For I I h
n i
X ∂L dS = Ṡi + Ṡe dt = 0
H≡ q̇i − L(q, q̇, t) = H(q, q̇, t). (7)
i=1
∂ q̇i
then Ṡe ≤ 0 and Ṡi ≥ 0.
The Hamiltonian in terms of the canonical coordinates (q, p)
is n
D. Analogies and Connections
X
H(q, p, t) = pi q̇i − L(q, q̇, t) (8) Now the connections between thermodynamics and Hamil-
i=1 tonian mechanics are investigated.
where the canonical momentum is defined as 1) The irreversible entropy production rate can be ex-
∂L pressed as
pi = . (9)
∂ q̇i X 1 X
Ṡi = Fk Ẋk = Qk q̇k ≥ 0. (15)
Then Hamilton’s canonical equations of motion become To
k k
∂H
q̇i = ∂pi
2) The time derivative of the Hamiltonian is equivalent to
∂H (10) the exergy rate
ṗi = − ∂qi
+ Qi
P
where Qi is the generalized force vector. Next taking the time Ḣ = k Qk q̇k PN PM−N
derivative of (8) gives Ξ̇ = Ẇ − To Ṡi = j=1 Qj q̇j − l=N Ql q̇l
n   (16)
X ∂L ∂L ∂L Where N is the number of generators, P M − N the
Ḣ = ṗi q̇i + pi q̈i − − q̇i − q̈i . (11)
∂t ∂qi ∂ q̇i number of dissipators, and let Ẇ =
i=1 j Ẇj . The
Then substitute (10) into (11) and simplifying gives following assumptions apply when utilizing the exergy
n
rate equation (5) for Hamiltonian systems:
X ∂L a) No substantial heat flow:
Ḣ = Qi q̇i − . (12)
i=1
∂t
Q̇i ≈ 0.
Hamiltonians for most natural systems are not explicit func-
tions of time (or ∂L/∂t = 0). Then for b) No substantial exergy flow or assume Ti is only
slightly greater than To :
L = L(q, q̇) (13) To
1− ≈ 0.
the power (work/energy) equation becomes Ti
n
X c) No po V̄ work on the environment:
Ḣ(q, p) = Qi q̇i . (14)
i=1 dV̄
po = 0.
IV. T HERMO -M ECHANICAL R ELATIONSHIPS dt
d) No mass flow rate:
A. Conservative Mechanical Systems X
A system is conservative if ṁk ζkf low = 0.
k
Ḣ = 0 and H = constant.
e) Then define:
A force is conservative if
I I I Ẇ ≥ 0 power input/generated
F · dx = F · vdt = Qj q̇j dt = 0 To Ṡi ≥ 0 power dissipated.
3) A conservative system is equivalent to a reversible
where F is the force, dx the displacement, and v the velocity.
system when
Basically, all of the forces can be modeled as potential force
fields which are storage devices. Ḣ = 0 and S˙e = 0
 then A. Stability and Instability Theorems
S˙i = 0 and Ẇ = 0. To describe a nonlinear system’s behavior two theorems [20]
help to characterize the essential features of its motion. In
4) For a system that “appears to be conservative”, but is addition, by bounding the Lyapunov function between these
not reversible is defined as: Theorems, both necessary and sufficient conditions are a
result of the transition of the time derivative of the Lyapunov
Ḣave = Pave
H (over a cycle) = 0 function from stable to unstable.
= 1 ˙
τ [Ẇ − To Si ]dt 1) Lyapunov Theorem for Stability Assume that there
= (Ẇ )ave − (To Ṡi )ave exists a scalar function V of the state x, with continuous
1
H PN PM−N
= τ [ j Qj q̇j − l=N Ql q̇l ]dt first order derivatives such that
V (x) is positive definite
where τ is the period of the cycle. To be more specific
about the average power calculations, the AC power V̇ (x) is negative definite
factor [19] provides an excellent example. For the gen- V (x) → ∞ as kxk → ∞
eral case of alternating current supplied to a complex Then the equilibrium at the origin is globally asymptot-
impedance the voltage and current differ in phase by an ically stable.
angle θ. For 2) Chetaev Theorem for Instability Considering the equa-
√ √ tions of disturbed motion, let V be zero on the boundary
Ẇ = P = Qq̇ = v i = 2v̄ cos(ωt + θ) · 2ī cos ωt of a region R which has the origin as a boundary point,
= v̄ ī [cos θ + cos(2ωt + θ)] and let both V and V̇ be positive-definite in R; then the
undisturbed motion is unstable at the origin.
where P is power, v is voltage (v̄), i is current (ī), θ
is the phase angle, and ω is the frequency. Integrating B. Stability Lemma for Nonlinear Systems
over a cycle gives Based on the relationship between thermodynamic exergy
and Hamiltonian systems a fundamental stability Lemma can
(Ẇ )ave = v̄ ī cos θ be formulated.
Fundamental Stability Lemma for Hamiltonian Systems
where for the second term The stability of Hamiltonian systems is bounded between
I Theorems 1 and 2. Given the Lyapunov derivative as a
cos(2ωt + θ)dt = 0. decomposition and sum of exergy generation rate and exergy
dissipation rate then:
This is an important set of conditions that will be N
X M−N
X
used in the next section to find the generalized stability V̇ = Ẇ − To Ṡi = Qj q̇j − Ql q̇l (19)
boundary. j=1 l=N
5) Finally, the power terms are sorted into three categories: that is subject to the following general necessary and sufficient
a) (Ẇ )ave - power generators; (Qj q̇j )ave > 0, conditions:
b) (To Ṡi )ave - power dissipators; (Ql q̇l )ave < 0,
To Ṡi ≥ 0 Positive semi-definite, always true
c) (To Ṡrev )ave - reversible/conservative exergy stor-
Ẇ ≥ 0 Positive semi-definite; exergy pumped in.
age terms; (Qk q̇k )ave = 0.
These three categories are fundamental terms in the following The following corollaries encompass both stability and in-
design procedures. stability for Hamiltonian systems which utilize AC power
concepts [19]:
Cor 1: For (To Ṡi )ave = 0 and (Ẇ )ave = 0 then V̇ = 0 the
V. N ECESSARY AND S UFFICIENT C ONDITIONS FOR
Hamiltonian system is neutrally stable, conservative
S TABILITY
and reversible.
The Lyapunov function is defined as the total energy which Cor 2: For (To Ṡi )ave = 0 and (Ẇ )ave > 0 then V̇ > 0 the
for most mechanical systems is equivalent to an appropriate Hamiltonian system is unstable.
Hamiltonian function Cor 3: For (To Ṡi )ave > 0 and (Ẇ )ave = 0 then V̇ <
0 the Hamiltonian system is asymptotically stable
V =H (17)
and a passive system in the general sense (passivity
controllers).
which is positive definite. The time derivative is
Cor 4: Given apriori (To Ṡi )ave > 0 and (Ẇ )ave > 0 then
P PN PM−N the Hamiltonian system
 is further
  subdivided into:
V̇ = Ḣ = k Qk q̇k = j Qj q̇j − l=N Ql q̇l
= Ẇ − To Ṡi . 4.1: For To Ṡi > Ẇ with V̇ < 0
ave ave
(18) yields asymptotic stability.
    
4.2: For To Ṡi = Ẇ with V̇ = 0 nonlinear stiffness and Coulomb friction coefficients are KN L
ave ave
yields
 neutral
 stability.
  and CN L , respectively. The PID controller is defined as
Z t
4.3: For To Ṡi < Ẇ with V̇ > 0
ave ave u = −KP x − KI xdτ − KD ẋ (23)
yields an unstable system. 0
The bottom line is that stability is defined in terms of power where KP , KI , and KD are the proportional, integral and
flow which determines whether the system is moving toward derivative controller gains, respectively.
or away from its minimum energy and maximum entropy state. Initially, the nonlinear Duffing oscillator is investigated as
a neutrally stable, reversible conservative system or
VI. LYAPUNOV O PTIMAL AND PASSIVITY C ONTROL
M ẍ + Kx + KN L x3 = −KP x
Present day robotic and aerospace applications use feedback
controller designs that are Lyapunov Optimal [21]. A control subject to the initial condition x(0) = xo = 1.0. Now
law is Lyapunov Optimal if it minimizes the first time deriv- apply exergy/entropy control design and the derivative of the
ative of the Lyapunov function over a space of admissible Lyapunov function/Hamiltonian becomes
controls. In general, a set of feedback gains are optimized N
X M−N
X
by minimizing the regulating and/or tracking error of the V̇ = Ḣ = Ẇ − To Ṡi = Qj q̇j − Ql q̇l
feedback controller while regulating to zero and/or tracking j=1 l=N
a desired reference input. The Lyapunov function is the total which yields
error energy which for most mechanical systems is equivalent
To Ṡi =0
to an appropriate Hamiltonian function
Ẇ =0
V = H. (20) (To Ṡrev )ave = (M ẍ · ẋ + (K + KP )x · ẋ + KN L x3 · ẋ)ave
= 0.
Then the concept of Lyapunov Optimal [21] follows directly
from setting Ẇ = 0 in (19) and maximizing To Ṡi for which Numerical simulations are performed with the numerical
the time derivative of the Lyapunov function (Hamiltonian) or values listed in Table I. Note that for all cases that M = 10.0
the modified power (work/energy) equation is written as kg, K = 10.0 N/m, and KN L = 100.0 N/m3 . For this
initial Case 1 the phase plane plot and the potential and
N
X N
X kinetic energy rate plots are shown in Fig. 2 (top row).
V̇ = Ḣ = −To Ṡi = − Qj q̇j = − Fj Ṙj (21) This run demonstrates Corollary 1 and a stable orbit for the
j=1 j=1 nonlinear system with offsetting potential and kinetic energy
which is independent of system dynamics and is a kinematic rates responses.
quantity that applies to any system. Note that Fj denotes a set Next, consider the additional PID, linear, and Coulomb
of forces acting on a mechanical system and Ṙj denotes the friction effects applied to the Duffing oscillator and partition
inertial linear velocity of the point where Fj is applied. into exergy generation and exergy dissipation terms. Now
Passivity control [9] for robotic systems follows directly apply the exergy/entropy control design and the derivative of
from setting Ẇ = 0 in (19). the Lyapunov function/Hamiltonian becomes
N
X M−N
X
VII. R EGULATOR C ONTROL D ESIGN E XAMPLES V̇ = Ḣ = Ẇ − To Ṡi = Qj q̇j − Ql q̇l
j=1 l=N
Two nonlinear dynamic systems are investigated to demon-
strate exergy/entropy control design analogies for control which yields
design theory and to provide unique insights as well. These To Ṡi = (C + KD )ẋ · ẋ + CN L sign(ẋ) · ẋ
examples are based on 1) a PID regulator control for nonlinear Rt
Ẇ = −KI o xdτ · ẋ
Duffing oscillator/Coulomb friction dynamic system and 2) a
(To Ṡrev )ave = (M ẍ · ẋ + (K + KP )x · ẋ + KN L x3 · ẋ)ave
van der Pol nonlinear system.
= 0.
A. Duffing Oscillator/Coulomb Friction with PID Control To determine the nonlinear stability boundary from the ex-
System ergy/entropy control design
This example is the design of a control law for a single V̇ = Ḣ = Ẇ − To Ṡi
degree of freedom nonlinear oscillator. The Duffing oscilla-
which gives
tor/Coulomb friction dynamic model (see Fig. 1 - right) is
defined as (Ẇ )ave = (To Ṡi )ave .
Substituting the actual terms yields the following:
M ẍ + C ẋ + CN L sign(ẋ) + Kx + KN L x3 = u (22)  Z t 
where M, C, K, and u are the mass, damper, stiffness co- −KI xdτ · ẋ = [(C + KD )ẋ · ẋ + CN L sign(ẋ) · ẋ]ave
o ave
efficients and external force input terms, respectively. The (24)
which is the nonlinear stability boundary. To best understand Corollary 4.3.
how the boundary is determined, concepts and analogies
from electric AC power have been introduced earlier. Essen- B. Van der Pol Nonlinear System
tially, when the average powerin is equivalent to the average The classic van der Pol’s equation [22] is analyzed using
powerdissipated over a cycle, then the system is operating the techniques of this section. Originally, the “van der Pol
at the stability boundary. Later, in the exergy and exergy equation” is credited to van der Pol, and is a model of an
rate responses for the nonlinear system, one may observe electronic circuit for early radio vacuum tubes of a triode
that the area under the curves for the exergy rate generation electronic oscillator [22]. The tube acts like a normal resistor
and the exergy rate dissipation are equivalent and for the when the current is high, but acts as a negative resistor if the
corresponding exergy responses the slopes will be equal and current is low. The main feature is that electrical circuits that
opposite. This helps to explain why PID control works well contain these elements pump up small oscillations due to a
for nonlinear systems. negative resistance when currents are small, but drag down
Numerical simulations are performed to demonstrate where large amplitude oscillations due to positive resistance when
the nonlinear stability boundary lies for the Duffing oscilla- the currents are large. This behavior is known as a relaxation
tor/Coulomb friction dynamic model subject to PID control. oscillation, as each period of the oscillation consists of a slow
Three separate cases are conducted with the numerical values buildup of energy (’stress phase’) followed by a phase in
listed in Table I. The nonlinear system is subject to an initial which energy is discharged (’relaxation phase’). This particular
condition of x0 = 1.0. For Case 2 the integral of position, system has played a large role in nonlinear dynamics and has
position, velocity, and acceleration responses along with the been used to study limit cycles and self-sustained oscillatory
exergy and exergy rate responses are plotted in Fig. 2 (second phenomena in nonlinear systems.
row from top). For this case, the dissipative term is greater Consider the van der Pol equation with mass (m) and
than the generative term. This is observed from the decaying stiffness (k) values other than unity and a nonlinear damping
system responses. In Case 3 the system responses along with term (µ) to be defined as:
the exergy and exergy rate responses are shown in Fig. 2 (third
mẍ + µ(1 − x2 )ẋ + kx = 0.
row from top). In this case, the average exergy slopes and
integrated power areas for the dissipative and generative terms The appropriate Hamiltonian/Lyapunov function is defined as:
are equivalent which demonstrates (24). This results in system 1 1
responses that do not decay, displaying constant nonlinear H=V = mẋ2 + kx2 > 0.
2 2
oscillatory behavior. In final Case 4, the system responses
along with the exergy and exergy rate responses are shown in Then the corresponding time derivative of the Lyapunov
Fig. 2 (bottom row). In this case, the dissipative term is less function/Hamiltonian becomes
than the generative term which results in a system response V̇ = [mẍ + kx]2ẋ
with increasing nonlinear oscillatory behavior. In conclusion, = µẋ(1 − x ) ẋ
Fig. 3 shows the responses for the total exergy with respect = µẋ2 − µx2 ẋ2 .
to each case along with the phase plane plot for the nonlinear
Next identifying the generator and dissipator terms yields
system. For Case 3 the nonlinear stability boundary (or neutral
stability) is characteristic of an average zero output for the To Ṡi = µx2 ẋ2
total exergy response or validation of (24). For the phase plane Ẇ = µẋ2
plot, Case 2 demonstrates an asymptotically stable decaying (To Ṡrev )ave = (mẍ · ẋ + kx · ẋ)ave = 0.
response, Case 3 a neutrally stable orbital response, and Case
The nonlinear stability boundary can be determined as
4 an asymptotically unstable increasing orbit response. h i h i
The last three cases for the PID control regulator Duffing Ẇ = To S˙i
oscillator/Coulomb friction dynamic system demonstrates the  2 ave
  ave

µẋ ave = µx2 ẋ2 ave
three subcases for Corollary 4: Given apriori (To Ṡi )ave >
0 and (Ẇ )ave > 0 then the nonlinear system showed the By investigating several initial conditions both inside, on,
following: and outside the limit cycle then three separate regions can
i. Case 2 yielded (To Ṡi )ave > (Ẇ )ave ; asymptotic sta- be observed. Figure 4 shows these conditions with the corre-
bility; damped stable nonlinear response and demon- sponding numerical values given in Table I. The responses are
stration of Corollary 4.1. plotted on the Hamiltonian 3D surface (top) with the projection
ii. Case 3 yielded (To S˙i )ave = (Ẇ )ave ; neutral stabil- onto the phase plane shown on the 2D plot (middle). For the
ity; and demonstration of Corollary 4.2. This case case outside the limit cycle, the dissipator term dominates
is the nonlinear stability boundary where dissipation and for the case inside the limit cycle the generator term
and generation terms cancel each other out on the dominates. For both cases inside and outside the limit cycle,
average. the system migrates back to the stability boundary. For the case
iii. Case 4 yielded (To Ṡi )ave < (Ẇ )ave ; increasingly already on the limit cycle then the system is already at neutral
unstable towards another orbit; and demonstration of stability. The neutral exergy-rate and exergy plots are shown in
Fig. 4 (bottom). The cycle is defined at approximately τ = 3.5 [13] K.-H. Anthony, Hamilton’s Action Principle and Thermodynamics
seconds. For the neutral pair the terms cancel each other out at of Irreversible Processes - A Unifying Procedure for Reversible and
Irreversible Processes, J. Non-Newtonian Fluid Mechanics, Vol. 96,
the end of the cycle or [Ẇ ]ave = [To Ṡi ]ave . For the generator 2001, pp. 291-339.
case then [Ẇ ]ave > [To Ṡi ]ave and for the dissipator case then [14] D.S. Scott, Links and Lies, International Journal of Hydrogen Energy,
[Ẇ ]ave < [To Ṡi ]ave , respectively. Eventually, given enough Vol. 28, 2003, pp. 473-476.
[15] D. Kondepudi and I. Prigogine, Modern Thermodynamics: From Heat
cycles both the generator and dissipator cases will converge Engines to Dissipative Structures, John Wiley & Sons, N.Y., N.Y., 1999.
to the neutral case. [16] A. Greven, G. Keller, and G. Warnecke, Entropy, Princeton University
Press, New Jersey, 2003.
VIII. S UMMARY AND C ONCLUSIONS [17] D.S. Scott, Exergy, International Journal of Hydrogen Energy, Vol. 28,
2003, pp. 369-375.
A novel control system design methodology was devel- [18] L. Meirovitch, Methods of Analytical Dynamics, McGraw-Hill, New
oped that uniquely combined: concepts from thermodynamic York, 1970.
[19] R.J. Smith, Circuits, Devices, and Systems: A First Course in Electrical
exergy and entropy; Hamiltonian systems; Lyapunov’s direct Engineering, John Wiley & Sons, Third Edition, 1976.
method and Lyapunov optimal analysis; electric AC power [20] T.L. Saaty and J. Bram, Nonlinear Mathematics, McGraw-Hill, New
concepts; and power flow analysis. Relationships were derived York, 1964.
[21] R.D. Robinett, III, G.G. Parker, H. Schaub, and J.L. Junkins, Lyapunov
between exergy/entropy and Lyapunov optimal functions for Optimal Saturated Control for Nonlinear Systems, AIAA Journal of
Hamiltonian systems. The methodology is demonstrated with Guidance, Control, and Dynamics, Vol. 20, No. 6, Nov-Dec 1997, pp.
two fundamental numerical simulation examples: 1) a Duffing 1083-1088.
[22] B. van der Pol, Radio Rev. 1, 704-754, 1920 and B. van der Pol, Phil.
oscillator/Coulomb friction nonlinear model that employs PID Mag., 3, 65, 1927.
regulator control and 2) a van der Pol nonlinear oscillator
system. The control system performance results and/or appro- x
Se C, C NL
priately identified terms were partitioned and evaluated based W
IN W
OUT u
Q M
on exergy generation and exergy dissipation terms. These IN E Q
OUT
Si
m K, K
numerical results showed the stability boundaries for each IN m
OUT
NL

nonlinear system. This novel nonlinear control methodology

Fig. 1. Energy flow control volume (left), second law entropy with
resulted in both necessary and sufficient conditions for stability flux exchange system (middle), and Duffing oscillator/Coulomb
of nonlinear systems. In the near future, this novel control friction system (right)
system design methodology will be extended to tracking and
adaptive control of multi-input/multi-output nonlinear systems.
TABLE I
This methodoloy is applicable to a large class of nonlinear
systems. Duffing oscillator/Coulomb friction model and PID control
system gains
R EFERENCES
. Case KP KI KD C CNL
[1] R.M. Murray, Ed., Control in an Information Rich World: Report of the
No. (kg/s2 ) (kg/s3 ) (kg/s) (kg/s) (N)
Panel on Future Directions in Control, Dynamics, and Systems, SIAM,
Philadelphia, PA, 2003.
[2] P.J. Moylan, Implications of Passivity in a Class of Nonlinear Systems, 1 10.0 0.0 0.0 0.0 0.0
IEEE Transactions of Automatic Control, Vol. AC-19, No. 4, August 2 10.0 20.0 2.0 0.1 5.0
1974, pp. 373-381. 3 10.0 40.05 2.0 0.1 5.0
[3] J.L. Wyatt, Jr., L.O. Chua, J.W. Gannett, I.C. Goknar, and D.N. Green, 4 10.0 80.0 2.0 0.1 5.0
Energy Concepts in the State-Space Theory of Nonlinear n-Ports: Part
I - Passivity, IEEE Transactions on Circuits and Systems, Vol. CAS-28,
No. 1, January 1981, pp. 48-61. TABLE II
[4] J.L. Wyatt, Jr., L.O. Chua, J.W. Gannett, I.C. Goknar, and D.N. Green, Van der Pol model numerical values
Energy Concepts in the State-Space Theory of Nonlinear n-Ports: Part II
- Losslessness, IEEE Transactions on Circuits and Systems, Vol. CAS-
29, No. 7, July 1982, pp. 417-430. Case xo ẋo µ m k
[5] R. Ortega, Z.P. Jiang, and D.J. Hill, Passivity-Based Control of Nonlinear (m) (m/s) (kg/s) (kg) (kg/s2 )
systems: A Tutorial, Proceedings of the American Control Conference,
Albuquerque, NM, June 1997, pp. 2633-237.
generate 0.1 −0.1 1.5 1.0 1.0
[6] J.C. Willems, Dissipative Dynamical Systems Part I: General Theory;
neutral 1.0 −1.0 1.5 1.0 1.0
Part II: Linear Systems with Quadratic Supply Rates, Archive for
dissipate 2.0 −2.0 1.5 1.0 1.0
Rational Mechanics and Analysis, Vol. 45, pp. 321-393, 1972.
[7] D. Hill and P.J. Moylan, The Stability of Nonlinear Dissipative Systems,
IEEE Transactions on Automatic Control, October 1976, pp. 708-710.
[8] P. Kokotovic and M. Arcak, Constructive Nonlinear Control: A Histor-
ical Perspective, Preprint submitted to Elsevier, August 2000.
[9] J.-J. E. Slotine and W. Li, Applied Nonlinear Control, Prentice Hall,
Inc., N.J., 1991.
[10] M. Kristic, I. Kanellakopoulos, and P. Kokotovic, Nonlinear and
Adaptive Control Design, John Wiley & Sons, Inc., New York, 1995.
[11] P.A. Ioannou and J. Sun, Robust Adaptive Control, Prentice Hall,
Englewood Cliffs, N.J., 1996.
[12] A.A. Alonso and B.E. Ydstie, Stabilization of Distributed Systems Using
Irreversible Thermodynamics, Automatica, Vol. 37, 2k1, pp. 1739-55.
 Duffing Oscillator Phase Plane Duffing Oscillator
3 150
Potential/Kinetic Energy Rates (W)
3
(K+K )xv+K xv
P NL
2 100 (Ma)v
1 50
v (m/s)

0 0

−1 −50

−2 −100

−3 −150
−1 −0.5 0 0.5 1 0 2 4 6 8 10
x (m) Time (sec)
Integral, Position, Velocity and Acceleration Integral, Position, Velocity and Acceleration Integral, Position, Velocity and Acceleration

CASE 2 PID Control Nonlinear System CASE 2 PID Control Nonlinear System
Exergy (Joules) − Exergy Rate (Watts)

15 60
x (m−s)
int
10 x (m) 40
v (m/s) 20
5 a (m/s/s)
0
0
−20 T S damper
o i
−5 W integral
−40 T d/dt(S) damper
o i
−10 −60 d/dt(W) integral H = 0.5*k*x2 + 0.5*m*xdot2 4
generate
−15 −80 neutral
0 2
4 6 8 10 0 2 4 6 8 10 2 dissipate
Time (sec) Time (sec)

Xdot (m/s)
CASE 3 PID Control Nonlinear System CASE 3 PID Control Nonlinear System
Hamiltonian (J)
Exergy (Joules) − Exergy Rate (Watts)

15 200 0
x (m−s) 5
int 0 4
10 x (m) 4
v (m/s) 100 2 −2
2
5 a (m/s/s) 0
0
−2 −2 −4
0 0 Xdot (m/s) X (m) −3 −2 −1 0 1 2 3
−4 −4 X (m)
−5 T S damper 30 30
o i
Sdoti
−100 W integral ∫ Sdoti d τ
Wdot 20
−10 T d/dt(S) damper 20 ∫ Wdot d τ
Exergy Rate (Watts)

o i
Exergy (Joules)

d/dt(W) integral 10
−15 −200 10
0 2 4 6 8 10 0 2 4 6 8 10
Time (sec) Time (sec) 0
CASE 4 PID Control Nonlinear System CASE 4 PID Control Nonlinear System 0
Exergy (Joules) − Exergy Rate (Watts)

60 800 −10
−10 −20
40 600 Neutral Case
Neutral Case
400 −20 −30
20 0 2 4 6 8 0 2 4 6 8
Time (sec) Time (sec)
200
0 Fig. 4. Van der Pol responses - 3D Hamiltonian, phase plane plot
0
−20 x
int
(m−s) T S damper
o i (top), and exergy-rate and exergy plots (bottom)
−200
x (m) W integral
−40 v (m/s) −400 T d/dt(S) damper
o i
a (m/s/s) d/dt(W) integral
−60 −600
0 2 4 6 8 10 0 2 4 6 8 10
Time (sec) Time (sec)

Fig. 2. Cases 1-4: Duffing oscillator/Coulomb friction with PID

control numerical results

All Cases PID Control Nonlinear System All Cases Phase Plane Nonlinear System
250 10
EX Case 2
Case 2
200 EXCase 3 Case 3
Total Exergy (Joules)

5 Case 4
150 EX
Case 4
v (m/sec)

100 0

50
−5
0

−50 −10
0 2 4 6 8 10 −2 −1 0 1 2
Time (sec) x (m)

Fig. 3. Cases 2:4 - Duffing oscillator/Coulomb friction numerical

results