Recent advances in multi-body dynamics and nonlinear control

TECHNICAL PAPERS

Recent advances in multi-body dynamics and nonlinear control

Firdaus E. Udwadia

Aerospace and Mechanical Engineering; Civil Engineering,, Mathematics; Systems Architecture Engineering and Information and Operations Management; University of Southern California, Los Angeles; CA 90089-1453; [e-mail]

ABSTRACT

This paper presents some recent advances in the dynamics and control of constrained multi-body systems. The constraints considered need not satisfy D'Alembert's principle and therefore the results are of general applicability. They show that in the presence of constraints, the constraint force acting on the multi-body system can always be viewed as made up of the sum of two components whose explicit form is provided. The first of these components consists of the constraint force that would have existed were all the constraints ideal; the second is caused by the non-ideal nature of the constraints, and though it needs specification by the mechanician who is modeling the specific system at hand, it nonetheless has a specific form. The general equations of motion obtained herein provide new insights into the simplicity with which Nature seems to operate. They are shown to provide new and exact methods for the tracking control of highly nonlinear mechanical and structural systems without recourse to the usual and approximate methods of linearization that are commonly in use.

Keywords: Constrained motion, multi-body dynamics, explicit equations of motion, exact tracking control of nonlinear systems

Introduction

The general problem of obtaining the equations of motion of a constrained discrete mechanical system is one of the central issues in multi-body dynamics. While it was formulated at least as far back as Lagrange (1811), the determination of the explicit equations of motion, even within the restricted compass of lagrangian dynamics, has been a major hurdle. The Lagrange multiplier method relies on problem-specific approaches to the determination of the multipliers which are often difficult to obtain for systems with a large number of degrees of freedom and many non-integrable constraints. Formulations offered by Gibbs (1879), Appell (1899), and Poincare (1901) require a felicitous choice of problem specific quasi-coordinates and suffer from similar problems in dealing with systems with large numbers of degrees of freedom and many non-integrable constraints. Gauss (1829) developed a general principle governing constrained motion for systems that satisfy D'Alembert's principle, and Dirac (1964) has offered a formulation for hamiltonian systems with singular lagrangians where the constraints do not explicitly depend on time.

The explicit equations of motion obtained by Udwadia and Kalaba (1992) provide a new and different perspective on the constrained motion of multi-body systems. They introduce the notion of generalized inverses in the description of such motion and, through their use, obtain a simple and general explicit equation of motion for constrained multi-body mechanical systems without the use of, or any need for, the notion of Lagrange multipliers. Their approach has allowed us, for the first time, to obtain the explicit equations of motion for multi-body systems with constraints that may be: (1) nonlinear functions of the velocities, (2) explicitly dependent on time, and, (3) functionally dependent. However, their equations deal only with systems where the constraints are ideal and satisfy D'Alembert's principle, as do all the other formulations/equations developed so far (e.g., Lagrange (1811), Gibbs (1879), Appell (1899), Poincare (1901), Gauss (1829), Dirac (1964), Chataev (1989), and Synge (1927)). D'Alembert's principle says that the motion of a constrained mechanical system occurs in such a way that at every instant of time the sum total of the work done under virtual displacements by the forces of constraint is zero.

In this paper we extend these results along two directions. First, we extend D'Alembert's Principle to include constraints that may be, in general, non-ideal so that the forces of constraint may therefore do positive, negative, or zero work under virtual displacements at any given instant of time during the motion of the constrained system. We thus expand lagrangian mechanics to include non-ideal constraint forces within its compass. Second, the explicit equations of motion are obtained. They lead to deeper insights into the way Nature seems to work. With the help of these equations we provide a new fundamental, general principle governing constrained multi-body dynamics.

Nomenclature

A = m by n matrix

a = n-component acceleration vector of unconstrained system

B = AM^–1/2

b = m-component vector

B^T = Transpose of matrix B

B⁺ = generalized Moore-Penrose inverse of matrix B

C = given n-component vector describing work done by non-ideal constraints

c = M^–1C

h = number of holonomic constraints

M = positive definite n by n mass matrix

m = total number of constraints, holonomic and nonholonomic

n = number of generalized coordinates

Q = right hand side of unconstrained equation of motion

q = n-component vector of generalized coordinates

q_i = component of generalized coordinate vector

= vector of generalized velocity

= vector of generalized acceleration

Q^c = constraint force , or control force, n-component vector

= n-component constraint force vector cause by ideal constraints

= n-component constraint force cause by presence of non-ideal constraints

t = time

Greek Symbols

D = difference between acceleration of constrained system and that of unconstrained system

j_i = i-th constraint

µ = M^1/2v

v = n-component virtual displacement vector

Subscripts

i relative to ideal constraint force

ni relative to nonideal constraint force

0 relative to initial time

Superscripts

T the transpose of a matrix

+ the Moore-Penrose inverse of a matrix

c the constraint force, or, the control force

Statement of the Problem of Constrained Motion

Consider first an unconstrained, multi-body system whose configuration is described by the n generalized coordinates q = [q₁,q₂,...,q_n]^T. By 'unconstrained' we mean that the components, _i, of the velocity of the system can be independently assigned at any given initial time, say, t=t₀. Its equation of motion can be obtained, using newtonian or lagrangian mechanics, by the relation

Where the n by n matrix M is symmetric and positive definite. It is indeed possible for the matrix M to be singular when dealing with some special substructure models of multi-body systems, but we shall not discuss this here. We refer the reader to Udwadia and Phohomsiri (2006) for the general, constrained equations of motion when such nonsingular matrices appear in the formulation of the unconstrained equations of motion of a mechanical system. The matrix M(q,t) and the generalized force n-vector (n by 1 matrix), Q(q,,t) are known. In this paper, by 'known' we shall mean known functions of their arguments. The generalized acceleration of the unconstrained system, which we denote by the n-vector a, is then given by

We next suppose that the system is subjected to h holonomic constraints of the form

and m – h nonholonomic constraints of the form

The initial conditions q₀ = q(t=t₀) and are assumed to satisfy these constraints so that j_i(q₀,t₀)=0, i = 1, 2, . . . , h, and , i = h + 1, h + 2,...,m. These constraints encompass all the usual holonomic and nonholonomic constraints (or combinations thereof) that the multi-body system may be subjected to. We note that the constraints may also be explicit functions of time, and the nonholonomic constraints may be nonlinear in the velocity components _i. Under the assumption of sufficient smoothness, we can differentiate equations (3) twice with respect to time and equations (4) once with respect to time to obtain the consistent equation set

where the constraint matrix, A, is a known m by n matrix and b is a known m-vector. It is important to note that for a given set of initial conditions, equation set (5) is equivalent to equations (3) and (4), which can be obtained by appropriately integrating the set (5).

The presence of the constraints (5) imposes additional forces of constraint on the multi-body system that alter its acceleration so that the explicit equation of motion of the constrained system becomes

The additional term, Q^c, on the right hand side arises by virtue of the imposed constraints prescribed by equations (5).

We begin by generalizing D'Alembert's Principle to include forces of constraint that may do positive, negative, or zero work under virtual displacements.

We assume that for any virtual displacement vector, v(t), the total work done, , by the forces of constraint at each instant of time t, is prescribed (for the given, specific dynamical system under consideration) through the specification of a known n-vector C(q,,t) such that

Equation (7) reduces to the usual D'Alembert's Principle when C(t) º 0, for then the total work done under virtual displacements is prescribed to be zero, and the constraints are then said to be ideal. In general, the prescription of C is the task of the mechanician who is modeling the specific constrained system whose equation of motion is to be found. It may be determined for the specific system at hand through experimentation, analogy with other systems, or otherwise. We include the situation here when the constraints may be ideal over certain intervals of time and non-ideal over other intervals. Also, W at any given instant of time may be negative, positive, or zero, allowing us to include multi-body systems where energy may be extracted from, or fed into, them through the presence of the constraints. We shall denote the acceleration of the unconstrained system subjected to this prescribed force C by c(q,,t)= M^–1C. In what follows, we shall omit the arguments of the various quantities, except when needed for clarity.

Equations of Motion for Constrained Systems

We begin by stating our result for the constrained multi-body system described above. For convenience we state it in two equivalent forms (Udwadia and Kalaba, 2002a and 2002b)

1. The explicit equation of motion that governs the evolution of the constrained system is:

or

Equation (9) can also be expressed as

In equations (8)-(10), B = AM^-1/2, and B⁺ denotes the Moore-Penrose inverse of the constraint matrix A (Udwadia and Kalaba., 1996); D(t) denotes the deviation of the acceleration of the constrained system, , at time t from its unconstrained value, a(t), at that time; and, the quantity e(t):= (b – Aa) represents the extent to which the acceleration a, at the time t, corresponding to the unconstrained motion does not satisfy the constraint equation (5). Later on, from a controls perspective we will call e(t) the 'error signal.'

2. At each instant of time

   t, the total force of constraint,

   Q^c, is made up of two additive parts. The first part,

   , is the force of constraint that would have been generated were the constraints ideal at the time

   t; the second part,

   , is created by the non-ideal nature of the constraints at the time

   t. These two contributions to the total constraint force are explicitly given by

and

where Q^{c =} + . The subscripts i and ni refer to ideal and non-ideal, respectively. When C(t) = 0, the constraints are all ideal and then Q^{c =} .

Equation (10) leads to the following new fundamental principle of motion for constrained multi-body mechanical systems:

The motion of a discrete dynamical system subjected to constraints evolves, at each instant in time, in such a way that the deviation in its acceleration from what it would have at that instant if there were no constraints on it, is the sum of two M-orthogonal components; the first component is directly proportional to the extent, e, to which the accelerations corresponding to its unconstrained motion, at that instant, do not satisfy the constraints, the matrix of proportionality being M^–1/2B⁺; and, the second component is proportional to the given n-vector c, the matrix of proportionality being M^–1/2(I – B⁺ B)M^1/2.

We define two n-vectors u and w to be M-orthogonal if u^TMw = 0. Since the Moore-Penrose inverse of a matrix, B⁺, may be unfamiliar to some, I provide here some of its properties, which will be used later on. Given an m by n matrix B⁺, the n by m matrix is a unique matrix that satisfies the following four relations:

As stated in our fundamental principle above, the two components of acceleration engendered by the presence of the constraints are explicitly given by the last two members on the right hand side of equation (9). The M-orthogonality of these two members follows from the relations where we have used relation 13(c) in the second equality and equation 13(b) in the last.

The derivation of our result is as follows.

The acceleration, , of the constrained system must satisfy two requirements. It must be such that :

(1) at each instant of time, t, it must satisfy the constraints given by equation (5), and,

(2) the work W done under any virtual displacement by the force of constraint, , must, at each instant of time t, be as prescribed by relation (7).

Since we require the acceleration of the constrained system to satisfy the consistent set of equations. A = A(D+a) = B(M^1/2D) + Aa= b, we have, from the theory of generalized inverses,

where z is any arbitrary n-vector, and B⁺ is the Moore-Penrose inverse (of the matrix B = AM^–1/2) whose properties are described in equations 13(a)-13(d). From equation (14) we then have

so that

To explicitly find , we next determine the second member on the right in equation (16) in such a way as to ensure that the second of the above-mentioned requirements is satisfied.

A virtual displacement at time t is any displacement that satisfies the relation An = 0 at that time (Udwadia, et al., 1997). Since Av= B(M^1/2v) := Bµ the explicit solution of the homogeneous set of equations Bµ=0 is simply

where y is any arbitrary n-vector. And so from relation (7), we require that

where, at each instant of time, C is specified by the mechanician who is modeling the specific mechanical system. Using equation (17) in the last equality in (18) we get

which, because y is arbitrary, yields

Relation (20) follows from (19) through the use of relations (13d) and (13b) because

and,

Using (20) we then get

which when used in the second member on the right in equation (16) gives

and the result given by equation (8) now follows from equation (15).

The explicit equations of motion obtained herein, like those obtained earlier for ideal constraints (Udwadia and Kalaba 1992), are completely innocent of the notion of Lagrange multipliers. Over the last 200 years, Lagrange multipliers have been so widely used in the development of the equations of motion of constrained multi-body systems that it is sometimes tempting to mistakenly believe that they have an instrinsic presence in the description of constrained motion. This is not true. As shown in this paper, neither in the formulation of the physical problem of the motion of constrained multi-body systems nor in the equations governing their motion are any Lagrange multipliers involved. The use of Lagrange multipliers (a mathematical tool invented by Lagrange (1811)) constitutes just one of the several intermediary mathematical devices invented for handling constraints. And, in fact, the direct use of this device appears difficult when the constraints are functionally dependent. Lagrange multipliers do not appear in the physical description of constrained motion, and therefore cannot, and do not, ultimately appear in the equations governing such motion.

Conclusions

The simplicity of the general explicit equation of motion obtained herein relies on the interplay of four central observations:

(1) No transformation of coordinates, or their elimination, is undertaken when constraints are present; the coordinates in which the unconstrained multi-body system is described are the same as those used to describe the constrained system. This, at first, appears to be counter-intuitive and indeed goes against a 200 year-old, well-accepted current of practice in dynamics and theoretical physics that was first initiated by Lagrange. Such transformations and eliminations are often useful in handling problems of mathematical physics. However, it is the fact that we do not use them that appears to be ultimately responsible for the simplicity of the explicit equation obtained herein, and the fundamental insights about the nature of constrained motion provided by it.

(2) The constraints are described in their differentiated form by equation (5); this a consequence of the realization that, at any instant of time t, the 'state' of the system, (q(t)

(t)), is assumed known, and it is the state immediately following this instant that must then be the focus of our inquiry. Our attention must then naturally focus on the system's acceleration, .

(3) For a physical system where the constraint forces do work the equations of motion cannot be obtained solely through knowledge of the kinematical nature of the constraints as described by equations (3) and (4); one needs to have an additional dynamical characterization of the constraints given by the extension of D'Alembert's principle (or some equivalent of it), as stated in equation (7). Such a characterization yields a unique equation of motion, as expected from, and consistent with, practical observation.

(4) The Moore-Penrose inverse of a matrix shows an intrinsic presence in the equations of motion. It manages to sort out the manner in which the constraints interact with the given forces (known acceleration, a(t)) to yield an equation of motion that is both simple and provides new physical insights.

Lastly, it is worth mentioning that the general equations of motion obtained here have immediate application to the tracking-control of nonlinear multi-body systems (Udwadia, 2003, 2004) — a problem that has been worked on for many decades, with weak success, by control theorists. For, the constraint force can be interpreted as the control force required to be applied to the nonlinear multi-body system which is described by equation (1) so that it 'exactly' satisfies the trajectory requirements imposed by equation (5) (equivalently, by equations (3) and (4)) at each instant of time. One then obtains the closed-form control force, given by equation (24). And this for a general, nonlinear multi-body system! In fact, this control force is exactly what Nature "would use" were it required to satisfy the constraint equations (3) and (4) (also thought of now as the trajectory requirements!) along with relation (7). Furthermore, were we to set Cº0 (the ideal constraint case), we would obtain the force that Nature would employ to control the nonlinear multi-body system described by equation (1) with (ideal) constraints described by equations (3) and (4). We would then have

And so we see that Nature appears to be actually behaving much as a control engineer would! For, the second member on the right in equation (25) can be thought of as providing 'feedback control,' using feedback proportional to the 'error signal' e(t) := (b– Aa) , which measures the extent to which the acceleration that we know at time t, namely a(t), does not satisfy our trajectory requirement (5). However, it is in the choice of the 'gain matrix,' M^1/2(q,t)B⁺(q,,t), that Nature seems to really excel! She picks the control gain with incredible ingenuity so as to exactly satisfy the trajectory requirement (5) at each instant of time. It is the choice of this matrix, which, in general, is a highly nonlinear function of q,, and t, that would most likely baffle our best control theorists! Such reinterpretations of the equations obtained in this paper within the framework of control theory show their considerable scope of applicability and utility. The details of this approach to the control of multi-body systems (accuracy and robustness, etc.) would be too long a story to present here. The interested reader may find them in Udwadia (2003, 2004).

In this paper we have thus extended the lagrangian formulation of mechanics to include constraints that may be ideal and/or non-ideal, and the equations of motion presented in this paper are applicable to multi-body mechanical systems that include such constraints. They appear to be the simplest and most comprehensive equations of motion so far discovered for such systems.

Paper accepted: June, 2005.

Presented at XI DINAME – International Symposium on Dynamic Problems of Mechanics, February 28th - March 4th, 2005, Ouro Preto. MG. Brazil.

Technical Editor: José Roberto de França Arruda.

• Appell, P., 1899, "Sur une Forme Generale des Equations de la Dynamique," C. R. Acad. Sci, Paris, 129 pp. 459-460.
• Chetaev, N. G., 1989, "Theoretical Mechanics, Mir Publisher, Moscow".
• Dirac, P. A. M., 1964, "Lectures in Quantum Mechanics", Yeshiva University, New York.
• Gauss, C. F., 1829, "Uber Ein Neues Allgemeines Grundgsetz der Mechanik," J. Reine Ang. Math, Vol. 4, pp. 232-235.
• Gibbs, J. W., 1879 "On the Fundamental Formulae of Dynamics," Amer. Journ. Math, Vol. 2, pp. 49-64.
• Lagrange J. L., 1811, "Mechanique analytique, Paris: Mme Ve Coureier".
• Poincare H. 1901, "Su rune forme nouvelle des equations de la mechanique," C. R. Acad. Sci, Paris, Vol. 132, pp. 369-371.
• Synge J. L., 1927, "On the Geometry of Dynamics," Philosophical Transactions of the Royal Society of London, Series A, Vol. 226, pp. 31-106.
• Udwadia F. E. and Kalaba R. E., 1992, "A new perspective on constrained motion," Proceedings of the Royal Society of London, Series A, Vol. 439, pp. 407-410.
• Udwadia, F. E., and Kalaba, R. E., 2002a, "On the Foundations of Analytical Dynamics," International Journal of Non-Linear Mechanics, Vol. 37, pp. 1079-1090.
• Udwadia, F. E., and Kalaba, R. E., 2002b, "What is the General Form of the Explicit Equations of Motion for Mechanical Systems?," Journal of Applied Mechanics, Vol. 39, pp. 335-339.
• Udwadia, F. E., and Kalaba, R. E., 1996, Analytical Dynamics: A New Approach, Cambridge University Press, Cambridge, UK.
• Udwadia F. E., Kalaba R. E., and Hee-Chang E., 1997, "Equations of Motion for Constrained Mechanical Systems and the Extended D'Alembert Principle," Quarterly of Applied Mathematics, Vol. 56, pp. 321-331.
• Udwadia F. E., 2003, "A New Perspective on the Tracking Control of Nonliner Structural and Mechanical Systems," Proceedings of the Royal Society of London, Series A, Vol. 459, pp. 1783-1800.
• Udwadia F. E., 2004, "Exact Tracking Control for Nonlinear Structural and Mechanical Systems," Proceedings of the International Congress on Applied and Theoretical Mechanics, Warsaw, Poland, Aug. 15-21.
• Udwadia F. E. and P. Phohomsiri, 2006, "Equations of Motion for Systems with Singular Mass Matrices and Their Application to Multi-body Dynamics", to appear in Proceedings of the Royal Society of London, Series A.

Publication Dates

• Publication in this collection
  18 Aug 2006
• Date of issue
  Sept 2006