V. Feliu Protesor, Department of ingeniriaEletea, fecronieaY Ho Conta SITS. lndusvites Oe La Ure, Matne-78080, Spain K. S. Rattan Professor, Department ol Eletizal Engoe ‘Wiihl State Universi. Dayion, GH 45435 H. B. Brown, Jr. Robotics Istitee, Cammepie Maton University Pisburgh, Ponasyiana 15273 Modeling and Control of Single- Link Flexible Arms With Lumped Masses This paper deals with the modeling and control of a special class of single-tint Flexible arms. These arms consist of flexible massless siructures having some masses concentrated at certain points of the beam. In this paper, the dynainie model of such flevible arms is developed and some of the conirol properties are dedced. A robust control scheme fo remove the effects of friction in the joins fs proposed. The control scheme consists of two nested feedback loops, an inner loop 10 control the ‘position of the motor and an outer lodp fo contral the tip position. The inner loop ‘is deseribed in other publications. A simple feedforward.feedback controller is de signed for the outer loop to drive the bean accurately along a desired trajectory Effects of the changes in the tip’s mass are studied. This modeling and control ‘method is then generalized to the distributed-mass flexible beam case. Finally, exe perimental results are presented, 1 Introduet This paper deals with the modeling and control of a special class of single-link, lumped-mass, flexible arms. These arms ‘consist of massless flexible structures that have masses cot centrated at certain points of the beam (see Fig. 1). Although the translations of these masses produce stresses in the flexible structure, their rotations do not generate any torgute in the beam. Therefore, the number of vibrational modes in the struc ture coincides with the number of lumped masses. Book (1979) studied the case of two rigid masses connected by a chain of massless beams having an arbitrary number of rotation joint Our problem differs from this in the sense that our structure hhas only one rotation joint and an arbiceary number of lumped masses, These twa particular structures are studied because + Some lightweight robots and other applications can be reasonably approximated by these models. ‘+ Their dynamics may be easily modeled as compared to dlistributed-mass flexible arms Interesting properties for the control of flexible arms are educed from their dynamic models ‘A method to control these arms is inferred from the steue ture of the model ‘The influence of changes in the tip's mass are easily char: acterized, Given a distributed-mass flexible arm, there always exits 4 truncated dynamie model which is of the same form as the lumped-mass flexible arm model und which reproduces the dynamics of the measured variables. This allows us to generate the above mentioned control method to the case of disisibuted-mast flexible arms. march 32,199; reed Imanaserp eecivedAtsch Ta, B91. Amore Feta, ayes Journal of Dynamic Systems, Measurement, and Control A method 0 control these flexible arms ie proposed in this paper and is based on a robust control scheme developed by Ratian eral. (1988, 1988) This method compensates for the effects of Coulomb and dynamic (viscous damping) frletions in the joints. Coulomb friction isa nonlinearity which is *yp- ically not taken into account by other methods to control flexible arms, This method proposes the use of two nested feedback loops, an inner loop to control the motor position land ax outer loop to control the tip position. ‘A new method to control the tip position of Nexible arms js also proposed in this paper. This method is simpler than the other existing methods but, in turn, needs more sensing. Most ‘methods (Cannon and Schmitz, 1985; Matsuno et al., 1987 Kotnick et al., 1988; etc.) use measurements of the tip a ‘motor variables. This proposed method also uses some sen at intermediate points of the beam. The number of sensi elements placed on the beam are equal to the number of con ‘centrated masses of the beam, The sensors are located on the masses. Two approaches may be used—sensing positions or umped massos exible beam MARCH 1992, Vol. 114/59sensing torques. In the second approach, a sensor is instalted at the base of the beam instead of the tip mass. The control scheme developed in this paper is based on position sensing. [Effects of the changes in the carried load on the dyn: of the system are studies. In order co keep the systems response to a reference input approximately constant, the feedforward component of the controller is tuned for a particular payload, Section Zestablishes the dynamic model of the Flexiale arms. Some properties that are useful when controlling these arms are deduced in Section 3. The control scheme is proposed in Section 4, Modeling and control methods are extended to the case of distributed mass flexible beams in Section 8. Expe mental results are presented in Section 6 and conclusions are drawn in Section 7. 2 Modeling The model of our flexible arms is divided into two sub- models; one describes the behavior of the motor while the other describes the behavior of the mechanical structure using the angle of the motor as its input. These two submodels are coupled by the reaction torque of the bean on the motor (see Fig. 2). This model is quite different from the models normally used in the control of flexible arms that consider the applied orque as the input to the beam (Truckenbrodt, 1979 and Low. 1987]. Our representation has an advantage in that it identifies Hlexible arms with friction in the joints (Feliu et al., 1988), ‘which allows us to compensate For this friction (Rattan et al, 1988). We show that another advantage of our model is that iv allows us to separate the dynaniic model terms that depend ‘on the geometry of the beam from the terms that depend on the lumped masses of the beam. Special attention is paid 10 this issue because this separation of terms will be used in the control design. 2.1 Beam Modeling. Consider the system of Fig. 1. It represents a massless flexible beam with m point masses dis tributed along the structure with the last mass located at the tip of the beam. The inertia of the motor is included in the motor submodel. Let mj, 1=5/51, be the values of the masses; Je the distances between consecutive masses /— 1 and where 1 Js the distance between the rotation axis of the motor and the first mass. let Z, be the distance between the mass m, and the axis of the motor. We assume that beam deflections are sinall enough 50 that the distances between masses a, (me: ured along length of beam) are essentially equal to thelr pro- jections on the Y-axis, We establish two coordinate systems, both with origins at the motor axis. The coordinate system 2— Ys fixed in space, whereas the coordinate system XY moves with the motor shalt. Thus, » is the distance of a point on the bear froma the X-axis, We denote #(x) and T(x) as the force and torque at this point; ic, the force and torque acting on the beam just to the left of this point due to the action of the beam just to he right of the point. External forces and torques, F, and Ty, applied at the tip Of the beam are considered in our model. F,, represents the ‘component of the resultant applied force that is normal to the ‘beam and t0 the joint rotation velocity vector simultaneously Tris the component of the resuitant torque normal to the X- ¥ plane. These two terms represents the interaction of the ‘beam with the environment and may be produced by the fol: lowing: reaction forces when following a surface, the reaction (of the next joint when dealing with a multi-link arm, a load, frictional effects on the tip (for example in flexible arms ‘mounted on an air table), etc ‘Assuming sinall deflections, & massless beam can be de- sctibed by the state defleetion relation 601 Vol. 114, MARCH 1992 lg, 2_Oynamic moo af the arn If the stiffness FY is constant through each interval of the ‘beam, the deflection inthe interval [/— 1, i]s given by a thirel order polynomial DHA) =the Fw Ly 1) Pale LW ma Lia? o) where tj are the polynomial coefficients and are different for each interval, and La = 0. Weassumi in the following analysis that £Tisconstant throughout the beam. Notice that expression (1) is similar to the equations obtained from finite element model 24.1. Geomette Equations, Imposing continuity cond tions betwen two consecutive intervals up 0 the second de: rivative, we get 3(0~1) equations PAL) = Sex D> thet tealet Maal + ial) = Miso Gay is ee By) =P (earns Quah Stoll =taews @) ay, ae for the joint between {i~1, f) and fi, (+1) intervats, where Lsicn. The continuity condition between the mator and the ‘beam yields | FE (Ld 22+ Bish Wiss de a r0)= 2 == 160, m= °% and the condition of the applied torque atthe tip gives fy Prats Lous tien ® ae 5 This set of geometrical equations is completed by expressing deflections 3 specific points / in terms af these polynomials tug t tial tial tual =yLa Lise © 2.12 Dynamic Equatio tions to the m masses, we get If we apply Newion's equa ei ie Tebad= By my ty PPD ‘ —Th-Fr(Le-L) Osicn (6) Osicn where (Lj) = y(L,)4-0yL) i the ate position of mass j with respect to the origin ans (isthe angle of the motor). Taking ince account that n w= ED, y= Flee) o ati GET and combining (6)-(8), we get PHL) _6ET FE lagenols isisn 8) where ugg = ~ B/E is defined according 0 (8). Bxpes sion (9) i obtained from (7) using the second equation of 8) ‘An equivalent expression may be obtained from (6) since expressions (2)-(8) are not independent. But we used (7) be- cause the resulting expression is simpler Transactions of the ASME2.1.3 Dynamic Model of the Beam. Equations (2)-(5) a low us to express coefficients wy as linear functions of (Ey) and T,. The u:3 coefficients can be expressed in compact forit ny Pen) igs} __ [2a 29) euf 1 (0) (Le) where wis a constant matrix that depends only on the dimen- sions of the beamn and the location ofthe mastes, Denoting 0 as the ange between the X-axis and the radial line from the origin to mass i (See Fig. 1), using the approximation 8 S(L,) Ls, and substituting (26) into (9), we get linear eau tions of the form mah Bar A fj+ DOA BET Lsicn a a a et Year be Be Tot ‘Equations (11) and (12) can be expressed in compact form as Mt EO BIAG+ Boa) +P, OF, (3) where ME diagtms ma, «me ©” = @y Bs 89) Ais an nx constant matrix, and B, P and Q are constant ‘nx1 column vectors. In particular Q” = (0,0, «40, Ly"). Inexpression (13). A, B, P and Q depend only on the geometry Of the beam. The values of the lumped masses influence only matrix M, 2.2. Motor Modeling, tor may be modeled as Kins Bee ve vce The dynamic behavior of a de mo- ay, where X is the electromechianical constant of the motor, / is, the current, Jis the polar inertia of the motor, Vis the dynami ttiction coefficient, Cy is the coupling torque between motor and beam and CFs the Coulomb friction. Taking into account that C, = ~7(0) = —28/u,., we ean express this torque as linear function’ C= HOA hing Buy ne To <3) where H=(hiy fy. «5 Madi hig Lsisim-42 are parameters ‘hat do not depend on the masses of the beam. Expressions (13)-(15) suggest an easy way of extending this ‘method of modeling to multi ink arms, Expression (13) is obtained for each link, and expressions (14) and (15) for each joint. The coupling between each pair of links is obtained by (waking the 7, of the ith link equal to the C, of the (1+ Ist link 3. Properties of the Lumped-Mass Model ‘The dynamics of the exible arm as described by Eag.(13)- (18) are represented in Fig. 2. We assume that Fa Ty are Berturbations in the system, In order to perform the andiyss 314 design of the contol system, we make F, = Ty = 0. Five interesting properties from the control point of view ate siven below, om - ‘ t gqlfom (13) and using Laplace transforms, we find that Ue aster functions Gifs) = 04(s)Am(s), for I sin, have Heats only of the form 5%; O0 ~ two conjugate roots on th Imaginary axis ‘ournal of Dynamic Systems, Measurement, and Control (b) 268, 2<0 » two roots same magnitude but opposite sige (©) 2€C, F(z) # 0 = sets of four symmetrical roots with respect to both real and imaginary axes. n the real axis, being of the ‘The mechanical structure is marginally stable because we as- sume that there is no friction along the beam (no energy dis sipating phenomena) and the friction on the tip is treated as ‘perturbation. The poles are always of class (a), but the zeros, may be of any of the three classes. Case (B) appears quite often and produces nonminimum phase systems, Control of these systems presents some difficulties as stated in Cannon, and Schmitz (1985), 2. The difference between the orders of the denominator ‘and numerator of G(s), Vi, is at least two. This is a conse- quence of dealing with Mexible arms. If numerator and de nominator were of the same order, it would mean that an instantaneous change in the position of the motor would pro duce an instantaneous change in the position of the th point of the flexible arm (Initial Value Theorem of Laplace trans: forms, see Kuo (1982), which is not possible because flexible beams need some time to propagate the motion along the structure. 3. Separation Property. {was mentioned in the previous Section that the influence ofthe lumped masseson the dyitamies ‘of the arm may be perfectly seprated from the influence of the geometrical dimensions of the beam (see tgs. (13) and (15)). This occurs because the beam is modeled as massless; thus, ts shape fs given by static deflection equations that de- pend only on the position coordinates. This property is used in the controller design, 4. Zeros invariance Property. Zeros of G(s), the transfer funetion between the tip position 8, and the angle of the motor 8, femain constant and are independent of the payload of the tip. 5. The coupling torque C, does not depend explicitly on the external force, Fy. applied at the tip (see (I5)). This is because the force applied to the tip does not appear in the scometric equations. These properties are also verified for distributed-mass flex ible arms and will be used in the next section, 4 Control Method To control our exible arms, a scheme based on to nested feedback loops is proposed. An inner loop is used to contol ‘the position of the motor and an outer loop is wsed to contral the tip position (see Fig. 3). The controller of this outer loop generates a control signal which i the reference for the position Of the motor (,,) ia the inner loop. This scheme was shown to be robust in the sense that it minimizes the effeets of the Coulomb friction in the control ag well as the effects of un ‘expected changes in the dynamic friction (Rattan etal. 19883), Details of the design of the inner loop may be found in Rattan et a. (1988). Because the motor is a minimum phase system, the inner loop uses very high feedback gains relative ‘o the gain ofthe outer ioop. The inner loop also incorporates ‘compensation terms for the Coulomb friction and the conpling torque between the motor and the beam (ee Fig. 4). Th: compensation tetm for the motor-beam coupling may be easily Implemented in our arms using (15). Because of these high ain, the positioning of the motor can be made substantially faster than the dynamics of the vibration modes of the beam. {the positioning of the motor is much faster than the dynamics ‘of the beam, the dynamics of the inner loop can be neglected Gn = Gn) and the design of the tip position control loop is significantly simplified. We assume, in what follows, that € = Gor, ad also that Ty = Fy = 0. ‘Thecontroller for the tip position is composed of a combined feectforward/feedback law (Fig. 5). The feedforward com- MARCH 1992, Vol, 114/61Computer conto! loop of the motor position Fig. Proposed tp postion controler ff Puyo oy A——_ r0. euapimote ny Fig, 8 Nomina trajectories forthe tp position (F. ponent is cesponsible for driving the tip of the arm close to the desired trajectory and the feedback term is responsible for correcting tracking. errors, Both components are designed com sidering thatthe input of the system is the desired angle of the MROLOE Bye 4.1 Feedforward Term, if the transfer function between the angle of the motor and the angie of the tip Ga(s) is min- imum phase, then a second order parabolic profile ean be used, ‘and the feedforward term can be Gy '(s). But if this transfer Function is non-minimum phase, then a quesi-parabolic profile with derivatives bounded up Co the fourth order (sce Fig. 6) is used in order to guarantee the implementability ofthe feed- foward term (Fel et al., 1988) and the nominal quasi-para- bolie profile s passed through a special filter in order to avoid unbounded contral signals, We denote that nominal trajectory a8 Pp ‘The necessity of the above tioned filter may be justified 62 Vol. 114, MARCH 1992 a a ee Fi.7 Feedlorwerd contat from Fig. 7. Assume.an open-loop control forthe case without ‘external perturbations, If we want the tip to follow the ref rence exactly, then the contol signal Bq, (which is same as 8, neplecting the dynamics of the inner loop) is obtained by passing the desieed profile P, through a block Ga() that im plements the inverse ofthe plant Gy(s)- If this plant had zeros inthe right half-plane, they would become unstable poles it G,"(s) producing an unbounded 9, control signal. 1n order to'avoid this, a modified G,(s) term must be used and the tip reference would now be aiven by Onl8)=GyVGuls)Pol8) 6 where Pp(s) i the Laplace transform of the parabolic or quasi parabolic profile. Ths iter is chosen in sich a way to get 2 reference trajectory By 85 close as possible to the desired ref rence Pp, taking into account the consiraine of a bounded Bi. We Ehoose a8 a representative index of closeness between trajectories, the integral of the squared difference between botk profiles (ee Fig. 7). This can be weitten as 2 2 XU Gala) Gals) 5 (1 Gy —29Gal = De, (OD herea parabolic profile 2/s*has been assumed for Pybecause, from Fig. 6, the qasi-parabalie profile behaves as a'parabolic trajectory mott ofthe time, becoming a fourth order parabola only during the short transitions feom maxim ro minimin acceleration and vice versa, The transfer funetion Ga(s) that minimizes (17) isa Weiner filter and gives s better performance than any simple feedback controler, provided theve are no perturbations inthe system. A feedback version of the Wiener filter also exists, but it gives conaplex controllers that are it ficult to implement in a real-time control, An advantage of feedforward controller is that if we know the reference tra jectory, the feedforward signal can be computed offline in auvance of the motion. Assuming that G(s) is of the form Tl wa Tf «oo Be) Kat where a/<0, 1sisnj B20, 1s j [aa | Assuming T> = F; = 0, we obtain (from (32) the following tearsfer functions Outs) __ 588.7154 106.1) a8 (637. 15+ 37903. oa ffs) —d5.5(9- 1273.3 (5) =48.5¢2— 1278, Ba 0,4(8) SF 1637.14 57903.3 ‘Theoretical natural frequencies of the beam were obtained from the poles of the above transfer functions. They are 6.018 rad/s {0.957 Hz) and 40.0116 rad/s (6.368 Hz). ‘These fre quencies were ater measured experimentally and deviations of ‘the measured frequencies irom the theoretical values were about 4 percent. The transfer function given by Eq. @5) is non minimum phase, as expected, exhibiting a positive zero at 35.683. Using the identification procedure described earlier, the pa: rameters of the motor model (Eq. (14)) were determined as: J ~ 0.005529 Ib in. s*, V = 0.01216 Ib in./rad./s, K = 2184 1b in./amp and CF/K = 0.132 amp, 6.3. Motor Position Control Loop. ‘The motor position control loop is shown in Fig. 4. The parameters of the eon trollers were designed (Rattan et al., 1988) to minimize the effects of the friction and to make the motor positioning as fast as possible. The sampling period was 3 ms. The response Of the motor position when @ step input of +0.d rad was applied as references du to the motor is showa in Fig, 11. In this experiment, the tip of the arm was kept fixed in the zero angle position. There wasa significant coupling torque between the motor and the beam (4 beam deflection of 0.2 rad). The setling time was about 50 ms. This is Fast as compared 10 the first natural frequency of the beam, but is not much faster a6 compared to the second. However, we will assume that the transfer Tunetion of the inner loop is 1 and we will show that ood results are achieved with our method even in this case 6a Position Control Loop, 64.1 Trajectory Design. The transfer function of the beam is non-ininimum phase, Thus, according to Subsection 4.1, we choose a quas\-parabolic profile. Acceleration and second derivative of the acceleration are chosen taking into account the mechanical constraints (maximum allowable de. flection of the beam). These are: acceleration = 7875 rad/s, and 2nd derivative of acceleration = 8648,646 rad/s*. In this ‘case, using (1B), filter can be obxained as G,{IGq(s) = + 0.05615 40.0015 66 / Vol. 114, MARCH 1992 Figure 12 shows the ideal quasi-parabolie trajectory Pp(s) tat drives the tip from ~ 0.1 ad to 0.1 rad, and the optimized reference trajectory Jy(¢) resulting from passing Py through fiker G6). 642. Feedforward Term. ‘The feedforward term can be ‘obtained using expression (19) (expressed as a function of the payload m;) as Gufs} = (01121363 + (176.6052 + 2.6191)? 1 +0.05615-+ 0.001573" 4 852.6157 BO onstas.cay” ‘mm appeats only in the first factor. This feedforward term may 6.2262 | | Sis 106 and hy ~ 0.1429, be easily tuned as @ function of the payload using the scheme shown in Fig, 13, This scheme implements expression (37) in such a way that the parameter to be tuned omly effects the {gains of three signals that are known before the motion (ref- erence trajectory and its second and fourth derivatives). The dynamic component of the filter shown in Fig. 13, which re- quires most of the computations of the feedforward term, is independent of the payload value, Substituting mt; ~ 0.12136 1p for our beam, we get 140,05618-+ 0.001571 Gals) = (34+ 1637.18 + 57903, 3) EVES 0.001 45.5(5 + 35.683)" (8) Frnt) On(t) Fig. 11, Experimental response of the motor pestion contol loop to Dap rteroncee Fig. 12 Nomina tajectory 2, and optimum esl tory rence alee ‘Transactions of the ASME_ os nt [aoseis oan? Ye Bin som ate? BE TT Ye, “| 10 fea} —_} ai FIQ.13 Reference trajectories: Fu 88 Oy and Fee Morward term i 3 te os é 6-18 Rotten tales nd enponnlponee oh ip os Gel) Fo 3° ond 1 — ' zd hie g i a aa time (seconds) poe Br ronson oan Fig, 14_ Implementation ofthe fedora Fig 16 E43 Feedback Controller. We first need to caleulate DE wie tives the references for the measured vais Using (23), we get [2888+ 12733) [1204 1273.3) F-12733 F+RB3| 5) 1 ro} l APE 4 et Function 20) ofthe form: Oy = dag 2.5, Se iBete me Weighted the tip position wie the midis issacttd MOlot positions, The optimum controller that wi {7 Compson of he espones of rere imizes the ot unaton Go) Ui 17 Comparison of the responses ofthe motor, mide mass, a A=(0.4428 0.5572 0.0534 0.1828), (40) Ths pllees the closed-loop poles at —3.794/7.395, ana shat tsH0-506. The closed-loop frequencies ac of te sat so celtude as the open-loop trequencies ofthe beara, {it the oscillations are substantially damped this coed loop in navir is experienced only when perturbations ace sraces {hal fast and accurate motions are achieved for this arm. Track- {ngerrorsin Fig. 15 are significantly smaller when the tajecons 5s slower. Notice the high level of noise of the camera mene ‘urements 8 and 6 {eure 18 shows the simulation results of the tip response yrth and without the feedforward term. It can be secs Dens {his figure that the use of the feedforward term remeree vat Spsdy state erzor which is otherwise noticeable, This is became ofthe low gains ofthe Feedback controler. Noe also thar the transfer funetion between the motor and tip positions i ct type zero whereas the transfer function between the mote cals ieanbletely free of oscillation, Because of the limited oy inpiege capability of our compute, the feedback iso mente! with a sampling period of 6 ms (nee that a og itt 8 Fis. 16. Figure 17 compares the resporse ov a ‘™EFRE fo the responses of the middle mass and tp, TH Louth oder parabola reaches the target postion in 0.36 sine he fastest motion that ean be achieved foe eh oe chasatl Producing deflections in the structure beyond sagkalfinit Figures shows hat thetipresches tee yn “NUR an evror of less than 2 perce in Gas Hs ee identification and control of single-tink flexible Youmsl of Dynamic Systems, Messurement, and Controt MARCH 1999, Vol. 1141/67AEE Eee : teenie ipa 7. = ectoroattesdback scive a frectorward wacve neces) Fig. 18 Tip responses with and without th ssorward term ‘rms with lumped masses is studied in this paper. Most of the flexible arms have their mass distributed along the structure. But there are some applications in wiiich a model using lumped masses may be useful. A typical example is the case of @ lightweight flexible arm carrying a heavy load. Here, only one vibrational mode is normally important allowing an approx: imate modeling by a massless beam with a mass attached to the tip. Another reason to study flexible arms with lumped imasses is that some properties and control methods for these systems may be extended 0 the distributed mass case. Tech niques described in this paper refer to the following three aspects of flexible ams: modeling, identification, and control. In the ease of modeling, the division of the arm model into motor and beam submodels is general, but the modeling of ‘the beam submodel is specific to lumped mass flexible arms. ‘The identification technique described in Section $ can be applied fo both lumped and distributed-mass flexible arms. ‘The result is a dynamic model of the form (24) and (25) Provided that the mode! of a flexible arm is expressed in the form given by Es, (24) and (25), the control scheme deseribed jn Seetion 4 is general and may be applied to any flexible arm. ‘A method of modeling these systems has been presented in ‘Section 2. in order to deal with nonlinear Coulomb friction, ‘hwo submodels have been defined, one that describes the be- hhavior of the motor (includes friction nonlinearities), and an- other linear submodel that describes the mechanical behavior of the beam. Both submodels are coupled by the torque at the base of the beam, In order to make the model more general, perturbations in the tip, represented by a torque and a force, have been considered, These perturbations allow us to consider the effects of Friction in the tip in the model, changes in the carried load, or reaction forces produced by the nest joint in the case of @ multiplelink flexible arm. Although not used in this paper, some properties have been deduced from these ‘dynamic models in Seetion 3. The most interesting one is the Separation Property that defines the influence of the masses fof the bean in the model. It permits representing the product Of the mass i and its angular acceleration by a linear combi: nation of the deflections of the beam at the points where the asses are located, This linear combination depends only on. the geometry of the beam. ‘A control scheme is proposed that minimizes the effects of, the frietion in the joints. This control scheme is composed of twvo feedback nested loops, a inner loop to control the motor position and an outer loop to contro! the tip position. Tip position control loop is designed in this paper. The design of the inner loop (motor position) was presented in previous pe 68 / Vol. 114, MARCH 1982 pets (Rattan et al, 19888, 1988, 1988). The control scheme for the tip position has been developed from our model and its properties, The controller for the tip position control loop is composed! of feedforward and feedback terms. Nominal trajectory and feedforward term are designed taking into ac ‘count whether the system is minimum or non-mninimum phase ‘A method to optimize the feedforward term was presented for the non-minimum phase case. Other feedforward control methods for flexible arms are based on generating a current (orque) to drive the arm (Meck! and Seering, 1988 and Far: renkopf, 1979). But, they cannot be used when the joint has 2 significant Coulomb friction. Our method is based on gen- trating a motor position command. This simplifies signifi canily the command signal, allowing it to be generated in real time, Our control scheme fs robust to nonlinear frition in the joint and is taken careof by the inside loop. Another important ‘advantage not exhibited by other methods is that this command profile may be easily modified as a function of the actual payload, making it suitable for real-time adaptive control. This 4s a consequence of the Zeros invariancy property AA feedback controller was designed using an optimization criterion, Effects of changes in the tip’s mass were also studied ‘Modeling and control methods have been generalized in Section 5 to the distributed-mass flexible arm ease. Therefore, the resitts presented in this paper are valid for all single-link flew ible arms. Controllers designed here are simpler than the ones dbtained using other methods. This is because of the assump: tion that the inner loop makes the motor respond to position ‘comtiands without any delay and by using more sensing than in other methods. Reconstruction of nonmeasured states from, only motor and tip measurements, using filters or observers, is avoided. Consequent!:, litle computational effort is needed and the measured signcis can be sampled at a very high rate improving the performances of the computer controlled arm. (Franklin and Powell, 1980). A limitation of our method is, the necessity of achieving a motor response faster than the vibvationat modes considered in the weidel of the beau. Finally, some experimental results have been. presented. Modeling and control methods have been applied to @ lumped. tso-mass Flexible arm that we builtin our laboratory. Results ‘atthe limitation mencioned in the above paragraph may be relaxed in many cases. The control system also proved 10 be ‘quite insensitive to noise in the measurements, and to errors in modeling (about 5 percent in the estimation of tatural fre. quencies). The arm used in this paper is very Mlexible. ‘This means that in order to quickly remove small errors in the tip position, relatively larg: defections were needed, and the a tor had fo experience lerge motions ‘Acknowledgments, “The research described in this paper was performed at the Robotics Institute, Ceenegie-Melfon University, Pietsburgb Pennsylvania, The autos are grateful to the associate editor, Dr. Suhads Jayasuriy, and the reviewers for their encouragine ‘and helpful comment: References Book, Ws 1979, “Ana us of Mass lotic Chains wah Servo Contos eiuss! ASME Jounsn or Brune Saves, STeACURENES, ASD CONAN Val ta “Canton, RH, and Sc, E1985, “Peis Cont of Flee Manip sion," Rear’ Reser in 1b, farodtsion to Robotic Mechanesand Contr, AGO Farcnbopi Rt 1972."Optial Oen-Loop Ataacoee Profiles for Hewtlé Sparel Soul of Gudene nd Contra We. 2 80:6 Felis Raton, Sy and Hrowa, 8, TSH, "A New Aporosch © ‘cool Sgn Fue ns Pac 1; Comrl er ae Tip Poston Transactions of the ASME eeFresece of Joint Friction" Techaeal Report FCMU-RL-TR-8G.5, Robotics Insitute, Pustreh, Pensa Foy, VY Raa, KS. and Brown, HB 198, “Mode deifiation of « SiobLink Fee Manipulator in the Pucca of Pilon es oe Sth 184 Annual Modeling nd Silation Conternes,Pasborse Pa Fata v Brown, HB and Rata, Kalai S 1969, "Devine soe Ghatak hg yee Best of reedom Festi A,” Tahoe! Repo HEMT 892, Robots lastiate, Psborgh, PA Ekle, G. Fen Powel J.D 98D, Dig Cantol af Dynanies Stems, Addison Wise. Resdag. MA. “Gupta S.C. and Hasforll Sohn ey & Son, ine asnot TY for Comet oa Hex i So Kuo, 5. C:, 1982, Anvomate Con! Systems, Prentice a {socaster and Tienes, 183, The Theory of Metres, Second don wits Appleton, heavier Foundations of Opimal Control Theos +1970, Fundomenas of Automatic Cone, Sand O1pyee, U., 198, “Assertion Feedba lator Aran" Zn of Robot Stems Vo Tee an Mar, eg ublsing Coy Fotide ow, Ke Hy 1987, A Syseraie Formulation of Dynamic Equations fr ‘overs Manipulators wih Ele Link," Jounal of Robot Systems, Yo. Neo “Matson, F. Fukusing, 8. and camovsers, 19, Fesile Minipalfr sth 3 Pall Dee Michaon 2f Robots Retr, Val 6, Non. Mec, PH cb Seeriag, WP 1885 Stems with Uncenain Resonances (2B Sivor. Ratan KS. Fei "Feedback Contl of "Internationa oven! “Reding Residual Vibration in Contra Systems Magsene, Vol Vand Brown, HB, 188,“ Robust Conte Scheme fer Flexble Ans wiih Frktion Inthe lent" roe ofthe Set sinned Workshop on Spe, Operation, Aeration and Reeves Dae ln aan, K-S. Feb, V. and Brown, He By 38h "Fup te eg ‘Sine Lnk Lihoweaht esis Manpsstor with One Vibrations Nie Pio of the 184 International Conjeone on Rebs end Essel Stes, Howson, "Tex Ratan, -S. Fei, V.,and Brown, 1.1, 18, “deaticaion and Convo! ora si elie Menipuotr, Pro. ef te 27 IEEE Conference ot Decion and Cone, Auth Tena, Fruskesbrodt 199, “Dyaunks and Conte Methed for Moving Flex: ine Sucures sod Ther Appleton ta mds Rete Word Conares on Thvars of Machines and ects For Your ASME Bookshelf DSC-Vol. 32 Micromechanical Sensors, Actuators, Editors: D. Cho, R. Warrington, Jr., A. Pisano, H. Bau, C, Friedrich, J. Jara-Almonte, and J, Liburdy and Systems The papers that cover this multidisciplinary research field rey well as Asia and Europe. They cover: chanical mechanisms, devices, Cations of advanced energy micrc 1991 Order No. HOO695 $102 List / $51 ASME Members present divisions of ASME, IEEE, as fundamentals of micro heat and mass transfer; microme. and actuators; modeling of microdynamical systems; and appli: 0 systems, 362 pp. ISBN No, 0-7918.0863-7 {Per write ASME Order Department, 22 Law Drive, Box 2800, Fired, NJ 07007-2300 ©" call S00-THE-ASME (843-2763) or fex 201-882-1717, #8 beck cover for new Journal covoring miereelectiomechanica systems ‘Journal of Dynamic Systems, Measurement, and Control MARCH 1992, Vol. 114 69