JOURNAL OF THEORETICAL AND APPLIED MECHANICS 43, 3, pp. 593-607, Warsaw 2005 MODELLING AND CONTROL OF MECHATRONIC SYSTEMS BY THE DESCRIPTOR APPROACH Peter C. Müller Safty Control Engineering, University of Wuppertal, Germany e-mail: mueller@srm.uni-wuppertal.de In recent years, the analysis and synthesis of control systems in a de- scriptor formhas been established.The general description of dynamical systems by differential-algebraic equations (DAE) is important formany applications in various disciplines, but particularly in mechatronics. In this contribution, the pros and cons of themodelling ofmechatronic sys- tems by differential-algebraic equations are discussedwith application of subsystemmodelling.Additionally, the actual state of the art simulation, analysis and design of descriptor systems are presented. Key words: descriptor systems, mechatronic systems, optimal control 1. Modelling The investigation of dynamical systems inmechanical or electrical engine- ering usually requires mathematical modelling of the system behaviour. The increasing complexity of these processes leads on the one hand to the deve- lopment of computer programs automatically generating the governing system equations (Schiehlen, 1990) for multibody systems, or on the other hand to an increase of modular subsystem modelling of which the complete model is composed. Usually, this interconnection-oriented modelling describes dyna- mic behaviour of single components by differential equations and the coupling of the subsystems by algebraic equations. All over, a mathematical model is represented by a combined set of differential and algebraic, i.e. by differential- algebraic equations (DAE). In control engineering, we speak about singular control systems or descriptor systems (Luenberger, 1977). Mechatronic systems usually consist of a large number of mechanical, hy- draulic, electrical and electronic components where the subsystem modelling 594 P.C.Müller represents the most clear and manageable way of modelling that maintains the physical character of the components. For example, electrical networks can be considered to be composed of subsystems of network elements (like re- sistors, capacitors, inductors described by different types of equations) and by coupling due toKirchhoff’s laws (describedby algebraic equations) (Kampow- ski et al., 1992; Mathis, 1992). In mechanical systems, differential equations usually describe the dynamics of subsystems and algebraic equations charac- terise couplings by constraints such as joints. A general approach to handle mechanical systems as an interconnected set of dynamic modules has been given in Rüekgauer and Schiehlen (1997). In the following, three examples of the descriptor modelling are dealt with for illustration. 1.1. Lagrange’s equations of the first kind Lagrange’s equations of the first and second kind are well established in analytical mechanics (Rosenberg, 1977). They describe dynamic behaviour of discrete systems, particularly of multibody systems. The difference of the two kinds consists of the manipulation of kinematic constraints. If a kinematic description of the system has been performed by generalised coordinates con- sistent with the constraints, Lagrange’s equation of the second kind can be applied which leads to a set of differential equations only. But if a redundant set of coordinates is used to describe kinematically the system regarding still some constraints explicitly, then the Lagrange’s equations of the first kind hold. In the case of holonomic constraints f(q)=0 (1.1) we have d dt (∂L ∂q̇ ) − ∂L ∂q =Q+F>λ (1.2) where the Lagrangian function L=T−U consists of the kinetic andpotential energies T and U, Q represents nonconservative forces acting on the system. F = F(q) = ∂f/∂q> is the Jacobian matrix of the constraints and λ is the vector of Lagrange’s multiplyers. They represent the constraint forces if the column vectors of F> are normalized. While the variables q describe motion of the system, the Lagrange’s multiplyers λ give some information on the load of the mechanical structure. Therefore, critical loads due to motion may be considered simultaneously. Equations (1.1) and (1.2) represent a system of DAE. If Q includes some actuators to control the multibody system, then a descriptor system is under consideration. Modelling and control of mechatronic systems... 595 1.2. Subsystem modelling If the interconnection-oriented modelling approach is applied (Müller, 1995), usually the dynamics of N subsystems is described by sets of diffe- rential equations ẋi =ai(xi,ui) i=1, . . . ,N (1.3) where xi are the internal state vectors and ui the control vectors of the corre- sponding subsystems. The couplings among the subsystemsmay be obtained kinematically by ”constraints” or kinetically by ”forces” leading to ẋi =ai(xi,ui)+ N ∑ j=1 aij(xi,xj)+ N ∑ j=1 Lij(xj)λj (1.4) 0= N ∑ j=1 fij(xj) i=1, . . . ,N The additional terms compared to (1.3) are the kinetic couplings aij between subsystems no. i and j, and the kinematic couplings (1.4)2 which have to be considered in dynamic balance equations (1.4)1 by some Lagrange’s mul- tipliers λj with some input matrices Lij due to coupling requirements. How Lij is definedmore precisely depends on physical principles behind the system discipline; equations (1.1) and (1.2) show an example of mechanical systems. All over, equations (1.4) represent again the descriptor system. 1.3. Tracking control Inmechatronics, often the problem of tracking control arises, e.g. the pre- scribed path control of a robot. In this case, the process dynamics may be described in the state space by ẋ=a(x,u,t) (1.5) and it is asked for the control uwhich guarantees that some output variables y= c(x,u,t) follow a prescribed reference path yref(t) 0= c(x,u,t)−yref (1.6) This descriptor system (1.5) and (1.6) can be described by [ Ix 0 0 0 ][ ẋ ẋ ] = [ a(x,x,t) c(x,x,t)−yref(t) ] (1.7) 596 P.C.Müller which defines explicitly the desired tracking control u(t)= [ 0 Iu ] [ ẋ ẋ ] (1.8) where Ix, Iu are identity matrices of dim(x) and dim(u), respectively. Theexact tracking control (1.7) and(1.8) canbeextendedtoanasymptotic tracking control introducing for the tracking error e(t)=y(t)−yref(t) (1.9) additional error dynamics ė(t)=f(e) f(0)=0 (1.10) which are asymptotically stable in the large. Then (1.7) and (1.8) are replaced by    Ix 0 0 0 Iy 0 0 0 0       ẋ ė ẋ    =    a(x,x,t) f(e) c(x,x,t)−yref(t)−e    (1.11) u(t)= [ 0 0 Iu ]    ẋ ė ẋ    1.4. Advantages With respect to tasks of themodelling ofmechatronic systems, thedescrip- tor approach has many advantages. It is a very natural way to model process dynamics. It refers much more to physical behaviour of a system and gives more physical insight. The interpretation of results is alsomore simple than in a case of a more abstract description by state space models. In the opposite, the state space system approach was mainly required by mathematical tools available until 1980 to simulate, to analyse and to design such systems. But today, also many tools have been prepared for the simulation, analysis and control design of descriptor systems. 2. Simulation As long as it was not possible to simulate descriptor systems, the very efficient and very accurate state space approach was still superior according Modelling and control of mechatronic systems... 597 to well established tools of numerical integration of ordinary differential equ- ations. But in the 1970s, the simultaneous numerical solution to differential and algebraic equations was firstly considered (Gear, 1971). Step by step, nu- merical system solvers were developed. For index-1-problems (see below), the code DASSL was presented (Petzold, 1983), stimulating more research also for higher index problems. The first code for mechanical index-3-systems was presented by Führer (1988). In the meantime, a lot of efficient solvers for DAE have been developed (cf. Brenan et al., 1989; Hairer andWanner, 1991; Simeon, 1994). In the Ph.D. thesis by Rükgauer (1997) on themodular simu- lation of mechatronic systems, several solvers were compared resulting in the recommendation of the codes SDOP853 and SDOPPRI5 which are modified versions of Runge-Kutta solvers for ordinary differential equations including projection stepswith respect to constraints of the algebraic equation.A survey on solvers of higher index DAEs was given byArnold (1988). With respect to these results, today a number of stable and efficientDAE solvers exist and can be applied as naturally as ODE solvers for state space models. Such solvers are included inmany programpackages to generate and simulate equations of motion of dynamical systems, e.g. in ADAMS and SIMPACK (cf. Schiehlen, 1990), for multibody systems. 3. Analysis and synthesis The tools for the analysis and control design of descriptor systems have been developed enormously in the last two decades. As usual, linear theory was in the foregroundof thediscussion, but results onnonlinear problemswere reported as well. In the following, the well established results of the linear theory are touched only shortly, but the optimal nonlinear control design of descriptor systems will be discussed in more detail. 3.1. Linear systems Linear time-invariant descriptor systems are presented by Eẋ=Ax(t)+Bu(t) (3.1) y(t)=Cx(t)+Du(t) where x is an n-dimensional descriptor vector, u denotes the r-dimensional control input vector, and y characterises the m-dimensional measurement 598 P.C.Müller output vector. The matrices E, A are n× n-matrices, and B, C, D have dimensions n×r,m×n,m×r, respectively.Theessential propertyofdescriptor systems is that E is a singular matrix. rankE PE+E>PA=−Q (3.5) where definiteness properties of P and Q with respect to certain subspaces assure stability (Müller, 1993). Modelling and control of mechatronic systems... 599 First results on the design of the linear feedback control by pole placement were presented by Dai (1989). But the main problem of the synthesis of the feedback control consists in the possibility of non-proper system behaviour. This can be seen immediately by the consistent solution to the fast subsystem (3.4)2 (cf. Dai, 1989) x2(t)=−B2u(t)−NkB2u̇− . . .−N j−1 k B2u j−1(t) (3.6) which includes generally higher-order time-derivatives of the control input until to the order j − 1 with j ¬ k and N j k B2 = 0. The two cases have to be distinguished, where solution (3.6) depends either only on u(t) but not on its derivatives u̇(t), . . . ,uj−1(t) (i.e. j =1) or on u(t) and its derivatives u̇(t), . . . ,uj−1(t) (i.e. j > 1) according to the general case (3.6). In the first case, the system is called proper, in the second case non-proper according to related proper and non-proper transfer matrix functions. System (3.1)1 is proper if and only if in representations (3.4)1,2, the equation NkB2 =0 (3.7) holds. The distinction between proper and non-proper descriptor systems and its consequences were discussed by Müller (1998a, 1999a, 2000). Regarding proper and non-proper systems, the linear quadratic optimal regulator pro- blem (Müller, 1998a, 1999a, 2000) and the descriptor state estimation problem (Müller,1999b) were discussed in detail and properly solved. Some initial re- sults on the robust control design exist as well. The H∞-control problemwas considered in Masubuchi et al. (1997), Rehm (2003), Takaba et al. (1994). In case of stabilizable, detectable, impulse-controllable and impulse-observable descriptor systems necessary and sufficient conditions for the solvability of the control problem are given. For the analysis and systhesis of linear descriptor systems, usual theoretical tools are (more or less) available and related program packages have been developed (Bunse-Gerstner et al., 2000; Varga, 2000). 3.2. Nonlinear systems First results onnonlinear descriptor systemswere reported inBajic (1992), Campbell et al. (1999), Müller (1998b). Also, first attempts of the optimal control designwere given inCobb (1983), Jonckheere (1988). Particularly, the linear-quadratic optimal regulator for descriptor systems was dealt with in Bender and Laub (1987). More recently, theoretical papers on optimal control problems with sta- te constraints have been published, see Dmitruk (1993), Hartl et al. (1995), 600 P.C.Müller Stefani and Zezza (1996). However, usually the papers put severe restrictions on the type of the control problem (such as properness or index one) in or- der to secure some necessary optimality conditions. Such restrictions are often not satisfied by real practical problems, therefore the above-mentioned results cannot be applied in practice. The main weakness here is the lack of distinc- tion between proper and non-proper descriptor systems. If such distinction is regarded, the optimal control problems can be correctly dealt with. For the linear-quadratic optimal control design of linear descriptor systems this has been fully implemented in Müller (1999a, 2000). In a recent paper (Müller, 2003), the nonlinear optimal control problem has been discussed in detail. In the present contribution, we will present a survey on these results as mentioned above. Controlled time-invariant finite-dimensional descriptor systems can be de- scribed in a semi-explicit form by ẋ1 =f1(x1,x2,u) (3.8) 0=f2(x1,x2,u) where xi, i=1,2 are ni-dimensional vectors and n1+n2 =n. Usually u is an r-dimensional control input vector. Theproblemof the optimal control of nonlinear descriptor systems consists in the control design for (3.8) with respect to the performance criterion J = T ∫ 0 f0(x1,x2,u) dt ⇒ minimum (3.9) where u belongs to a set U of bounded or unbounded control functions: u∈U. Additionally, boundary conditions for the optimal control design have to be considered. The boundary conditions are given either geometrically or there are dynamical boundary conditions which have to be determined by an optimization procedure. In both cases they have to be consistentwith algebra- ic equations (3.8)2, which have to be satisfied. The corresponding boundary conditions will be not regarded explicitly in the following, because the han- dling follows from usual rules of the optimal control if the descriptor control problem has been properly formulated. Furthermore, we confine ourselves to principles of correct formulation of the optimal control design of nonlinear descriptor systems. The notation is simplified assuming that all algebraic equ- ations (3.8)2 have a uniform index, i.e. that they have the same index k. Then Modelling and control of mechatronic systems... 601 the underlying set of ordinary differential equations runs as ẋ2 =− ( ∂ ∂x>2 Lk−1 f1 (f2) )−1 Lk(f2)=f2(x1,x2,u, . . . ,u (s)) (3.10) on a manifold defined by algebraic equations (invariants) djf2 dtj ≡Lj(f2)=0 j =0, . . . ,k−1 (3.11) where L,Lf1 are suitably defined operators L(·)=Lf1(·)+Lu̇(·)+ ∆ ∆t (·) Lf1(·) = ∂(·) ∂x>1 f1 Lu(j)(·) = ∂(·) ∂u> u (j) (3.12) ∆ ∆t Lu(j)(·)=Lu(j+1)(·) Thefirst integrals Lj(f2)=0, j =0, . . . ,k−2, of (3.10) dependon x1 butnot on x2. The function L k−1(f2) depends on x2 for the first time such that the related Jacobian is regular anddifferential equation (3.10) canbederived.Also Lk−1(f2) = 0 is a first integral of (3.10). Additionally, the function L j(f2) dependsgenerally on the time-derivatives of the control inputu: u, u̇, . . . ,u(j), j = 0, . . . ,s, 0 ¬ s ¬ k. If u appears explicitly in Lp(f2) = 0 for the first time, then s= k−pholds. Systemrepresentations (3.8)1 and (3.10) have to be supplementedbyconsistent initial or boundaryconditions satisfying invariants (3.11) Lj(f2) ∣ ∣ ∣ t=0 =0 Lj(f2) ∣ ∣ ∣ t=T =0 j =0, . . . ,k−1 (3.13) The different representations of the control system in the preceding Section show that the system behaviour may depend not only on the control input u but also on its time-derivatives u̇,ü, . . . ,u(s−1),u(s). Although DAE de- scription (3.8) shows explicitly only the input u, there may be hidden effects related to time-derivatives u̇, . . . ,u(s) as it is shown by representations (3.8)1, (3.10), (3.11). This situation is very different from the common state space di- scussions. Any control design method has to take care of this unconventional problem. Therefore, it is necessary to clarify this unusual situation. For these systems, the notion of ”properness” is introduced according to the definition of frequency domain for linear systems as it was done in Section 3.1. Descrip- tor system (3.8)1 is called proper if the solution x1, x2 does not depend on 602 P.C.Müller u̇,ü, . . . ,u(s−1) but only on u (and/or on weighted integrals of u). Other- wise the system is called non-proper. The notion of proper behaviour can be sharpened for strictly proper systems, if system (3.8)1 is proper and additio- nally x2 depends on x1 but not on u, according to L k−1(f2)=0. Obviously, the description of the dynamical system by equations (3.8)1, (3.10) with the additional boundary conditions is the most convenient one to perform the optimization with respect to performance criterion (3.9) applying well-known methods of calculus of variations or the maximum principle of Pontryagin (Funk, 1970; Leitman, 1981; Pontryagin et al., 1962). The only one problem, which still appears, is correct handling of the time-derivatives u̇, . . . ,u(s) in the non-proper case. But this problem is easily solved by introducing some extended variables ξ1 =u, ξ2 = u̇, . . . , ξs =u (s−1), v=u(s) (3.14) defining a multi-dimensional integrator chain ξ̇1 = ξ2, ξ̇2 = ξ3, . . . , ξ̇s−1 = ξs, ξ̇s =v (3.15) By these variables, we are able to state the optimal control problem properly: Minimize the performance criterion, i.e. J = T ∫ 0 f0(x1,x2,ξ1) dt ⇒ minimum (3.16) regarding the differential constraints of an extended dynamical system resul- ting from ẋe =           f1(x1,x2,ξ1) f2(x1,x2,ξ1, . . . ,v) ξ2 ... ξs v           (3.17) where the extended state vector is defined as x>e = [ x>1 x > 2 ξ > 1 · · · ξ > s ] (3.18) The new (fictitious) control input is v. The original control constraints u∈U appearas state constraints ξ1 ∈U.Thegivengeometrical consistent boundary conditions have to be supplemented. Modelling and control of mechatronic systems... 603 The necessary conditions for the optimal control design follow from the application of the conditions of Pontryagin’s maximum principle, which was shown in Müller (2003). The necessary condition for the solution of optimal control problem (3.16) and (3.17) is the existence of a non-vanishing adjoint vector [ λ>1 λ > 2 Ψ > 1 · · · Ψ > s ] 6=0 (3.19) satisfying the relations λ̇1 =− ∂Hnp ∂x1 0=− ∂Hnp ∂x2 Ψ̇1 =− ∂Hnp ∂ξ1 (3.20) Ψ̇i =− ∂Hnp ∂ξi =−Ψi−1 i=2, . . . ,s and Hnpmax =Max v Hnp : Ψ > s v ⇒ Maximum v (3.21) where the Hamiltonian reads Hnp =λ > 1f1+λ > 2f2+Ψ > 1 ξ2+ · · ·+Ψ > s v−f0 (3.22) Therefore, for non-proper descriptor systems it is necessary to include higher- order time-derivatives into the optimization procedure according to integrator chain (3.14) and (3.15). Additionally, the control constraint u ∈ U has to be considered as a constraint ξ1 ∈ U of extended state (3.18). A practical problem ofmaximum condition (3.21) is that the constraint on v is not given in advance. Therefore, it is strongly recommended to think about a properly defined optimization problem. Condition (3.21) is an indirect hint that the problem is not well-posed from the very beginning. Inmany applications, one may be easily convinced that the constraint u∈U is not reasonable but the constraint v ∈ V does make sense. If the condition u ∈ U is replaced by v ∈ V , then a usual optimization problem arrives, which is solved by (3.20) and (3.21). After all, a two-point-boundary problem has to be considered to get the optimal (fictitious) control v. According to integrator chain (3.14) and (3.15) this is not static (proportional) control but dynamic control. A special case of the above result, see (3.20) and (3.21), is a solution to the unconstrained optimization problem under sufficient smoothness conditions such that calculus of variationsmay be applied. InMüller (2003) it was shown that conditions (3.20) and (3.21) can be essentially simplified to necessary 604 P.C.Müller conditions of the existence of non-vanishing adjoint variables λ1,λ2 satisfying the differential-algebraic equations λ̇1 =− ∂Hr ∂x1 0=− ∂Hr ∂x2 0=− ∂Hr ∂u (3.23) Here, the reduced Hamiltonian Hr =Hr(x1,x2,u,λ1,λ2)=λ > 1f1+λ > 2f2−f0 (3.24) is applied. The boundary conditions of the adjoint variables follow usual rules and are not considered explicitly. The solution to the unconstrained optimization problem consists in the solution to a two-point-boundary problem with respect to two sets (3.8) and (3.23)1,2 of differential-algebraic equations having regard to the control from (3.23)3. Unconstrained optimization problems can be solved by the usual ap- proach of the calculus of variations independently of the proper or non-proper behaviour of dynamical system (3.8). 4. Conclusions In this contribution, an effort to characterise the state of the art of model- ling, analysing and designing dynamical processes, particularly mechatronic systems by the descriptor system approach has been made. Without any do- ubts, the modelling of dynamical systems by differential-algebraic equations has many advantages and is superior to the state space modelling. The si- mulation tools for DAEs are well established and are comparable with ODE solvers. For linear descriptor systems, the tools for analysis and the control design are available, too. For nonlinear descriptor systems the optimal control design has been shown, but still more research work has to be performed. References 1. ArnoldM., 1998,ZurTheorie und zur numerischen Lösung vonAnfangswert- problemen für differentiell-algebraische Systeme von höhrem Index,VDI-Verlag, Fortschr.-Ber. VDI, Reihe 20, Nr. 264, Düsseldorf 2. Bajic V., 1992,Lyapunov’s DirectMethod in the Analysis of Singular Systems and Networks, Shades Technical Publications, Hillcrest, RSA Modelling and control of mechatronic systems... 605 3. Bender D.J, Laub A.J., 1987, The linear-quadratic optimal regulator for descriptor systems, IEEE Trans. Autom. Control, 32, 672-688 4. Brenan K.E., Campbell S.L., Petzold L.R., 1989, Numerical Solution of Initial-Value Problems in Differential-Algebraic Equations, North-Holland, NewYork 5. Bunse-Gerstner A.,Mehrmann V., Varga A., 2000,Numerische Metho- den zur robusten Steuerung. DFG-Forschungsprojekt, 1999/2000, Zentrum für Technomathematik, Tätigkeitsbericht, Universität Bremen 6. CampbellS.L.,NikoukhahR.,DelebecqueF., 1999,Nonlineardescriptor systems, In: Advances in Control-Highlights of ECC’99, P.M. Frank (Edit.), Springer, London, 247-281 7. CobbD., 1983,Descriptorvariable systemsandoptimal state regulation, IEEE Trans. Autom. Control, 28, 601-611 8. DaiL., 1989,SingularControl Systems,Volume118ofLectureNotes inControl and Information Sciences, Springer, Berlin-Heidelberg 9. Dmitruk A.V., 1993,Maximumprinciple for the general optimal control pro- blem with phase and regular mixed constraints, Comput. Math. Modelling, 4, 364-377 10. Führer C., 1988, Differential-algebraische Gleichungssysteme in mechani- schen Mehrkörpersystemen: Theorie, numerische Ansätze und Anwendun- gen, Dissertation, Mathematisches Institut und Institut für Informatik, TU München 11. Funk P., 1970,Variationsrechnung und ihre Anwendung in Physik und Tech- nik, (2nd ed.), Springer, Berlin-Heidelberg 12. Gear C.W., 1971, The simultaneous numerical solution of differential- algebraic equations, IEEE Trans. Circuit Theory, 18, 89-95 13. Hairer E., Wanner G., 1991, Solving Ordinary Differential Equations II, Stiff and Differential-Algebraic Problems, Springer, Berlin 14. Hartl R.F., Sethi S.P., Vickson R.G., 1995, A survey of the maximum principles for optimal control problems with state constraints, SIAM Review, 37, 181-218 15. Jonckheere E., 1988, Variational calculus for descriptor problems, IEEE Trans. Autom. Control, 33, 491-495 16. Kampowski W., Rentrop P., Schmidt W., 1992, Classification and nume- rical simulation of electric circuits, Surv. Math. Ind., 2, 23-65 17. LeitmannG., 1981,The Calculus of Variations and Optimal Control, Plenum Press, NewYork 606 P.C.Müller 18. Lewis F.L., 1986, A survey of linear singular systems, Circuits, Syst. Signal Processing, 5, 3-36 19. LuenbergerD.G., 1977,Dynamic equations in descriptor form, IEEETrans. Autom. Control, 22, 312-321 20. Masubuchi I., Kamitane Y., Ohara A., Suda N., 1997, H ∞ -Control for descriptor systems: a matrix inequaltiy approach,Automatica, 33, 669-373 21. Mathis W., 1992, Analysis of linear time-invariant network in the frequan- cy domain, In: Mathematical Modelling and Simulation of Electrical Circuits and Semiconductor Devices, R.E. Bank, R. Bulirsch, H. Gajewski, K.Mertens (Edit.), Birkhäuser, Basel, 83-90 22. Müller P.C., 1993, Stability of linear mechanical systems with holonomic constraints,Appl. Mech Rev., 46, 11, Part 2, 160-164 23. Müller P.C., 1995, Descriptor systems: a new way to model mechatronic systems?,Proc. 3rd European Control Conference, Rome, 3, Part 2, 2725-2729 24. MüllerP.C., 1998a,Analysis and control design of linear descriptor systems, In:Advances in Systems, Signals, Control and Computers, V. Bajic (Edit.), 1, 11-17, Center for Engineering Research, Technikon Natal, Durban, RSA 25. Müller P.C., 1998b, Stability and optimal control of nonlinear descriptor systems: a survey,Appl. Math. Comp. Sci., 8, 269-286 26. Müller P.C., 1999a, Linear control design of linear descriptor systems,Proc. 14th IFAC World Congress, Vol. C, 31-36, Pergamon, Beijing 27. Müller P.C., 1999b,Verallgemeinerte Luenberger-Beobachter für lineareDe- skriptorsysteme, Z. Angew. Math. Mech., 79, Suppl. 1, S9-S12 28. Müller P.C. 2000, Linear-quadratic optimal control of non-proper descriptor systems, In:CD-Proc. 14th Internat. Symp. Mathematical Theory of Networks and Systems (MTNS 2000), Université de Perpignan 29. Müller P.C., 2003, Optimal control of proper and nonproper descriptor sys- tems,Arch. Appl. Mech., 72, 875-884 30. PetzoldL.R., 1983,AdescriptionofDASSL:ADifferential/AlgebraicSystem Solver, In: Scientific Computing, R.S. Stepleman et al. (Edit.), 65-68, North- Holland, Amsterdam 31. Pontryagin L.S., Boltyanskii V.G., Gamkrelidze R.V., Mishchenko E.F., 1962, The Mathematical Theory of Optimal Processes, Interscience Pu- blishers, NewYork 32. Rehm A., 2003, Control of Linear Descriptor System: A Matrix Inequality Approach, Dissertation, Universität Stuttgart 33. Rosenberg R.M., 1977, Analytical Dynamics of Discrete Systems, Plenum Press, NewYork Modelling and control of mechatronic systems... 607 34. Rükgauer A., 1997, Modulare Simulation mechanischer Systeme mit An- wendung in der Fahrzeugdynamik, VDI, 248, VDI-Fortschr.-Ber., Reihe 20, Düsseldorf 35. Rükgauer A., Schiehlen W., 1997, Simulation of mdular dnamic sstems, Proc. 2nd MATHMOD, Vienna, 329-334 36. Schiehlen W., 1990,Multibody Systems Handbook, Springer, Berlin 37. Simeon B., 1994, Numerische Integration mechanischer Mehrkörpersysteme: Projizierende Deskriptorformen, Algorithmen und Rechenprogramme, VDI, 130, VDI-Fortschr.-Ber., Reihe 20, Düsseldorf 38. Stefani G., Zezza P., 1996, Optimality conditions for a constrained control problem, SIAM J. Control and Optimization, 34, 635-659 39. Takaba K., Morikira N., Katayama T., 1994,H ∞ -Control for descriptor systems – a J-spectral factorization approach,Proc. 33rd IEEEConf. Decision and Control, 2251-2256 40. Varga A., 2000,Descriptor System Toolbox for MATLAB, Institute of Robo- tics andMechatronics, German Aerospace Center, Oberpfaffenhofen Modelowanie i sterowanie układów mechatronicznych metodą deskrypcyjną Streszczenie W ostatnich latach sformułowano i spopularyzowano problem analizy i syntezy układów sterujących w postaci deskrypcyjnej. Ogólny opis układów dynamicznych za pomocą równań różniczkowo-algebraicznych (DAE)ma ogromne znaczenie aplika- cyjne w różnych dziedzinach nauki, w szczególności w zakresie mechatroniki.W pre- zentowanej pracy przedyskutowanowszystkie „za” i „przeciw”modelowania układów mechatronicznych równaniami różniczkowo-algebraicznymi z zastosowaniem podzia- łu opisywanego układu na podsystemy. Ponadto przedstawiono najnowocześniejsze metody symulacji, analizy i projektowania układów deskrypcyjnych. Manuscript received April 18, 2005; accepted for print May 16, 2005