42 IEEE Transactions on Energy Conversion, Vol. 13, No. 1, March 1998 cient and Accurate Model for the Simulation Synchronous MachindconverterSystems H. J. Hegner, Member Naval Surface Warfare Center Annapolis, Maryland S. D. Pekarek, Student Member 0.Wasynczuk, Senior Member School of Electrical and Computer Engineering Purdue University West Ldayette, Iadiana Abstract - A new synchronous machine model is presented which is readily implemented in either circuit-based or differentialequation-based simulation programs. This model is well suited for the simulation and analysis of synchronous machine - converter systems. It is based upon standard representations and no a p p r o h a tions are made in its derivation. However, the numerical implementation is shown to be significantly more eEcient An example is provided which demonstrates a 1700%increase in si" lation speed with no observable lws in accuracy. The model includes provisions for an arbitrary number of damper or rotor windings and may be easily modified to represent synchronous or induction machines with an arbitrary number of stator phases. I. INTRODUCTION Although there are a large number of techniques and computer programs available for the simulation of power circuits and systems, many fall into one of two general categories. In circuitbased simulation languages such as SABER [l] or SPICE [2], the circuit data (resistances, inductances, capacitances) are specified branch by branch. Topological information may be defined graphically or by a branch-to-node incidence matrix which is easily established from the equivalent circuit. Independent and dependent sources and time varying-parameters are easily incorporated. In order to establish the time-domain response, the differential equations are discretized at the branch level providing an algebraic equation relating branch voltages and currents at any given instant of time to their past values. These difference equations may be assembled numerically using nodal techniques to form a set of coupled difference equations of the interconnected system. In differential-equation-based languages such as MATLAB [3] or ACSL [4],the system is described by its differential equations or transfer functions. The differential equations are converted to difference equations in accordance with the specific integration algorithm selected (e.g. Euler, Runge Kutta, Gears, etc.). The disadvantage of this approach is that in many converter-machine systems, it is difficult to derive the differential equations for all the potential topological modes. However, an algorithmic method of developing the state equations of complex power circuits and systems was recently set forth in [5]. PE-689-EC-0-04-1997 A paper recommended and approved by the IEEE Electric Machinery Committee of the IEEE Power Engineering Society for publication in the IEEE Transactions on Energy Conversion. Manuscript submitted November 13, 1996; made available for printing March 26, 1997 Using this method, circuits may be described by the pertinent branch parameters and a branch-to-node incidence matrix as in circuit-based languages. The composite system state equations are then established algorithmically, and are subsequently solved using any one of a number of well established techniques. Regardless of the simulation approach selected, the model structure and choice of state variables can have a significant impact upon simulation speed and accuracy. In thi synchronous machine model is presented which is shown to have significant advantages relative to existing models. It is readily implemented in either circuit-based languages or in differential-equation-based languages using the state model generation algorithm in [5]. This model is based upon s representations and no approximations are made in its derivation. In fact, the solution of the corresponding state equations yields identical results for the same inputs and initial conditions. However, the numerical implementation is significantly more efficient. In the example system sfudied, the given model gives rise to a 1700%increase in simulation speed with no discernible loss in accuracy. II. AUTOMATED STATE MODEL GENERATION Before introducing the new synchronous machine model, it is useful to describe briefly the state model generation algorithm set forth in [5]. In switched inductive circuits, the state equations may be expressed explicitly as pi, = [--Li1(r,+pLx)]ix-LilBbebr (1) T T where L, = BbLbrBb, rx = BbrbrBb, and B, is the socalled basic loop matrix which is numerically established from the branch-to-node incidence matrix. Also, rbr and L,, are the branch resistance and inductance matrix, respectively, which are readily established from branch data. The currents and voltages of all branches in the system may be obtained using T. ibr= Bbi, and Capacitive elements may be incorporated into the state model as described in [5]. An alternative formulation may be derived in which flux linkage, rather than current is used as the state variable. Therein, the current vector is written in terms of flux linkage as i, = L,-1 h, 0885-8969/98/$10.00 0 1997 IEEE (4) 43 and the resulting state equation written d-axis rotor windings -1 p h , = - r,L, h, - B b e b r .\\ I Lbs ' b r = [ ( r b r + P L b r ) B E - L b r B bTL ,-1 (r, + p L , , ] L i l h + [I - L b , B bT L,-1 B J e b r ) stator windings 'kdN (5) The current and voltage of all branches in the system may be obtained using T -1 i b , = B b L , h, (6) and q-axis rotor windings (7) Equation (5), (6), and (7) define the state model of the system with flux linkages as the state variables and branch voltages and currents as outputs. m. CIRCUIT-BASED SYNCHRONOUS MACHINE MODEL For notational purposes, it is assumed that the synchronous machine has three stator windings, one rotating field winding, and an arbitrary number of short-circuited damper windings along the q and d axes. Machines with other than three stator windings require only a minor change of notation. The equivalent circuit is depicted in Fig. 1. Therein, the orientation of each winding (depicted in Fig. 1 as inductors) portrays the physical dmction of the Corresponding magnetic axis. The voltage equations may be expressed Fig. 1 Coupled equivalent circuit of synchronous machm. state model with voltages as inputs and winding currents as state variables. This representation of the machine is subsequently referred to as the coupled circuit (CC) model. The corresponding equations may be solved numerically using either circuit-based or differential-equation-based languages. In circuit-based languages, the nodes and branches of Fig. 1 are first labelled and/or numbered, the branch parameters (resistances, self- and mutualinductances) are specified and topological information is defined graphically or by a branch-to-node incidence list. In differentialequation based approaches, equations (8) and (11) are programmed using syntax specific to the selected language. In either case, a variety of numerical algorithms (e.g. Runge-Kutta, Gears, Euler) may be applied to establish the time-domain response. IV. VOLTAGE BEBIND REACTANCE REPRESENTATION where Designating 8, as the electrical position of the rotor, the stator variables may be transformed to the rotor reference frame using Park's transformation [SI. <dos = K:(er)fabcs Here, f can be i , or h . The stator resistance matrix is rs13 . The rotor resistance matrix rr is diagonal with entries corresponding to the appropriate field or damper winding resisU, (14) tance. The flux linkage equations may be expressed where f may be a voltage, current, or flux linkage, and - _I Icos(Qri cos(€+-$) Where Lss is the stator inductance matrix, L,, and L,, repre- cos(Or+q sent the mutual inductances between the stator and rotor windings, and L,, represents the rotor inductance matrix. The stator -1 inductance matrix is of the form 2 (12) i The equations of the synchronous machine may be expressed in the rotor reference frame as [6] .r Phis (17) %gs P h i , (18) vis! = rs'slqs+ % G s where, for example La,,, = LlS+ LA - LgCOS28, (13) Expressions for the remaining stator and rotor self- and mutualinductances are given in [6] and will not be repeated here due to space considerations. Equations (8) and (11) are readily manipulated to form a l2 .r = r.2d.S VOS UJ. = QOS +Phos = rJ. iJ. + p h j + 4- (19) (20) where j = kql, ...,k q M , fd, kdl, ..., k d N . Here, the super- ables are expressed in the this paper, rotor variables to the stator by the appropriate turns he stator and rotor flux linkages per sec- Substituting (31) and (32 equations may be re r the .r r ~ s qs .r ' d s = 's'ds- U qs r + = In the analysis of machin an approximate voltage-b (37) and (38) wherein it is constant during fast switch A.J = L Ui j + A m d ; j = fd,kdl, ...kdN (24) where p h i terms may be neglected. H algebraically incorporated which case an exact voltag be obtained. This is achieved oltage equations in a voltagebehind-reactance form, the magnetizing flux linkages are f i t expressed in terms of rotor flux linkages. Solving (21)-(24) for currents and substituting into (25) and (26) yields, after algebraic manipulation, Expressions for the derivat obtained by manipulating the (24). Using 1 i . = -(A.J Lij J M 1-1 ,,.r .r r uLs = rqLqs+ q-LJdtds + p L q i q s + e; 1 1 N -1 w .r (30) j= (42) for the rotor currents and substituting the resulting expressions into (37) and (38) yields, after algebraic manipulation, an exact voltage-behind-reactanceform of the stator voltage equations where r hmd);j = fd, kdl, ...kdN 1 fi .r uLs = r i i & - o , . L ~+LP L~d~i d s + e ; (43) (44) where The stator flux linkage equations can then be expressed as xis = L i i i , + h ; (32) r $ = r s +Lm%fd - where the double primes ("1 are used to denote subtransient (46) L;fd quantities. The subtransient inductances, L i and Ld are given by and ,, L; = L l s + L m 4 (33) (34) and the subtransientflux linkages by (35) These equations applying the inverse (44), which yields, Y d 45 where Am, and hmd are given by (27)-(28),define the so-called voltage-behind-reactance (VBR) model of the synchronous machine. It is important to note that no approximations were made in its derivation. Neglecting numerical error, the solution of the corresponding equations should yield the same time-domain response as the CC model upon which it is based. Moreover, no assumptions have been made in regard to the stator winding configuration. The windings may be connected in wye, delta, or the individual windings may be supplied to isolated converter circuits. With only minor modification, the model may be extended to include machines with an arbitrary number of stator phases. When implementing the VBR model, only the stator branches and nodes are included when defining the circuit topology. The rotor voltage equations are expressed explicitly in state model form with rotor flux linkages as state variables. The subtransient voltages represent outputs of the rotor model and are incorporated in the stator circuit as dependent sources. The stator branch currents are transformed into the rotor reference frame and represent inputs to the rotor state model. A circuit/blockdiagram of the VBR model is given in Fig. 2 In circuit-based approaches, it is necessary to solve an ndimensional set of linear equations at each time step where n corresponds to the number of nodes. In differential-equation-based approaches using the state model generation algorithm, it is necessary to solve a k-dimensional set of linear equations where k = b-n+l.Here,bisthenumberofbranchesannisthe number of nodes. The elimination of rotor branches and nodes reduces the dimensionality of the linear equations that need to be solved at each time step in either approach. The resistance and inductance matrices are given by -, r 1 n n La = Lmq + xkz&-I Lmd ;'I-' I l l 3 . It is interesting to note that the stator voltage equations are complicated by the addition of the time varying "mutual" resistances contained in (51). However, these are easily incorporated by introducing off-diagonal terms in the branch resistance matrix rbr. The stator voltage equations given by (49), along with the rotor voltage equations r. p h . = -L(hj-hmd)+Vj; j = fd, k d l , ...k d N Llj _ . , U Fig. 2 Voltage-behind-reactancemachine model. V. EIGENSYSTEM ANALYSIS Although the CC and VBR models are equivalent in the sense that the solutions are identical for the same inputs and initial conditions, the eigenstmcturesmay be substantially different. For example, if the CC model is implemented with currents as state variables, the state model (1) can be expressed symbolically as p i , = A{@,) i, + Bier) ebr (63) where (62) A(er) = -Lil(O,)[r, +pL,(e,)l (64) 46 B@,) = -Lil(Q,) (65) Also, ebr is a vector of dependent and independent branch voltages which are assumed to be inputs in this analysis. If the CC model is expressed with flux linkages as the state variables, the state equations may be expressed p h , = 40.1 h, where + Bhebr (66) Ah(@,)= -IzL;?e,) (67) (68) Although (63) and (66) are algebraically related, the eigenstructures are substantially different. To show this, it is convenient to assume that the rotor of the synchronous machine is held fixed at an arbitrary position ern. Here, the subscript denotes where pL(8,) = 2wFBsin(28,) If the rotor is f i e d at an arbitrary position, the inductance is constant and the resulting eigenvalue [-r/L(B,@)I is negative. However, if the rotor speed is such that 2 0 r L B > r , then for some rotor positions the resulting eigenvalue will be positive. The presence of positive real eigenvalues can lead to a significant error growth when integrating the state equations numerically. Generally, a smaller time step is needed in order to maintain a specified truncation error. A detailed explanation of the relationship between eigenstructure and time step is contained in [9]. This analysis suggests that flux linkages represent a better choice of state variables. In the VBR model with (5) used to represent the stator dynamics, the rotor and stator flux linkages are state variables. that w, = 0 . With the rotor fixed, the corresponding state matrices A$o,fo) and A,(0,@) are timeinvariant. Moreover, they are similar since q1 Ai @,fd= (e,fo)A,( 6,fO )L,( % f O ) (69) Therefore, at zero speed the two models have identical eigenvalues. Since the machine is stationary, the eigenvalues are negative regardless of the selected rotor position. VI. COMPUTER STUDY In order to illustrate the advantages of the new model, an experimental system consisting of a 3.7 kW synchronous machine connected to a line-commutator converter was simulated. The system studied is the same as that described in [5]. A circuit diagram of this system is shown in Fig. 3. The synchronous machine parameters, as determined by standstill frequency response testing, are summarized in Table 1. If the machine is allowed to rotate, the matrices Aje,) and Ah(e), at any rotor position e,, may be expressed as -4'(Or0)rX-Li1 (6,0)~L,1( )e, ~$0~= 0) (77) Table 1: Synchronous machine parameters. Ir, = 382 mQ IL,, = 1.12 mH IL,, = 24.9 mH (70) (71) Comparing (71) with (67) it can be seen that if ,e, A~(O,,) = A,@,f,) = then (rkd3= 1.58 Q (72) Thus, the real parts of the eigenvalues of A,(e,) are negative regardless of speed or rotor position. The same cannot be said for 4 (0,). Comparing (70) with (64)it can be seen that if ,e, = = 4.52 m~ Irkq3= 447 mQ I L,, = 39.3 mH (Llkol= 4.21 mH (Llka2= 3.5 mH then ~p,,) = ~ p -L;l(e,,)pL;l(e,,) ,~) (73) Although the first term on the right of (73) was shown to have eigenvalues that are negative, the addition of the second term, which is a function of the changing inductances, can shift the eigenvalues of 4 (0,) to the right-half plane. This can be shown In the computer studies, it is assumed that the system is initially operating in the steady state with a base load resistance of 21 Q connected to the dc output terminals. A second load resistance of 4.04 Q is then connected in parallel with the original load. The CC and VBR models were implemented in ACSL [4] using a simple example. using the state model generation algorithm. The simulated In a single-phase reluctance machine the voltage equations may be expressed as response using the CC model is depicted in Figs. 4-5. The experimentally measured response is shown in Fig. 6 [5]. As shown, the measured and simulated responses are in excellent agreement. In calculating the response depicted in Figs. 4-5, Gear's algorithm U = ri+p[L(O,)i] where the winding self-inductance is given by ~ ( 0 ,= ) L,,+ L~ - L ~ C O S ( ~ ~ , ) (74) (75) with LA > LB ,the resulting state equation can be written in terms was used with a maximum and minimum time step of 1 x and 1 x sec, respectively. The local truncation error, which -4 of currents as 1 LWU (76) is used to determine the actual time step, was set to 1 x 10 for all state variables. When a change in topology (change in diode conduction state) is sensed, the time step is reduced and the pre- I ' b VP l h t) f)n 12 b 19 n16 b23 n20 nl - - vbn b 10 i b8 vca 47 i. U n3 --, == vdc n4 Fig. 3 Example system. *O 1 2o 1 0 ' 0.75 1 0.751 0.0 I 40ms I ' Fig. 4 Simulated dc and field currents. 2o 1 -20 J 50 7 -50' Fig. 5 Simulated ac current and vdtage. ceding calculation is repeated so as to limit the uncertainty in switching time to less than the minimum time step. Following each topological change, the time step is set to its minimum value and the integration algorithm is reset. Subsequent time steps are adjusted in accordance with Gear's criteria [ 101. Fig. 6 Measured dc and field currents. For comparison purposes, the same study was repeated using the VBR model. The results are indiscernible from those depicted in Figs. 4-5. However, the computer time is significantly less. The CC model required 212 sec and the VBR model required 12.3 sec of CPU time with both models implemented on a 133MHz Pentium-based personal computer. The average time step selected by Gear's algorithm was 300% larger in the VBR model with no observable difference in response. The further savings is attributed to the reduction in the number of nodes and branches in the VBR representation. The CC model includes 24 branches and 20 nodes whereas the VBR model includes 10 branches and 6 nodes. Although the eigenvalues vary with time, it is interesting to compare them at a specific instant of time. The eigenvalues associated with the two models are summarized in Table 2 for t = 0.02. While the CC model has two eigenvalues with positive real parts, all of the eigenvalues of the VBR model have negative real parts, In fact, if the eigenvalues are calculated at other time instants, the CC model consistently has eigenvalues with positive real parts, while the eigenvalues of the VBR model stay in the left-half plane. Consequently, a larger integration time step may be used in the VBR model. This, coupred with the reduced dimensionality of the linear equations that must be solved at each time step, results in a much faster simulation. 48 Bble 2: System Eigenvalues using CC and VBR Models t I 326 -46 -279 -802 - 1,377 -4,718 -13,442 -218,060 I I I -52 -205 -370 -798 -1,176 -13,315 -214,489 [47 Mitchell and Gauthier Associates, Advanced Continuous Simulation Language Reference Manual,” Concord, MA, 1993. IS] 0. Wasynczuk and S. D.Sudhoff, “Automated State Model Generation Algorithm for Power Circuits and Systems,” Paper 96 WM 259-2 PWRC presented at the IEEE Power Engineering Society Winter Meeting, Baltimore, MD, Jan. 21-25, 1996. Electric Machinery, nous Machine Systems ” version, Vol. 8, No. 1, scataway, N5, 1995. , ‘Analysis and Averageutated Converter-Synchroions on Energy Con- Vn. SUMMARY A new voltage-behind-reactance model of the synchronous machine is derived. Therein, the rotor dynamic equations are expressed explicitly in state model form while the stator equations are described in circuit form. The model is computationally more efficient than existing machine models without making any approximations, it is flexible with regard to the stator winding configuration, and it is compatible with circuit-based and differential-equation-based simulation approaches. age-Value Model of Line-CommutatedConverter - Synchronous Machine Systems,” IEEE Transactions on Energy Conversion, Vol. 8, No. 3, pp. 404-410, September 1993. [9] L. 0. Chua, P-M Lin, Computer Aided Analysis of Electronic Circuits: Algorithms and Computational Techniques, Prentice-Hall, Englewood Cliffs, New Jersey, 1975. [lo] C. W. Gear, Numerical Initial Value Problems in Ordinary Differential Equations, Prentice Hall, NJ, 1971. VIII. ACKKNOWLEDGEMENTS This work was supported in part by P. C . Krause and Associates under contract “24-93-C-4180 with Naval Surface Warfare Center and in part by University of South Carolina under Grant No. N00014-96-1-0926 with Office of Naval Research. Steven D. Pekarek was bom in Cicero, Illinois on December 22, 1968. He received the B.S.E.E, M.S.EE, degrees from Purdue University in 1991, 1993 respectively, and is currently pursuing his Ph.D. His interests include electric machines and automatic control. E.REFERENCES Analogy, Inc., “Introduction to the Saber Simulator,” Beaverton, Oregon, 1991. L. W. Nagel and D. 0. Pederson, “Simulation Program with Integrated Circuit Emphasis,” University of California Electronic Research Laboratory, Memorandum EAL-M382, April 1973. The Mathworks, Inc. “Simulink Dynamic System Simulation Software - Users Manual,” Natick, Massachusetts, 1993. 01% W a s p a u k ( M 76, S M 88) was bom in Chicago, Illinois on June 26,1954. He received the B.S.E.E. degree from Bradley University in 1976 an the M.S. and Ph.D. degrees from Purdue University in 1977 and 1979, respectively. Since 1979, he has been at Purdue where he is presently a Professor of Electrical and Computer Engineering. Dr. Wasynczuk is a member of Eta Kappa Nu, Tau Beta Pi and Phi Kappa Phi and is a Senior member of the IEEE Power Engineenng Society. H. J. Hegner (S’ 88, M’89) received the B.S.E.E. degree from the Virginia Polytechnic Institute and State University in 1983 and the M.S.E.E. degree from Purdue University in 1992. He is currently a member of the US. Navy Advanced Surface Machinery Programs in which he serves as a team leader of the DC Zonal Electric System Program. For the past 16 years, he has specialized in electrical systems and components for US.Navy shipboard systems