You are on page 1of 35
Chapter 7 Sliding Control In this chapter, we consider again the control of nonlinear systenis ofthe general form studied in chapter 6, but we now allow the models to be imprecise. Model imprecision may come from actual uncertainty sbout the plant (e.g. unknown plant ppatametets), or from the purposeful choice of a simplified representation of the system's dynamics (e g,, modeling friction as linear, or neglecting structural modes in 44 reasonably rigid mechanical systems). From a control point of view, modeling inaccuracies ean be clasified into two major kinds: + structured (or parametric) uncenainties *# wrsiewctured soncenainties (or womodeled dynamics) ‘The first kind corresponds to inaccuracies onthe terms actually included in the model ‘while the second kind corresponds to inaccuracies on (i.e, underestimation of) the system order. As discussed earlier, modeling inaccuracies can have strong adverse effects on nonlinear control aysiems. Therefore, any practical design must address them explicily. Two major and complementary approaches to dealing with mode! Uncertainty are robust control. which we discuss in this chapter, and adaptive control, which is he subject of chapter 8. The ypical structure of 2 rebust controler i composed of a nominal part, similar to a feedback linearizing or inverse control iam, and of additional terms aimed at dealing with model uncertainty. ‘The structure of a adaptive controller is similar, but in addition the model is actually updated during 276 Sect. 73 Siding Surfaces 277 operation, based on the measured performance. AA simple approach 1 robust conteol, and the main topic of this chapter, is the so-called sliding contol methodotogy. Inciively, itis based on the remark that tis rauch easier «9 control (S-order systems (e., systems described by. LS-order differential equations), be they nonlinear or uncertain, than itis 10 control general ridorder systems (i.e, systems described by nforder differential equations. Accordingly, & notational simplification is intrluced, which, in effect, sllows nihorder problems to be replaced by equivelent 15-order problems. It is then easy to shove that, for the mransformed problems, “perfect performance can in principle be schieved in the presence of arbitrary parameter inaccuracies. Such performance, however, is obtained at the price af extremely high contol setvity. Tis is typically at dds wit the other source of modeling uncertainty, namely the presence af neglected dynamies, which the high contol activity may excite. This leads us to a modification ofthe contol laws which, given the admissible control stivity. is aimed at achieving an effective wwade-off between tacking performance and parametric uncertinty. Furthermore, in some specific applications, partivularly those involving the contol of fectric motors, the unmodified contol laws ean be used directly. For the class of systems to which it applies, sliding controller design provides a systematic approach tthe problem of maintaining stability and consistent performance in the face of modeling imprecisions. Furthermore, by ailowing the trade-offs berween modeling and performance to be quantified in a simple fashion, it can illuminate the whole design process, Sliding control has been successfully applied ta robot manipulators, underwater vehicles, automotive iransmissions and engines, high-performance electric motors, and power systems. “The concepts are presented first for systoms with a single control input, which allows us to develop intuition about the basic aspects af nonlinear contreller design. ‘Specifically, section 7.1 introduces the main concepts and notations of sliding contrat, ‘and illustrates. the sssociated asic controller designs. Section 7.2. describes modifications of the control laws aimed at eliminating excessive control activity. Section 7.3 discusses the choice of controller design parameters. Section 7.4 then studies generalizations fo mult-input systems. 7.1 Sliding Surfaces Consider the single-input dynamic system 2100 f(x) +b) 0 av a 278 Sliding Control Chap.7 system), ihe scalar « is the cooteol input (for instance, a motor torque), and xefe £ ... x-D]P is the state vector. In equation (7.1) the function f09) Gn senerab nonlineae) is not exactly known, but the extent of the imprecision on fUX) is upper bounded by a known continuous function of x ; similarly, the control gain OC) is rot exactly known, but is of known sign and is bounded by known, continuous functions of x. For instance, typically, the inertia of a mechanical system is oaly known (@ & certain accuracy, and friction models ony describe part of the actual friction forces, The control problem is to get the state x to track a specie time- varying state y= [ay fy af-NYF in the presence of madel imprecision on fla) and 8. For the tracking task ¢o be achievable using a finite control «, the initial desired state x4(2) must be such that x40) = x0) a2) In a second-order system, for instance, position or vetoeity cannot “jump”, so that any desired trajectory feasible from time t= 0 necessarily starts with the same position and velocity as those of the plant. Otherwise, tracking can only be achieved afiec & ansieat ‘where the sealar x isthe output of interest (For instance, the position of a mechanical ¥ 7.1.4 A Notational Simplification Let ¥=x--.nybbe the wicking evr in the variable a let Eo. Hey be the tacking error vector. Furthermore, let os define a time-varying surface S(O in fhe state-space RI”) by the scalar equation s(x:1)= 0. where sone a3 and 2 is a sicictly positive constant, whose choice we shall interpret later: instance, if n= 2. +e ‘.e., 5 is simply a weighted sum of the position ervor and the velocity errors if FORME Sect. 7.4. Sliding Surfaces 279 Given initial condition (7.2), che problem of racking X= %y is equivalent to that of remining om the surface St for all 1> 0; indeed «0 represents & linea differcatal equation whose unique solution is 0, given initial conditions (7.2). Thus, the problem of tacking the n-dimensional vector x, can be reduced to that of keeping the Scalar quantity at 2e0. More precisely, the problem of tacking the w-dimensional vector xy (Le, the original n-order tracking problem in 1) can in effect be replaced by a order sabilization problem ins. Indeed, since from (7.3) dhe expression of 5 contains HD , we only need to differentiate s once forthe input u to appear. Furthermore, bounds on s can be directly translated into bounds on the tracking eeror vector %, and therefore the scalar s represents a tme measure of tracking performance. Specifically, assuming that %(0) = 0 (he effect of non-zero toitial conditions in X can be added separately), we have VIZO [1S => HIZO, [TH] < QFE aay F200. 1 n 1 : See zee Re ee 1 oaks Figure 7.1.2: Compating bounds on © whore © = /2-! Indeed, by definition (7.3), the tacking error ¥ is obtained from s through a sequence of first order lowpass filers (Figure 7-L.a, where p = (di) i the Laplace operator). Let» be the output ofthe firs filer. We have ios MENS aT From |s| <@ we thus get [yl < fe M-Dar=@ay eM) = OA We ean apply the same reasoning t0 the second filer, and so on, all the way to Ynot =F. We then get me 280 Sliding Controt Chap.7 LF] < ane! =e Similarly, 0 can be thought of as obtained through the sequence of Figure 7.1. From the previous result, one has || < M2!) , whore 2, is the output of the (=i 1)! fer, Furthermore, noting that D2 P#hmR pra peh 2 3 | Jem oe fh Bode 7 blocks Computing bounds om Figure 7.11 ‘one sees thatthe sequence of Figare 7.1-b implies that Ie) < got) One fe, bounds (7.4). Finally, in the case that %(0) #0, bounds 74) are obtained asymptotically, fe, within a shor time-constant (i ~ 1 3 “Thus, we have in effect replaced an al-order tracking problem by a ‘order stabilization problem, and have quantified with (7.4) the coresponding transformations of performance measure. ‘The simplifies, (Corder problem of keeping the scalar + at zero can now be achieved by choosing the eontral law w of (7.1) such that outside of S(t) S22 sab @s) ‘where 1) is strictly positive constant, Essentially, (7.5) states that the squared “distance” to the surface, as measured by 37, decreases afong all system trajectories. “Thus, it constrains trajectories to point towards the surface S(0), as illustrated in Figure 7.2, In particular, once on the surface, the system trajectories remain on the surface. In other words, satisfying condition (7.5), or sliding condition, makes the sutface an Jnvariant set, Furthermore, as we shall see, (7.5) also implies that some disturbances or dynamic uncertainties caa be telerated while still keeping the surface an invariant Sect. 7 ‘Sliding Surfaces 281 ‘el. Graphically, this corresponds to the fact dhat in Figure 7.2 the trajectories off the surface can “move” while still pointing towards the surface. S(e) veritying (7.5) is referred to asa sliding surface, and the systems behavior once on the surface is called siding regime or sliding mode. ‘su Figure 7.2: The sliding condition ‘The other interesting aspect of the invariant set (0) is that once on i, the system. Urajectories are defined by the equation ofthe set ise, namely (on! In other word, the surface S(@) is both a place and a dynamics. This fac is simply the geonieiri interpretation of our earlier remark that definition (7.3) allows us, in effect, {0 replace an n"-order problem by a "order one. o Fivally, satisfying (7.5) guarantees that if condition (7.2) is not exactly verified, jie. it {r=0) is actually off x,(r=0) , the surface 5(0) will nonetheless be reached in a Fini rime smaes than [9(7=0))n. Indeed, assurme for instance that s(7=0) > O, and let reach be the time required to hit the surface s=0. Integrating (7.5} beaween £0" and /= fegcy tends 10 0-00) = sf0tpegen) ~ S(0=0) 5 Mb peacty OP 282 Sting Control Chap. 7 which apes thar “each $ 50-0 ‘One would obtain a similar result stating with s(¢=0) <0 and thus im Furthermore, detniion (7.3) implies that once on the surface, the tracking enor tends exponentially to 2210, with time constant (a ~ 1A dfrom the sequence of (n= 1) filters of time constants equal to 1/3). ‘The typical system behavior implied by sacisfying sliding condition (7.5) is illustrated in Figure 7.3 for n= 2, The sting surface ts a line in the phase plane, of slope ~ 2 and containing the (cime-varying) point xg [xg 4I". Starting from any initial condition, the state trajectory reaches the time-varying surface in finite time ‘smaller than js(=4)V/n, and then slides along the surface towards Xq exponentially, fime-constant equal 10 17h. ao Figure 7.3 : Graphical interpretation of equations (7.3) and (75) (a= 2) ‘in suramary, the idea behind equations (7.3) and (7.5) is to pick-up 9 well behaved function of the tracking error, s, according to (7.3). and then select the feedback control law ein (71) such that s? remains 2 Lyapunov-like function of the closed-loop system, despite the prescnee of model imprecision and of disturbances. ‘The controler design procedure then consists of two steps. First, as will be itustrated in seetion 7.1.3, a feedback control law w is selected so a5 to verity sliding condition (7.5), However, in onder to account forthe presence of modeling imprecision and of Sect. 7.1 Siding Suefaces 283 disturbances, the contrel law has to be discontinuous across S(t). Since the implementation of the associsted control switchings is necessarily imperfect (for instance, in practice switching is not instantaneous, and the value of sis not known, ‘with infinite precision), this leads to chavering (Figure 7.4). Now (with a few important exceptions that we shall discuss in section 7.1.4), chattering is undesirable in practice, since it involves high control activity and further may excite high- frequency dynamics neglected in the couse of modeling (such as unmodeled sieuetural modes, neglected tne-delays, and so on). Thus, ia second siep detailed in section 72, the discontinuous contro! law w fs suitably smoothed to achieve an optimal trade-off heaseen control bandwidth and wracking precision: while the first step accounts for parametric uncertainty, the second step achieves robustness t high- frequency unmodeled dynamics. haering xo ° [igure 7.4: Chavering 25 «result of imperfect contol switehiags 7.1.2 * Filippov’s Construction of the Equivalent Dynamics ‘Te system's motion on the sliding surface can be given an interesting geometric imerpretation, as an “average” ofthe system's dynamics on both sides of the surface. ‘The dynarnics while in sliding mode can be weitten as s=0 76) By solving the above equation formally for the control input, we obtain an expression for called the equisafent control, ty. which can be interpreted as the continuous eu 284 Sliding Conerot Chap. 7 control law thar would maincain #20 if the dynamics were exactly known. For instance, fora system of the form vw have lagna ft iy ted he sym ynanics wien ding ode, of course, £f eg = 3 Geometicely, the equivalent control cen be constructed as, Meg BOtaty 0 on ice, 8a convex combination of the values of w on both sides of the surface St). The value of a. can again be obtained formally from (7.6), Which corresponds to requiring that the system trajectories be tangent to the surface. This intuitive construction is summarized ia Figure 7.5, where fal ft]? , and similarly FeLi fru Wand fag =Li f+ tg} Its formal justification was derived in the early 1860's by the Russian mathematician A. Filippov. igure 7.8: Fiippov's constuction ofthe equivalent dynamics in sliding mode Recalling thet the sliding motion om the surface cotresponds 10 a limiting ‘behavior as control switchings occur infinitely fast, dhe formal solution cof (7.6) and (7.7) cat be interpreted as the average "residence time” of the (rajectory on the side 520. Sect. 7.1 Sliding Surfaces 288 7.1.3 Perfect Performance - At a Price Given the bounds on uncertainties on f(x) and (x), constructing @ control law 1 verily the sliding condition (7.5) is straightforward, as we now iustrace. ABASIC EXAMPLE, Consider the second-order system 78) where wis the control input, « is the (Scalar) output of imerest, and the dynamics f {possibly nonlinear o time-varying) is aot exactly known, but estimated asf. The ‘estimation error on fis assumed to be bounded by some known function F = F(x,8): U-AsF a9 For instance, given te system ean 80830 a0, swece at is enon bt ris eau s2 Pets e0e3e F055 (co 361 In order 0 have the system cack a(o) x40), we define a sliding surface s =O according to (7.3), namely: sa (Geayi a BaF Gay We ten have: atotyentaseuntyont 12) “The best approximation of «continuous control law that would achieve $=0 is thus heap ety nF (7.13) ‘Note that im terms ofthe discussion of section 7.1.2, 8 can be interpreted a our best estimate of the equivalent contol, In order 10 satiny sliding condition (7.5) despite uncenainy on the dyamics f, we add tof aterm dnconinunus across the surface 286 Sliding Conirol Chap.7 w= lm k sans) aay ‘where sga isthe sign function: +1 ifsso <1 ifs4, . Thevefore, $0, and the actual state converges towards the desired siate. Furthermore, slkding mode apd its robust properias are maiatained on the surface ,=0, which tends towards the desired sliding surface a the observer converges SWITCHING CONTROL IN PLACE OF DITHER ‘When uncertainty consists of effects small in magninide but difficult to model, such as stiction or actuator ripple, switching in s may be advantageously used in place of a more standard “dither” signal. Ideally, the frequency of the switching should be chosen well beyond that of significant structaral vibration modes (in. mechanical systems), while remaining below the actuators’ bandwidih, "This assures again that ‘meaningful state estimates can be provided at she selected switching frequency. Such ‘an approach can particularly improve the quality of low-speed behavior, which ‘orherwise is extremely sensitive to friction. ‘The examples above represent the few specific applications where chattering ccan be tolerated and actually exploited. In the general case, however, the question is how to derive control laws that maintain the system close to the surface $= 0 while avoiding chattering altogether. Ths isthe subject of the next section, 7.2. Continuous Approximations of Switching Controt Laws In general, chattering must be eliminated forthe controler to perform properly. This can be achieved by smoothing out the control discontinuity in a thin Boundary layer Oe 25) Sect. 7.2 Continuous Approximations of Switching Control Laws 291 Figure 7.6. : The boundary layer where @ is the boundary layer thickness, and € = Y/R is the boundary layer width, 5 Figure 7.6. illustrates forthe case » = 2, In other words, outside of B(d), we choose Control law w as before (i... satistying sliding condition (7.5)), which guarantees that the boundary layer is attractive, ence invariant all trajectories starting inside B«¢=0) remain inside B(e) for ali £20 ; and we then interpolate w inside BCA) — for instance, replacing in the expression of u the term sgn(s) by sft, inside BUA), as ilastrated in Figure 7.6.. Given the results of section 7.1.1, this leads to sracking to within a guaranecd precision & (eather than “perfect” tacking), and mare generally guarantees that forall trajectories staring inside B(r=0) Wi20, Wms@aye 7 Example 7.2: Consider again the system (7.10), and assume shat the desired wayectory i q=8atHP2). Figure 77 shows the tacking erro and contol la using te switched conta Ja (with sn.) wah espe, . 1842 05 30+ 3y=20F~ (05 fos 3x14 01) aga # 207] 292 Sliding Coutrot Chap.7 Figuee 7.6.8: Congo intepolation inthe boundary layer at sompling rte of KHz, The acu value of o) used in the simulations is a(9 = sine + 1 ‘cic verifies the assumed bound on a(t). We se that racking performance i excelent, but is ‘btsined atthe pie of high control ehaterng _Assome now tat we interpolate be above cont input in thin boundary layer of thickness. o vette) 58 c08 304-202 ~( 05 fens 3x1 +01) sal +207) 704] |As shown in Figure 7.8, the tncking perfomance, while not as "prfet” as above, i stl very god, ads now achieved wig a smooth contol law, Note thatthe bounds on tacking ero ate consistent with (7.25), a Bo Bo Bg & seas ad a coun] a) ” s05| sa Mo saa timetsco) [igure 7.7: Switehed contra input and resuling tacking performance Sect. 7.2 Continuous Approsimations af Switching Control Laws 293 69) Ben aol F aa] al £ os] a ea ts 135 So ab ug asa timesecd Ais) Figure 7.8: Smooth contol input aod resulting Wacking performance ‘The intuitive understanding of the effect of control interpolation in # boundary layer can be carried on further, and guide the selection of the design parameters 2. and ®, As we now show, the smoothing of contral discontinuity inside Bit) essentially assigns a fowpass filter structure to the local dynamics of the variable s, thos climinating chatering, Recognizing this fillersike structure then allows us, in essence, to tune up the control law 50 as to achieve a trade-off between tracking precision and robustness 1o unmodeled dynamics. Boundary layer thickness 4 can be ‘made cime-varying. and can be monitored so as to well exploit the control “bandwidih” available. The development is first detailed for the case (= 1 (a0 gain margin), and then generalized, Consider again the system (7.1) with bBo t,t order mann tractive of de boundary lager nw ht slowed fo vary with ine, must scaly miy contin (79). now need paraee tha the ditnce 0 the boundary layer always decreases, 320 4 2 {so} 5 - Gl-e1 sn 4 ss-@ => Ls-C0) so Ge One a ‘Ths instead of simply requiring thst (7.5) be satisfied outside the boundary layer, we how fequie that (combining te above equations) id 2h seo ee (b- mis (7.26) “The additional term df in (7.26) celles the fac tha the Boundary layer aration condition is more stringent during boundary layer contraction 4 < 0) and less Stringent during boundary Laer expansion (® > 0) In onder to salisly (2.26), the 294 Sliding Comrot Chap.7 quantity ~ is ndded t0 condrot discontinity gain K(x) , Le. in our smoothed {implementation che term &(x) sgn(s) obtained from switched control law w is actually replaced by k (x) sat(si®) , where Fen) =x) - an ‘and sat isthe saturation function, which can be formally defined as sayy iTS sai(y) = sgnty) otherwise Accordingly, control law « becomes: - i uxt -Fis) sitet) il Lav us now conser te system aectories inside the boundary layer, where they le by cnsroton they ea be expressed rely in terms ofthe variable s 8 da Foy 5 > ase 7.28) where A=} —f. Now since F and Af re continuous in x, we can exploit (7.4) rewrite (7.28) in the form be-Fog £ + (Asay +00) 028) We see fiom (7.29) that rhe variable s (which is a measure ofthe algebraic distance 10 the surface (2) can be viewed as the ouput of a first-order filter. whose dynamics only depends on the desired state x49, and whose inputs are, to the fist order, “perturbations,” Ze, uncertainty Aflx,). Thus, catering can indeed be eliminated, a5 lang as high-frequency unmodeled dynamics are not excited. Conceptuatly, the structure of the closed-loop error dynamics can be summarized by Figure 7.9: perturbations are filtered according 10 (7.29) 10 give s, which io tum provides tracking error & by further lowposs filtering, aécording (0 definition (7.3). Controt action 1 is @ function of x and xg. Now. since A is the break-frequency of filter (7.3), i must be chosen to be "smal" with respect to high-frequency unmodeled dynamics (such a8 "unmodeled structural modes or neglected time delays). Farthermore. we can now tune the boundary layer thickness so that (7.29) also represents a first-order filter of Sect. 7.2 Continuous Approximations of Switching Controf Laws 295 bandwidth A. It suffices to let ‘ eo 030) which can be written from (7.27) a8 dhe = hx aay G+ Fear ster]_2 a ey one corcEor & DeFINTONOF 5 Figure 7.9: Sructute ofthe closed Joop sor dynamics Equation (7.51 defines the desired time-hstory of boundary layer thickness ¢b an, in ‘he light of Figure 7.8, shall be refered to as the balance condition. Inuitively, i mounts toning up the closed-loop system so that it mimics an nS order critically damped system, Further, definition 7.27) can then he rewten 38 Fe) = 4) - Hea 4 2) ‘The s-trajectons ie. the variation of with time, is a compact descriptor of the closed-loop behavior: control activity directly depends on x, while by definition (7.3) tracking error ¥ is merely a filtered version of s. Furthemore, the serajectary represents a time-varying measure of the validity of the assemptions on model uncertainty. Similarly, the boundary layer thickness @ describes the evolution of ‘dynamic model uncensiny with time. It is thus panicularly informative to plot {0}, ©, and ~ (0) on a single diagram, as illustrated in Figure 7.10. [Baample 73: Consider aga the system describe by (7.40). The complete contol la is nose «LSP eosde ~ (05H [cos e]4n ob) sacl «ADHD wee wih b= Hb + OS; Phe0edigh +m) a, assuming ett £{0)=1 initially, 0) «9/2. Asin Bsampe 7.2, we let =O and 2=20. Typically, the arbitrary constant] (which, formally, elects the time te reach the ‘ounvdary layer starting fom the outside) chosen to be salt 33 compared to the average vale of Kay), 50 as to filly exploit on knowledge of the sracture of parametric anceainy, The 296 Sliding Control Chap. 7 Figure 7.10 : The s-tsjectries can convey much information ona single plot ‘value of is select based on the fequency range of wnmodeleé dynamics, as we shall discuss bate. ‘The racking eror, contol input ads aaectovies ate ples in Figure 7.11 forthe same ested trajectory xy = sn(te/2) a in Example 7.2, We sce thas while the maximum vale of he time-varying boundary layer thickness ic the same a8 that originally chosen (pmposefull) 25 the constant value of @ in Example 7, the vacking eer is consistently better (upto 4 times ete) chan that n Example 7.2, cause varying the thickness ofthe boundary aye allows ws to make bet use ofthe avaiable bandwidth, a ie Emo B49 Pca Ba F ceee 0 1203 20] coy limetsee) Figure 7.Lta: Control input an resting iceking performance f Continuous Approximations of Switching Control Laws 297 sarajeeories £8 a 8 a 3 zo timetssey Figure 7A :s-trjectories with dime-varying boundary layers In the case that 8 #1, one can easly show chat (7.31) and (7.32) become (with Bas Bx) Ka) ae > brro=Buu) — Foo=uni~ovp (7.33) hy do b .A2 K é hays 2 bs Adis Fixyenny po a3) “Os Rr iB Be )~B ‘with initial condition (0) defined as (0) = By HAO (7.35) Indeed, inorder to satisfy (7.26 athe pesence of unceraiaty Gen the control aia we et @50 we Fayat-eip 738) <0 > Toyenn po 37) Purthennave, the balance condition ean be writen, instead of (730) 36 Kady ty a op CONG. De thats Top=207, 298 Stiding Controt Chap.7 Applying this relation (7.36), (77) leas to the desired behave of $50 wo Mens of, Be rey $20 > Betisy-B Which we can combiae with 736}0.37) aad rewsite 26 (7.33)47.38). Finally, remark that if B= By one has Fea) = (Fer -Fing) +g) = kod ~Koy) +20, Note that the balance conditions (7.33) and (7.34) imply that @ and thas are bounded for bounded xy ‘The balance conditions have a simple and intuitive physical imerpretation: neglecting time constants of order 174, they imply that We Bghingd thavis bandwidth" (tracking precision) (paramnesric uncertainty measured along the desised trajectory) Such trade-off is quite similar €0 the situation encoprtered in linc time-invariant systems, but hete applies to the more general nonlincar system structure (7.1), all slong the desired trajectory. In paicular, it shows that, in essence, the balance conditions specify the hest racking performance anainable, given the desired control Ihandwidth and the extent of parameter wicertainy. Similarly, expression (7.20) allows one to easily quantify the trade-off between accuracy and speed along a given path, for instance [Example 7a: Let us consider agai the underwater vebicle of Example 7.1, and smoothen the conuol input using ie-vasying Boundary Taser, 35 described above. "he a pier bounds on m! andoare leme5 assests and, secordinsly, Sect. 7.2 Continuous Approximations of Switching Comrol Laws 299 hate tet “The aciaal values sein the simaation are aS L Sif it 212+ 2sintity) which ae used as a metaphor of the complexicy of the actual hydrodynamic effets. We Bt J aad 4=10. ‘The desied trajectory consists of constant acceeration phase ai 2 mis force seconds, 8 constanievelocty phase (at-4 mys) fortwo seconds, and a constantoceleration phase at ~2 a fo wo seconds The corresponding hacking et, conta input, and etsjctris re plots in Figure 7.12. o By Boo z a eon zo 5 on « on 4 ee 2a ast to toe See ee linet) i) [igure 7.12 + Conteol input and resulting tacking petformance Aimesee) Figure 7.12b + s-4rajectvies wit time-varying hoondary ayers 300 Sliding Control Chap.7 REMARKS (i) The desired trajectory xy must itself be chosen smooth enough not to excite the high-frequency unmodeled dynamics. Gi) An argument similar to that ef the above discussion shows that the choice of dynamics (7.3) used to define sliding surfaces is the "best-condivioned” simone linear dynamics, in the sense that it guarantees the est tracking performance given the sired control bandwidth and the extent of parameter uncertainty. Gil) If the modet or ts bounds are so imprecise that F can only be chosen as a large constant, then @ from (731) is constant and large, so that the term F sats/4) simply equals 2.sfB in the boundary layer, and therefore acts asa simple PD. : there is no free lunch ie) A welldesigned controller should be capable of gracefully handiing exceptional disturbances, te,, disturbances of intensity higher than the predicted bbounds which are used in the derivation of the control law. For instance, somebod) may walk into the laboratory and past violently on the system "to see Wow stiff i is" an industrial robot may get jammed by the failure of some other machine; an actuator may saturate 96 the result of the specification of an unfeasible desired trajectory. IF integral control is used in such cases, the imtegrat cer in the control action may become unreasonably large, so that once the disturbance stops, che system goes ‘yough large amplitude esciltaions in onder 10 retum (0 che desired trajectory. This phenomenon, known as integrator windup, is a potential cause of instability because Of saturation effects and physical limits on the motion. 1 can be simply avoided by Stopping integration (ue, maintaining the integral term constant) as long as the system 1 ouside the boundary layer, Indeed, under normal circumstances the system does ‘emnain inthe boundary layer: on the other hand, when the conditions ret to normal after an exceptional disturbance, integration can resume as soon as ihe system is back in the boundary layer, since the integral term is defined to within an arbitrary constant, (u) In the case that A is time-varying (as further discussed in the next section), the term az should be added to the corresponding fi, while augmenting gain K(x) accordingly by. the quantity Ja (B= U ‘The degree of simplification in the system model may be varied according to the online computing power available: in essence, the balance conditions quantify the Sect. 7.3 The Modeling/Performance Trade-Offs 301 ‘uade-off between model precision and tracking accuracy, as further detailed next. Furthermore, dhe s-traectories provide a measure of the validity of the assumptions ‘on model uncertainty and of the adequacy of bound simplifications, 7.3. The Modeling/Performance Trade-Offs. “The balance conditions (7.33)-(7.38) have practical implications in terms of designimodeling/performance wade-oils. Neglecting time-consiants of order Wh, conditions (7.33) and (7.34) imply that Wem Boky 73) ‘as noticed in section 7.2, If we now consider the structure of control law (7.19), we see that the effects of parameter uncersimly on f have been “dumped” in gain k Conversely, better knowledge of f reduces & by a comparable quantity. Thus (7-39) is particularly usefut in an incremental mode, ‘e., (0 evaluate the effects of model simplification (or conversely of increased model sophistication) on cracking performance: Az = By ky") (7.405 I particular, marginal gains in performance ate critically dependent om contol bandwidth 2 if large 2's are available, poor dynamic models may lead co respectable tracking performance, and conversely large modeling efforts produce onty minor absolute impeevements in tracking accuracy. 113s of course not overly surprising that system performance be very sensitive to contol bandwidth 2 : (7.1) only represents part of the system dynamics ~ e2. its rigid-body component ~ while A accounts forthe unmodeled pact. a the righthand side of (7.40) the effects of parametric uncestainty in (7.1) are cefleeted in the rmerator, while the presence of dynamics neglected in the model is eeflected inthe py. then the above equation admits a unique solution y, whose components y; are all non-negative Applying the Frobenias-Pervon theocem to the malex of components Dj, a noticing that oar second assumpion on the system implies that py < 1, shows that equation (7-50) uniquely defines a set of non-negative &; . Thus the controt law (7.49), with k defined by (7.50), satisfies. the sliding condition in the presence of parametsie ‘uncertinies bounded by (7.40) As in the single-input case, the switching contrat laws derived above can be smoothly interpolated in boundary layers, so 35 to eliruinate chattering. chus leading to a unde-off bosween parametric uncertainty and performance, ‘The reader is referred to section 7.6 for details Note that de point of view taken in this section is essentially mathernatical Se 306 Sliding Controt Chap.7 Chapter 9 shall discuss how the exploitation of physical properties of the systems, such a5 conservation of energy, ofien makes mult-input designs simpler and more powerful. This will become particularly important in adaptive versions of the designs. 7.5 Summary ‘The aim ofa sliding controler isto (i) Design 8 contro law to effectively account for ‘¢ parameter uncertainty, e.g, imprecision on the mass properties or foads, inaccuracies on the torque constants of the actuators, friction, and s9 on. the presence of unmodeled dynamics, sch as structural resonant modes, neglected tume-delays (in the setuators, for instance), of finite sampling Fate. iy Quantify the resulting modetingiperformance trade-offs, and in particular, the effect on tracking performance of discarding any particular ‘erm in the dysamic mode ‘The methodology is fased on a notations! simplification, which amotmts to replacing an i onder tacking. problem by a first order stabilization problem Although "perfect" perfarance sn in principle he achicved in the presence of arbitrary parameter inaceacie, uncertainties inthe mode sruture (ie, unmodeled dynamics) lead to & tradeoff between tracking performance and. parametric ven the avilable “bandwidth "In practice, this comesponds 0 replacing control lew by is smonth”approximavion. In speciic where contol charg, is aoeptoble, the pure switching contol ws can yield extremely high performance. Stiding controller design provides a systematie epproach to the problem of tmaintaning stability in the face of modeling imprecisions. Furthermore, it quantifies the modeting/performance trade-offs, and in that sense ean illuminate the whole design and testing process. Finally, it offers the potential of simplifying higher-level programmning, by accepting reduced information about both task and system. Seo. 7.7 Exercises 307 7.6 Notes and References ‘The concept of ashing surface originsted inthe SovistIteraure (a, Aasrman apd Gaptmache, 1957: Eielyanov, 1957; Filippo, (960} (se also (Toypkin, 1955; Flugge-Low, e af. 1958%), mosily in the context of "varisblestucture™ regulation of linea systems, see [Utkin, 1977) fr a ‘eview (aso [Young, 1978)). Classica sliding mode cout, however, hud important drawbacks listing is pracsical applicability, such as chatering and large conto authority. The developrea! of sections 74-73 is based op [Slotne, 1984} The combination of sising consoles with ste observers is Wieessed in Bonar, et 1985) inthe linear ease, and fMedick and Ragaven, 1990) inthe noolinar ease. Observers based on Siding suciaes are dseussed in (Drakunos. 1983: Sloine, ef aly £985, 1987: Waleott and Zak, 1967], The development of section 7.4 i adapes fom (Sltine, 1985; Hedck and Gopulswamy, 1989}. The reader is elated 0, eg, [Luenberger, 1979] fo a simple peoot ofthe Frobenius-Peeroa theoren, Some desis on boundary faye inerpolaions in vie tmatiinput ease can be found in [Slsire, 1985) Practical implementations of sing coriol ane described, eg, in (Yoerper. eat, 1986) (underwater vehicles), (Hedrick, erat, 1988) automotive applications), (Hrashima, er al. 1988) (Godot manipolase). The literature inthe ied has boen extremely ative inthe pas ew Year elated spproaches to yobust conte! include, eg, [Coress end Leitmann, 1981; Gutoan and Patmor, 1982; Ma and Gi, 1985) 17 Exercises TA Consider tbe underwater vehicle of Example 74, and assume there is an unmodeled pendulum mex” of the vehicle at 2H2, Choose A cordingly, and simula the vehicle's pestoxmance on the same najstory as in Example 7.4. Wha isthe minimum sampling rate requced, {0 implemere your design? Discuss the performance of the sysem on various trajectories, which you may want 19 generate using a eeence mode, a in equation (ILS), with & = 22,8; = 22, ‘Simulate the unmodsted 2 He mode by fies passing che cotrotfaw through & second-order lowpass titer of unit. gain before actually inpoting i nthe system. Tune 2 “experimentally” arc the value given hy (7.41, forefront vales ofthe ites daiping 72 For the system by = anny fret fy = aqneytoosey + ayia 308 Sliding Comtrot Chap.7 design a contoller © wack an arbitrary deste tajectory y(@) Assume that he state (2) 23) Is measured, that ayyl) iC Kyl) are all know, on that aff) snd cs) ae unkown timex ‘varying functions verifying de know bounds VEQtonsia — LsaKns2 Write the fall expression ofthe controller, as a Funesion of the measured state [xj 1) - Check OU design in sip simwlations. (i: Fist feeack linearize the system by differentiating the ‘rst equation 3 73. Consider agsin the system of Execse 7.1, but define so as f0 use integral contro (as liscussed in section 7.13), Simulate the cormoller’s performance and compare with Uwe esl of [Exervge 7.1, on various tajactores. Show ta, besides allowing the sytem to ahways stan ats = 0, ‘he integral conzol term Keeps adjusting the contof law as Tong as the fests of bounded Aistorbances are not compensated. 74 Consider a system of We foam (7.1), but whiee is now constant ad of known constant posive hounds. Divide both ses ofthe uation by 4, ad write che Sing comfition a nit < on te nal where dre 7b. By represeuting uacensiny on b adtively. design a simple switching contr to sity the above condition. ‘Smoodh out the sitchingcomrollern bounary layers and derive te comresponting balance condi. Show that the accurtey of the sppravimate "bandwidth? analysis fn the boundagy yee Increases with 7S Consider a system ofthe form (7.1), and assume that not only but aso 5 can be muessured eg hae can be measured), An addtional term ofthe form = a5 (with 20} can then be included in the conta av Wit the coresponding expeston off Assuming for simplicity dat the pain margin Bis constant, show that, given the andwitth % and the parametric uncerainty. the eect of the iinet term in a 15 t0 eee the masimum value of by 8 factor C1 +18 ). Show that this ‘implies thatthe ticking ere can in principle he made eimai smal simply by nceasing 0 ‘What ae the tinitations ofthis approach? In parcularessume tha there is 1 uncertain ‘onthe value of 8). How malt must be to make the approach worthwhile? (Adapted from {Asda and Satine, 19868) Seet.7.7 Exercises 309 76 Shove tht a contin similar (7.68) can also be abiained by requiing that he system tke ‘at eas two sampling periods to cross the boundry layer. ‘Assume that sampling isthe ative limit on bandwidih, and tha chaveing i accepable ‘Based on the above result, how do the cling performances of a switching eonole and a smoot sliding consollor compare? 71 Design a switching controler forthe sytem ayia? + 0gt0.x%cos 2e = Site ‘where ay(0 and cf ae usknown time-varying functions verifying the known bounds Vi2e, faynisi — —Asaynss in: Tee 9= $+, Diseus the effet of chaning ¥ 78 In the context of section 72 define sy = 5 ~ Ossi) ssa measure of distance tothe boundary layer. Show shat the time derivative of 2 is well defined ‘continuous ne, ad, spcifclly, that one can write bade sad Uy pytat = 595 — leah ‘Show that equation (7.26) canbe writen td 18s s ~misgl (Adopted from [Sloine and Coetse, 1986) ) 79 othe context of wicking conte, diseus altemate definitions of = For instance, coos choices based on Buterwont fies rather than (7.3), and how they Would modify bounds (7) 710 Design sliding contoler forthe sytem why 9? +e,C) sin te = HO ‘where 04. eal). and 0) are unknown dime-varyingSureions verifying the known bounds Vi2G Janis — jain? shana 310 Sliding Controt Chap. 7 ‘Assume that the state 55 measured, end that the slowest unmodeled dynamics is the actutce lyons, witha time-constant of shout 1/SD. Simulate the performance ofthe system on various Urajecores (hich you may wan w generate using a eftence model

You might also like