TEBE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 3, NO.10,0CTORER 193, Stick Slip and Control in Low-Speed Motion Brian Armstrong-Hélouvry Absract— Dimensional and perturbation analysis are applica to the problem of stck sip encountered during the motion of ‘machines. The friction model studied motivated by correat {ribological results and is appropriate fr lubricated metal con tacts. The friction model Incorporates Coulomb, viscous and Strbeck friction with frictional memory and ising stale fic: tion, Through dimensional analysis an exact model ofthe nom. inear system can be formed in five parameters rather than en, ‘realy facilitating study and expliily revealing the Interaction ‘of parameters. By converting the system of differential equations into a set of integrations, the perturbation technique, makes approximate analysis possible where only numerical tcchaiques had been avallable before. The analysis predicts the onset of a a function of plant and controler parameters these ‘sults are compared with experimental data, Lsrropucrion HEN moving slowly, machines are likely to exhibit stick slip, a petiodie cyele of alternating motion and arrest. Stick slip determines the lower performance bounds of a machine: the lowest sustainable speed and the shortest governable motion. In applications that place 4 premium on precise motion, an understanding of the ddynamics of stick slip and its possible elimination take on ‘great value, A. Tribology and Controls Systematic study of tribology, the science of rubbing interfaces, began early in this century and has made very considerable progress in the understanding of frictional Phenomena, particularly of engineering materials. ‘The Controls literature concerned with friction is also consid rable; highlights are discussed in Section I-B below. The controls-based investigation of friction, however, has not reached its full potential. This is due in part because the friction models used are sometimes simplistic and inade- quate to represent the observed behavior of servo sys- tems, Advantage has not been taken of the progress of tribology (22) Space permits here only a brief outline ofthe tribologi- cal investigation of the frictional dynamics of lubricated ‘metal-metal contacts. A more complete discussion of the results of tribology important for control may be found in 1, (2). The evidence of the tibology literature justifies a friction model with Coulomb plus viscous frietion—the components normally considered in the controls literature Manasrpt eeived March 27, 192; evied December 8, 1992. Paper recommended by Pst Atwcate Ear) W. Ge The authori with the Department of Electra! Engineering sad ‘Sompuer Scene, University of Wsconsn Miraulee, Miwa, WI ‘Snr TEEE Log Nomber 211899, ind four additional components which shape the behav ior of stick-lip motion in machines: 1) Stribeck Friction; 2) Rising Static Friction; 3) Frictional Memory; 4) Presliding Displacement, ‘The Stribeck cure is the shape of the friction versus velocity curve for lubricated systems, [1], (2), [11], [26 30), and is illustrated in Fig. 1. The figure shows four regimes that will be observed in oil or grease lubricated contacts NNo Sliding, where motion exists as the interface bonding, sites deform elastically; Boundary Lubrication, where slid- ing occurs with solid-to-solid contact because the velocity is not adequate to entrain fuid lubricant into the junc tion; Partial Fluid Lubrication, where the velocity is ade~ quate to entrain some fuid into the junction, but not enough to fully separate the surfaces; and Full Fluid Lubrication, where the surfaces are fully separated by a fluid film. The negative going portion of the curve arises from the contact riding up on a lubricant film in the regime of partial Muid lubrication: as the lubricant film gr0ws thicker with increasing velocity, the friction de- treases. This portion of the (friction-velocity) curve gives 4 substantial destabilizing effect. Tribology can not yet provide a physically motivated model of Stribeck friction, (Of the empirical models that have been investigated, the ‘model employed by Hess and Soom, [19], is used here because of its analytic tractabilty; the third term models, the negative going regime of partial fluid lubrication: FO) = Fy sen (200) + F800) 1 +E) sen XO). “(2 ‘The notation is defined in Section LC below; we note here that, whereas the Coulomb and viscous friction, Fy and Fy are constant, the magnitude of the Stribeck fric- tion, F, i a function of two parameters, reflecting the sing sate triton, ‘The Model presented here has not considered time varying normal foee, which will inuence frictional mem cry and the Stribeck curve (23) Varying normal force is considered briefly, but from a controls perspective, in [1] and more thoroughly in (23) The static friction is the force required to initiate sliding itis the Breakaway force and the magnitude of the Sribeck friction evaluated with #~ 0, In Wbricated ma- chines the highest value of static fiction, F,. 18 not oni8.9286/93508.0 © 1093 IEEERegime I. Zero [Steady Sate Veloiy Friction Force. —» i i i Regime Il, Boundary bh ae ‘Steady State Velocity —= Fg. 1. The peneraize Sribeck ure, showing ton a anton of ‘elo or lon velo reached at the instant that the machine arrives at zero ‘locity (2, (20, 25} 261 stead, in metal-o-metal con tacts with standard lubricants, dhe static fiction rises with time from a lower kinetic friction value to the higher, steady-state static friction level. Fig, 2 is from Kato eta. (20), where a good discussion of this phenomena may be found. Physically, rising static friction arises from the time required to expel the fluid lubricant film from the contact interface, For control, the implication is that i the res dence time at zero velocity is sufficiently short, the stick slip inducing, excess state fition will be reduced, and stick slip willbe extinguished {13}, [261 As we wil see, this process dominates the extinction of stick sip as veloc "ses. Rising stati frietion is modeled (20) r, Fa A where F,s, is the magnitude of Strbeck friction at the beginning ofthe mth sip: F,, is aso the static friction oF breakaway force of the mth stck-sip cle. F,,, is the Stribeck friction magnitude at the end of then ~ 1) slip: f, the dvell time, is the period spent in the stuck condition; and +, a friction parameter, isthe rising static friction rate, A subscript ‘b'indicates 2 quantity evaluated at breakaway, and ‘a’ at arrest. Definitions of sub and superscripts and other symbols are provided in Section 1-C below. In sysicms ranging ftom rock mechanis, (271, through Iwbricated machines, (2, (6, [7 (19), [26h to numerical analysis of transient partial-clasto-hydrodynamic lubrica- tion, (36), a time lag, oF phase shift, has been observed between a change in the sliding velocity and the corre- sponding change in friction. When velocity changes, the friction does not change instantly, but adjusts t0 its new steady state value only after some time. Fig. 3 is taken from the experimental data of Hess and Soom (19] and 7 Ty seconds Fg. 2. Sle econ (breakaway force) ats nction of tine a 70 ‘elect el me) the ences stati ction over Kine cto 1b the duel une. a in ig 7. Labocais 4, B,C and D ate respecte, scout mineral ol cmmerelsidevay ica, estar (Stand paste ol From Kato a (20) courtesy ofthe polher) 40 F(N) 30 o1 Y (m/sec) 00 08 t (eee) 06 Fg. 3. Typical ttn peed ne sift tinal memory, contat Totd = 280 Newtons, scant vscoty = 0.82 Pas min requency "Ha FON): trction Newtons: Va/S eli. (Bom Hes ond Soom 19, courte ofthe publisher) shows clearly the frictional memory. Physically, frictional memory is the result of state in the interface (lubricant film thickness is almost certainly one state variable (23) ‘which must adjust to the new sliding condition before the friction force will attain its new value. Hess and Soom, [19}, present experimental support fora simple lag model ‘The experiments of Hess and Soom, which did not incor- porate breakaway from zero velocity, have recently been extended by Polycarpou and Soom, (24, to include de- tailed friction measurements of the transition from stick- tion to sliding. Dupont, (14), and Rice and Ruina, (271 argue for a state variable model. Here the simple lag. ‘model will be used (= BU ~ 2). @ Where F, is friction as a function of time, and F, is friction asa function of velocity, (1).Preslding displacement is a consequence of clastic de formation of the surface asperities where contact and sliding occur and significantly influences friction forces during velocity reversal (2), (5), [12], [18] I is the compli- ance of presiding displacement that softens the hard nonlinearity of friction at zero velocity, and may make feasible some of the high gain controllers that are found in applications (In thi paper, uniform sliding without ‘velocity reversal is considered and so the state of elastic deformation is constant. Presiding displacement will not, be considered further. ‘The absence of these thee frictional phenomena— Stribeck friction, rising static friction and frictional mem ‘ory—from the models investigated in the controls itera- ture has limited the ability ofthe control theoretic investi- gations to describe the physical behavior of systems. For ‘cxample, a considerable literature exists applying describ ing function analysis to a system with Coulomb plus viscous plus, most often, static friction (the C+ V+ S mode), and investigating the resulting amplitude-depen- dent sablity (atest for limit eyeling or stck-slip motion) ©, (10) (17), (29) (31), [321 A difficulty arises because these analyses do not predict stick-sip motion for a sec- ‘ond order system in the absence of integral control {10}; while in fact mechanisms governed by PD controllers are ‘observed to stick slip. A friction model incorporating nonlinear, low-relocity friction is required to account for the observed behavior. By applying. dimensional analysis, it has been shown that in a second-order system with C+ V +S friction, increasing stfiness should not extinguish stick sip over a broad range of system conditions [4]. Experience, how- ver, shows that increasing stifness often extinguishes stick slip (See Fig. 4 below; and, indeed, stiffening a component is a standard technique in mechanical design for eliminating chatter.) The interaction of stiffness and stick slp is @ consequence of frictional phenomena not considered in the C + V+ S model ‘And the extinction of stick slip with inereasng velocity, while qualitatively anticipated by investigation ofthe C + V 4 $ model, is not quantitatively described correctly. AS carly as the 1950, it was known within the tribology community that static friction rises as a junction spends ‘more time at rest, and that this rising static frietion has a pronounced influence on the continued oeeurrence of the Stick-slip limit cycle, Derjaguin eta, (13), investigate the influence of this phenomenon. In [25], Rabinowicz shows that rising static frietion plays an important role in ex- Plaining the force cycle observed during stick slip. This can be seen in Fig. 4, where increasing stiffness corre sponds to decreasing stick-slip amplitude, This relation arises because the stiffer sytem spends fess time at zero velocity and thus has less stati friction ‘The combined impact of these influences is that the literature concemed with servo control of machines with friction does not presenta predictive capability. The works of Tou and Schulthess, (31}, Shen, (29), Walrath, (35), Canudas and Seront,[10}, and others qualitatively describe Slip Amplitude Delta-F z £ Fie Stikslp ample a 2 funtion of velo, for several spring sins. From Rabinoc (28) coutesy othe pbiser) the observed limit-ycle behavior; but cannot quantita- tively predict the presence, absence or character of this bbchavior in real mechanisms based on parameters of the friction model and control system. B. Prior Work within Controls ‘An overview of the controls literature concerned with Irietion and stick slip can be found in (1, 2), 16) Several papers are mentioned here because they include parts of the friction model described, Tustin, 33), employs an exponential fretion-velocity relation that isa form of the Stribeck curve. He uses a graphie method of analysis to determine conditions for stick sip. Significant, his analy- sis shows that sick slip can oocur in the absence of an integral control term. Recently @ number of investigators have employed friction models incomporatingStribeck fic tion [251 10} [15] 16. These investigations have shown that the Stribeck friction influences the presence of stick slip and the stability of adaptive friction compensation, Derjaguin eta, [13], published in the tibology litera- ture, but their paper is concerned with the dynamics ofa mass-spring-ider sytem that mirrors a servo mechanism With PD contol. By employing a C+ V+ S model, with- ‘out Stribeck friction oF frictional memory but considering rising state friction, they are able to form exact integrals Of the sytem equations of motion. In this way they find velocity dependence in the presence of stick slip that is similar to that observed experimentally. The terms de rived in their dimensional analysis ere comparable tothe terms derived here Dupont, {14}, studies the influence of a state variable friction model on the stability of feedback control. The state variable friction model has its roots in rock mechan- ics [27] and is an alternative to the time lag employed here. As here, though by a very different method of analysis, Dupont finds that stck slip ean be eliminated by sufficiently stiff poston controlThe system to be analyzed consists of a sliding mass governed by a proportional-derivative (PD) controller. The sliding contact friction combines Stribeck friction, fri tional memory and rising static friction. In analysis, the «dynamics of ideal PD control are interchangeable with the those of a mass-spring system with damping. The com: bined dynamic is expressed: Mi = -kyx ~ kt Fysgn (i) — RS oO WOE, 7) sen Gs G4) 30-7 : wanna lis ED 9 Peete ee ‘The notation used is defined below. The friction mode! is made up of F, and F,, the customary Coulomb and viscous friction terms, and F, (VC), which models the Stibeck friction. The Stibeck tition model is divided into two components: the magnitude on the n'* sticksip le, F,,(), which has dimensions of force, depends ‘upon the well time and is used to model rising static fiction; and a velocity dependent term, YC), which is dimensionless, anges from 2ero to one and incorporates the Stribeck friction curve and frictional memory. F,.. is the magnitude of Strbeck friction at the end of thé (1 — DY" stickslip ole; the details of the Stribeck frie- tion magnitude model are presented more flyin Section IV below. The case statement inthe definition of WC) is used to reflect the fact that YC) = 1 prior sliding. The physical dimensions of linear sliding re used here: with appropriate changes of dimension, rotational motion mey be considered by the same analysis C. Nomenclature [units given for translational motions] M = mass al stiffness, [Newtons /m] deri. feedback or damping, [Newtons/m s~'] F = total frietion force, (Newtons) F, = kinetic friction, [Newions} FE = dimensionless kinetic friction; Ft (F/M/ap?) F, = viscous fition, [Newtons /m *] ,() = instantaneous magnitude of the Stribeck frie tion, [Newtons] F... ~ magnitude of the Strbeck friction after long time in contact, (Newtons) 5, = magnitude of the Stibeck friction at time ¢, ‘on the nth stick-slip jee, [Newtons] F | ~ magnitude of the Stibeck friction at time 4, EF... on the ( — 1)” stick-slip eel, Newtons} Fy, = steady state magnitude of the Stibeck fric- tion, [Newtons] F3 = dimensionless F, [1 Foy, = the dimensionless Stribeck friction at break- away ofthe 1 stick-lip cle, (1 dimensionless F, = velocity and frictional memory dependence in Stribeck fiction, ] “YC) in dimensionless coordinates] Ki, = ky + Fes merged damping and viscous friction Vk, 7M; the natural frequency, (5) p= ,/20; the damping factor, (] 5,84 = postion, velocity and acceleration in physical coordinates, mete £,£,£= position, velocity and acceleration in dimen- Sionless coordinates, & ~ dimensionless position at the moment of break-away & = dimensionless position at the moment of ar= rival in static friction {ig = desired sliding velocity, m s-"] = dimensionless desired velocity; &, = fy/%,. 4, = characteristic velocity ofthe Stibeckftietion, fms] = ,/05 sealing factor for length, fn} {B= w, scaling factor for time, [5] 1 = time, (s] 1 = duration of the slip portion of the stck-slip oye 5} duration of the stick portion of the stickstip gle, dwell time, [] -y~ characteristic time of the static friction rise, bs} 17, = magnitude of the time lag in sliding friction, i) 1.03 = ff, in dimensionless coordinates, [] yesri = yeti in dimensionless coordinates, 1’ radius vector of a trajectory in the phase plane, ig. 7 Ef ~ dimensionless kinetic energy, [] ez, ~ change in dimensionless kinetic energy during, slip.) yy = enceay dissipated during a slip eye by damp- ing. F1 ‘energy contributed to a slip eyele by spring windup, due to static fetion, [1] ‘Agq,~ perturbation tothe sip energy due to Stribeck frition, [] ‘Mgy,~ velocity. and frictional memory dependent portion of 3, 1, = time at the moment of break-away, [5] 1 = time at the moment of arrival in static frie- tion. fs)[ARMSTRONG-HELOUVRY: STICK SLIP AND CONTROL IN LOW-SPEED MOTION ‘ysle in dimensionless coordinates, (:) static friction reduction factor, [-] the signum function i sen) Subscripts: b marks a variable given at the moment of break-away, the commencement of slip in the stick-sip cele ‘@ marks a variable given at the moment of arrival in static friction, the end of slip in the stick-lip eyele Superscripts: * indicates a dimensionless quantity TL DiMESStoNAL ANALYSIS ‘The system under study is modeled as a nonlinear differential equation in the system state, (4), coupled to a nonlinear difference equation in the magnitude of static friction, (6). The combined system is one of 9 parameters, 1M, ns gs Foes Yo Eu Tas Fad F,,. We shall consider the situation in which the desired motion is steady, ie. 2, is a constant, creating a tenth parameter. The object of this paper is to determine the range of these parameters ‘over which such a system will exhibit stick slip. Dimensional analysis will provide @ reduction in the ‘number of independent parameters from 10 to 5, yielding a simpler model and greatly facilitating the perturbation analysis to follow. The first transformation is a shift of the Position coordinate to place the equilibrium point at the ‘origin. Physically, this shift corresponds to finding the point at which the proportional. term balances the Coulomb Friction and the viscous and Stribeck friction due to steady motion. The shift eliminates the parameter F, from the dynamic equation. Folding the viscous friction parameter into the derivative feedback, one may write Ko~k, + F,, Defining: Ft kiy + ntese(Eo}) ye then & o (ote: 2 = # because £, = 0, only steady desired motion is considered) With this shift of the position coordinate, (4) becomes: Me kya Ke Grey - reveal ‘The sgn(i) functions are eliminated: when the com- ‘manded motion, z,, is steady motion in one direction, the velocity will not Teverse under quite general. circum- stances [13]. Equation (8), coupled with (6), has 8 indepen- ent parameters, i, included, in 3 physical dimensions: mass, length and time. The Buckingham Pi theorem, [8] indicates that the system may be described by (8-3) mensionless groups. Scaling the position coordinate by a factor a, and the time coordinate by a factor 1/8 gives dimensionless position and time, € and 1*: ag=x' and 0 = pr o Forming the derivatives of £ wr the new time coordi- nate, we find that awe a ae oe, a0) Substituting & for x" and its derivatives in (8) gives: Map'é= ~k,aé —K,apé— Rly") p(atrSla)-e[-ztal} ow where & is the magnitude of the desired velocity in dimensionless coordinates; «BE, (Note: {= 0 is the desired dimensionless velocity, not § &. Dividing through by the mass yields system with scaled force constants. Writing (11) in terms of w and p, the customary second order system parameters, and divid- ing through by a8? gives: te Y= By, and af = Bry o, BOtaDM a | aplé+ 4.) i (4) «2 By choosing the time scale, 8, to eliminate w and the length scale, a, to climinate ,, the dimensionless dy- namic model is formed: B= w- yK/if and a (3) Giving the dynamic equation Fa -€-208- FL OADM(Ert), 0where p = (Kt/M)/2e and Poe Fh RR, as 1 1 Tere Te) eae eae V4(6) 1+ (8 and where F,Cy*, 3) = (CRCy*.12)/M) Lap?) = CF, ODM) Cok Ds tt = 0. ‘The dimensionless model is given by (13)-(16), a non- linear differential and difference system in the five param- eters p, Fla &. 7, and y*. By setting B= w time measurements ate scaled to units ofthe natural frequency of the system. Thus as stifness increases, perhaps due 10 an increase in control gain, rf and y* increase as well By scaling lengths according 10 a ~ &,/«, the characteris tic velocity of the Stribeck friction becomes the unit velocity. Thus, systems are scaled so that the range of ‘motion velocity influenced by Stribeck friction, a range ‘which will vary between systems, will correspond 10 di- rmensionless velocities of a fixed range. Roughly speaking, the dimensionless velocities influenced by Stribeck frie: tion lie in the range from zero to ten, The dimensionless system has unit sifines, that is ky, if it were considered, ‘would always equal one. As physical stifiness changes both a and 6 change, and thus the scaling is changed for all ofthe parameters und state variables. In the dimensionless coordinates energy is proportional to radius on the phase plane, (£? + £%) (when discussing the energy of a system under feedback contro, the pro: portional term is treated as a virtual spring), Writing in shifted physical coordinates, the system en: ergy is given by an) Energy in dimensionless coordinates is given by: B= 3(@ +2) «as) ‘The derivation of the dimensionless model does not depend on the structure of the friction model, except 10 the extent that the Stribeck friction is modeled as a function of (i/4,). Thus, as tribological understanding improves and new friction models are presented, the dimensional analysis ean be adapted, 1, Pexrunssrion ANALYSIS Equations (13)-{6) present the coupled differential and diflerence equations governing the low velocity mo- Ee" = st )( ee" of) +28) (1+ (Ge a9) + ay (t+ ()) + tt went 6) tion ofa servo controlled mechanism. No analytic solution of equation (14) is available. And, while simulation is a useful tool in specific instances, the model has five inde= pendent parameters, creating a space of systems too large ‘o fully explore numerically. To determine the parameter combinations that signal the onset of sticksip motion, a perturbation technique will be used. “The unperturbed system is taken to be the full system less the viscous and Streck friction terms and less static {rietion, which is modeled by F,,. the Stribeck fition at the beginning ofthe n' stick-sip cycle. The trajectory of the unperturbed system is simply a citele on the phase plan, centered at the origin and of radius fy. The trajec- tory of a sample stick-slip cycle and the corresponding Lnperturbed trajectory are shown in Fig. 5. The sample trajectory was generated by numerically integrating (14) With the parameters Fi = 14; y* = 20; rf = 10; & = 0.75 and p ~ 03. (A Tull investigation of friction in & Unimate PUMA robot arm is presented in [5], 2 these are reasonable data for the base joint ofthat mechanism.) The circle is the corresponding unperturbed trajectory This sample was selected to lie near the boundary at which stick slip is eliminated. Using. the unperturbed trajectory the influence of damping and the friction forces can be calculated as Perturbations. The problem is converted in this way from fone of analyzing a differential equation, (14), into one of teprating a set of equations along a fixed path. Stick slip can be detected as @ condition on the sum of the energy contributions of the perturbation terms. The usefulness of this analysis rests on the fact that the perturbation terms, viscous and Stibeck friction, depend only on velocity, and, for a broad range of system parameters, the velocity profile of the unperturbed system isa good approximation to the velocity profile of the full system. This is particu- larly true near the stability boundary, where the perturba: tions sum to zero, Fig. 6 isa plot ofthe velocity profile of the sample trajectory of Fig. 5 and the corresponding Uunperturbed trajectory. The sample trajectory has been adjusted in time to show the correspondence of the sam- ple and unperturbed trajectories. The two trajectories are substantially separated on the phase plane because their positions diverge asthe integral of differences in velocity, but the velocities themselves are quite similar. Note thatARMSTRONG HELOUVRY: STICK SLIP AND CONTROL IN LOWSPEED MOTION “ 15 i g > weet py | sept cat 8 3 05 ase Pane % STinecony § o & : a : (postion) 6 q 1a & -05 os Dimensionless Position Fig. 5. Phase plane tajcores of sample sickslip le and the ‘irespoding unperturbed ty ay 3 os £ oo 5 aii § 7-05 E ao Gere 0.00 010 0.20 030 0.40 Time [Seconds] Fi & typo esp pas of he sample sd wpe tuted trajectories of figute 8A ime offen he beet aed fo he Spl tayectory the perturbation term does not depend upon position, but only upon velocity. Experimental evidence also supports this choice of unperturbed trajectory. In data of (3), [4, it is observed that over a factor of 20 range in stifness and 30 range in velocity the slip distance of a stick-slip cycle is closely approximated by xy. = 2w%,; that is the period ‘of the unperturbed system times the desired velocity. Fig. 7 illustrates the definitions of terms used in this analysis. The axes of the plane are dimensionless position, &, and velocity, € Terms referring to breakaway are subscripted‘b, thus Fi the level of Stribeck friction at breakaway, and F, isthe level of friction at breakaway of the 1’ sticksip éyele. The position at the moment of breakaway is &; and 1, isthe radius of the trajectory at breakaway. The energy of the system at breakaway is Lat vr ; Ava att ysraton of symbols sein the sais siven by E* = (1/2)r. Because § = 0 describes the sy fem moving atthe desied velociy, é = ~ &, corresponds to the sjstem with a physical velocity of zero icy in the stuck condition. The time 1} isthe (dimensionless) time spent during the slip portion of the cycle, and 1 is the time ofthe stick portion ofthe cycle. The time of break- away is designated and the time of arzval into the stuck condition as ¢,; thus, the time of sliding is the interval from t, 10 t We are concerned with the kinetic energy atthe end of an orbit on the phase plane. The dimensionless energy at breakaway is Ef ~ 1/2r? = 1/21} + 62). To interrupt the limit-eycle of stick slip, the velocity after an orbit on the phase plane must not reach € = —éy Ci = O in physi- cal coordinates). This gives rise to the condition that: Aes Bes, < a9) Where A,, is the change in dimensionless kinetic energy ring the Mp eel; and ps = 1/245 isthe degree by which 1/27} exceeds 1/28; "The positon of breakaway, &, is the point at which the spring force exceeds the friction. In dimensionless coordinates, considering the shift of origin, it is given by: cD) Equation (19) isa statement of the condition that more energy must be dissipated during the slip cycle than is provided to the system by spring windup (pesition feed- back) at breakaway. Only systems which exhibit steady ‘motion or bounded frictional instability are considered here. Neglected are systems with divergent modes of instability, such as &,, negative‘The change during an orbit ofthe dimensionless kinetic ‘energy may be written: 2o6— FEW" ( Ett) Edis = = flr pédar ~ [Peer ate ért)éar en Or writen Sez 7 Arg + Sey, @ where A, and Az,, are symbols for the respective inte- grals: Ay, is the energy dissipated by the damping term; and g,, is the energy contributed or dissipated by the Stribeck friction. Exact evaluation of (21) would require knowledge ofthe state along the trajectory, which isto say the Solution of (14). As a calculable approximation, (21) can be integrated along the path of the unperturbed trajectory. ‘The specifications of the unperturbed trajectory ate: EU) sin); E04) = ~& cos 9; Ose <2; 3) ‘where fis the radius of the circle on the phase plane. Using the trajectory of (23) in the first integral of (22) ives the energy dissipated by damping: -f pébdt® = ~2ape} ‘The second integral in (22) is the change in system energy due to nonlinear friction. Integrating along the lnperturbed trajectory gives: des 2a) 4g, = -Fl [Ceora 2 euyaee re (G* — af) +8) 41 +f -glene| - nt sin (=f) an eos") det Lao tf cn he G0 cost — PT Separating the integral from the multiplicative factors yields: ep, = Fs, 26) where cose") sin (sp) + [5 — ae, ht BU cose = aD +1 en wah 20 4 for sj equals 00, 05, 10, 15, 19 20 30 50 Dimensionless Velocity Fig. 8 The integral of (27) evaluates numerically function of "8 ‘nd ‘The integral of (27) cannot be evaluated analytically, but can be evaluated numerically a8 a function of its two independent parameters, & and 7}. The result of this numerical evaluation is’ shown in Fig, 8, where radial Perturbation is plotted as a function of é, and 77. Fig. 8 shows that as the frictional memory grows larger the destabilizing influence of the nonlinear friction is re- duced. ‘Combining the terms of (22) and the condition for 0 and (29) is not verified for any value of slifiness. In the presence of frictional memory, however, A, and therefore a, decreases and finally becomes nega. tive' with increasing stiffness, as indicated by Fig, 8, and (a~b) <0 is assured. Frictional memory explains the observed fact that across a broad range of mechanical systems, stick slip can be eliminated by increasing stiff: ness, 25]. For low velocities in systems with moderate friction, the level at which stick slip will be eliminated corresponds roughly to: wen or kee (30) Iustrated in Fig. 9s the sticktip extinction boundary 8 a function of normalized stifness,k,/M, and dimen sionless desited velocity, &. The extingtion boundary is shown for several values of damping, p. The figure was obtained by evaluating (28) using the parameters of the PUMA joint 1, exclusive of which is et to zero for Fig. 9. The parameters of the PUMA joint 1 are: M~ 6. them’) F< = 16 fkgm/sh, x, ~ 0050 [sh 7 = 0264 Ish 2, = 016058 frad/sh, F, = 4.94 [N-m/rad/sh. [2h Sh The region below a stic-sip extinction boundary com prises combinations of system stiffness and desited veloc- ‘ty fr which stick slip will oceur, as indicated by 28) and the condition Ars, 2 — Age. In the region above an eX- tinction boundafy"Sy,, <~ Ape and stick slip will not ‘occur. As expected the regime of stick slip is reduced by increasing p. More intriguing is the fast thatthe required damping sa decreasing function of stifiues, and, pethaps unexpectedly, there are values of stiffness and desired velocity for which stick slip will not occur, even if the system possess no damping. Systems which exhibit smooth ‘motion in the absence of damping are posibie because of the influence of frictional memory. The unusual mix of Physical and dimensionless coordinates in Fig. 9, a well 4 Fig. 12,13, and 14, was chosen to enhance interpreta- tion. In ‘the dimensional analysis, sifiness is scaled 10 unity and itis frition, f(y", (2), which scales inversely with physical sifness and could be used a8 the dimen- sionless ordinate variable. Bu stiffness, rather than fre tion, is normally a variable under engineering. control. When 7, = 0.050 [s, ,/M = 19740 (rad/s*] conte: sponds t0 f= 1; this approximately the value of normalized stfness at which the undamped system of Fig. 9 is stabilized for low velocity motions TV. Ta Daracr oF RISING Stanic FRICTION ‘The level of static friction at breakaway, F,,,. is not constant. During motion the level of friction ‘d¢ereases. ‘When motion stops, the static friction builds with time from a reduced value to its ultimate steady state value. ‘This process is illustrated in Fig. 2. The process is one of the lubricant erceping out of the asperty junction sites. ‘The implications of rising static friction for stick slip are for pequals 0.0, 0.01, 03, 0.7 and 2.0 $000.0 ap = 20000 | % 10000 00 0.0 50 100 Dimensionless Velocity Fig 9. Sticktip eniaction boundary under the inten of rion ‘memory. Calelted wag equation (28) wit», 50 seconds end Foy, Fs (eng sae fton nt comer substantial: if the length of dwell time, ¢,, is short i relation t0 the static friction ise time, y, the static friction at breakaway is reduced and the amplitude of the stick-slip cycle diminished, As the amplitude of the stick: slip eye is diminished, the dwell time is reduced and thus the level of static friction further reduced. If an equilib rium is not reached in this cyele of diminishing static friction and dwell, stick slip will be extinguished. The tribology ofthis process is discussed in (25) [26] and [20] and an analysis of the impact of rising static friction on stick slip—not considering Stribeck friction or frictional ‘memory—is presented in [13]. ‘The interaction between sliding friction and the rise of static friction is not fully understood. One issue that is uncertain, but must be quantified to support analysis, is the level of friction from which the rise of static fiction begins, F,. Derjaguin et al. [13], employ a static plus Coulomb friction model and presume that the rise of Satic friction commences from the level of Coulomb friction. Rabinowicz, (26), suggests that at the instant of arriving at zero velocity, indicated by #,, or ty = 0, the increase of static friction over sliding friction wil be zero. If we consider a more complicated sliding friction model, fone governed by the Stribeck curve and delay, a more complicated friction condition exists at 4,: the level of sliding friction rises as the system decelerates. Rising sliding friction is shown by the data of Hess and Soom, {19}. The data from several sources, [28 [34}, [6 (21), [19} suggest thatthe sliding friction rises to the level of static friction, F,. as steady state velocity approaches zero. Based on these data, a model to be used for analysis is proposed here In Fig. 10 a hypothetical velocity profile and corre- sponding friction profiles are shown, Curve 1a) is the velocity profile and curves 10(b) and 10(c) are friction profiles of two possible friction models. Curves 106) and 10() differ at time ¢. The friction profile of curve 1b) is that given by the Stribeck friction curve and no frictional‘o Ee Fag enteme a Friction | > Faas Foy Foy a he ‘a LF sate a oa . Fretian fete ° Fig. 10. Two pose stating pois forthe se of sate tion (a) ‘Velocity profle () Mos! nepacing fiona lg tition dacontina ‘sa ©) Mode enscering tna Ip. ton coats memory (7, ~ 0) Ifthe stati friction were to start its rise from the level of Coulomb friction, as suggested in [13], a sudden drop in friction at ¢, would be required, this is because sliding friction increases from the level of Coulomb friction as velocity diminishes, ‘An alternative model provides that the rise of static friction begins from the level of sliding friction at ¢,, a8 shown schematically in Fig. 10(€). The level of sliding friction at 1, F,.,. will be less than the value of static friction at breakaway, F,,, because of the interaction of system deceleration, frictional memory and Stribeck fric- tion. Such a friction model is consistent with the dynamic friction model, allows sliding friction to go to the static friction value as steady-state velocity goes to 2er0, and provides that friction is not discontinuous at the instant £,. Using this model, the interaction of F,, y and 7, will be studied. This portion of the analysis i not facilitated by the dimensional analysis and will be carried out in physical coordinates. To represent the degree to which friction is reduced at time 1, the variable is introduced: GD The variable / ranges from zero to one. Physically, relates to the lubricant film thickness at the end of the slip. If deceleration is rapid and the film has not had time (relative to 7,) to become thin, the force required for breakaway—were breakaway to occur at the instant of arrest—would be the level of Coulomb friction: h = I, F,, ~ 0. If deceleration is slower, the lubricant film will have been thinned during the period of low velocity ‘motion and the force required for breakaway would be larger than the level of Coulomb friction: h = 0, F.,, = Faye In either case, after 1, F, begins to grow according t0 (6) and ultimately reaches F,.. its highest and steady-state value. Using the friction model, (6), and the unperturbed trajectory, (23), can be evaluated: ai os(r7)y 4]C — 00s (w 1. © B= casGpy + YU = cso) + ro) 14a a re(Jour ao ‘The approximation, (33), is the frst term of the Taylor series expansion and is valid when fis small. When the approximation would give a value of h larger than one, ‘= 1 may be used: ft cannot be greater than one. Com= bining the approximation and upper limit on yields: oo Combining the definition of h, (3D), and the rising static friction model, (2), gives a diserete recursive equation for the level of static friction at the moment of breakaway, 4. WR, # (Fa ty uation (35) is more complex than it appears, because f depends upon F,,,. A larger F, ,, contributes to a larger Ag: and Ay,; both of which contribute to a larger r,, resulting in'd longer distance to travel during the stick portion of the cycle and thus a longer dwell time. The Sependence of f5 upon F,., is given by =m, | as lke G6)