Sei sulla pagina 1di 22
HON AND FAULT IDENTIFICATION at LING, SIMU iin Move 358 BOND GRA Ea force actuator velocity pickup e } 3 | ky @5>—41 ——_S\ E-m, iy i HA Fig. 13.1 Fig. 13.3 | \ The bond graph model of the system is shown in Fig. 13.2. By a well judged intuition | { we can also draw the equivalent bond graph of the same system as shown in f Fig, 133, extracting the equivalence of a resistance to an activated gyrator, In Fig. 132, with the gyrator, effort e, = 1f,, which is also the same in the resistance of Fig. 13.3 (e;=- fy). If'we consider the two bond graph as shown in Fig. 13.2 distinct views of the system. In Fig. 13.2 the control system used. For instance, the (transconductance amplifier) that w. and Fig. 13.3 we can see two graph is topologically similar to the gyrator models a voltage-to-current amplifier ‘ould perhaps be actually used in the controller Ie is assumed that the velocity pickup gives a voltage proportional to velocity and the force exciter is a current driven device. In this case, to design the value of tansconductance gain (1), one would need to consider the open loop function G(s) between output (flow in bond 3) and input (effort in bond 4). signal flow graph of the system shown in Fig. 13.4 Ina block diagrammatic form the system is LATION AND FAULT IDENTIFICATION d= fe s(c)de (133) gain ive relation for a storage ment is Constituti jor storage Celen 1 = [sae : (134) {Kis the stiffness of the mechanical spring or an electrical capacitance for a where K is the stiffnes derivative control system. Derivative feedback effort (135) where Kp is the derivative feedback gain. The constitutive relation for an / element in differential causality is =m /(), (136 where m is the inertia of a mechanical mass or the inductance of an electrical inductor. In the previous examples the control variable was chosen to be a flow variable. However, in reality one encounters other variables like position (from positions es like encoders, potentiometers etc.) or force (from force sensors). The force variables map to the bond graph based effort variables. These equivalences are given in Table 13.1 and Table 13.2. Table 13.1 is a summary of the cases when the control variable is effort and Table 13.2 represents the cases when the control variable is flow. measuring dev Table 13.1 | Control Equation Constitutive E e=K,fO Bond graph Element | Resistance R | (resistive causality) Inertance / [= (in differential « usality) | (= Kp | Compliance C [Gin integral CONTROL STRATEGIES IN PHYSICAL DOMAINS 361 Table 13.1 | Control Equation Constitutive Equation ef) = K,e() ie ; : f= [Kietyde | sey= [ Leeyar a id SO=Ky ZO LO nz Eo position measurements represented in bond graphs by Q, one can choose te form of expression from derivative of state O(t) = f(t). Hence C can to represent differential feedback when an effort variable is used as the variable. However, we need to keep in mind that the interpretation of the variables will be different and dependant on the choice of the feedback variable, when using variables other than effort and flow. nce in case of the integral feedback of position error e= «ff Q.(e)dz, (13.7) (¢) is position error, it is difficult to find a equivalent physical concept. The tical tools to analyse such situation is approached. we keep aside such types of control scenarios. We would physically feasible systems and look at issues of control from {ULATION AND FAULT IDENTIFICATION ING, SIM! RAPHIN MODEL 362 BONG! ow may be, Schematically the control lay can jy et GY — shown as in Fig. 13.7. 7 Ibis not necessary, however, to confine to su flow y SS" = complementary use of variables. One pk eo transform ‘efforts to flows and flows to effrn using transconductance amplifiers represency by GY in Fig. 13.8 to achieve any other combin: et id flow measurement to flow nt to effort feedback an Fee rocker tor uth ystems iis diffe find eaten canbe eee comparison from the physical universe so that control engines readily © can easily visualize them. Fig. 13.8 ot of system power variables which are compatible Id therefore look at the set of sy: Biondi graphic representation of single port J, C and R elements. Each of these ie Binet or Rowris input variable and the complementary power variable for uses ei output. Further, with such a standpoint we are also confining our attention to driving point feedback control. These physical equivalences carry meaning when used in the driving point feedback control mode. We now can formulate some control schemes with this standpoint exploiting Physical system equivalences. The attention is once again on realizable systems with ‘natural or physical equivalences which are more friendly to a system designer from a physical standpoint. 13.3 Control system desi; ign from physical standpoint as an inverse problem AS discussed in the previous section one can see the rel Physical system elements and the control variables (effort or flow) for broad classes of physically realizable system elements ic Physical concepts (Table 13.1) one without taking re, theoretical tools of lationships between various and R. Also using expressions and can now assign gain values to control elements Neco mumerically and computationally intensive contel Nyquist, Bode, Root locus plots ete For instance in the proble made from damping eon arrive at u by placing the Move into the m given in Fig. 13.1 the siderations, From Pole inside the le left half plane the more ie the choice of parameter R (= 4) ca control theoretical stand, point one ft hand side of the s - plane, The damping factor, / CONTROL STRATEGIES IN PHYSICAL DOMAINS 363 ‘ cri gis physically relevant and would concept of critical damping is more physically the P sign judgement of an en e's tical damping condition in the parameter space. In our physical eters for natural frequencies (decided zineer more sound. It would be desirable to first nt we would therefore arrive at param\ stiffness values) and damping factors (decided from mass, stiffness and J. These parameters would then be used to decide the control law tategorise this activity as an inverse problem solution wherein we find Jaw gains from equivalent physical systems desired (as a result of the Pephis fs in contrast to obtaining an end physical system after designing the Sing In the former, physical concepts are used first and the design evolves, ithe latter the design is validated for physical equivalence performance ol law gains are obtained from control theoretical analysis fated that the control theoretical tools for analysis of such systems are still ‘can be used for optimal control, robust control and adaptive control of is. The design from a physical standpoint may be considered as a n evolving the control scheme for physically realizable systems. This is feasible due to the physical presentation of the bond g: tion of simple yet physically important concepts and the alg, in bond graphs keep those physical concepts in the foreground in the helming control scheme problems in solving the inverse dynamics of plants when Strategies, Typical plants have certain characteristics like friction, backlash, compliances arising out of coupled coriolis + difficult to estimate and are not constant. These make the inputs on the basis of inverse dynamics solution very nit ae ‘control schemes which make sensitivity to disturbances The designs of such control schemes are often. h-gaii theory. In this section we show how we Eemndpoint that gives rise to a similar control (HON AND FAULT IDENTIFICATION, 4464 poxn Geary MODELING x(t) m+Am et 0) ucla Fig. 13.10 ig. 13: il i ‘hat would be the force to be app i to describe a trajectory x(0) whal ; i ‘i re Csi force on the mass (m+ Am) required to obtain the desired trajectory on ? would be Fu()=(n-+an)sC). an Clearly, with an uncertainty and variation of Amthe determination of the force F(t) is difficult. Consider a case Fig. 13.10 where we rigidly attach the nominal mass m with Am Variations to a very large mass M which is known. M)\(m+ Am). We prescribe the desired trajectory x(1) to the M. Therefore, given the condition of the ‘€quation above, force needed to obtain the t rajectory would be F)=M x). (139) The force, /,(:), decided by the rea, nominal mass m, needed to drag along the nomi: ction force at the Ti Bond graphs of the sy: nal mass will be automatically igid link between the large mass M and the stem is shown in Fig. 13.11. Im+Am iM i FQ):SE—t_»y] K 3 O:SE—3_ yt he pe Fig. 13. 11(a) CONTROL STRATEGIES IN PHYSICAL DOMAINS 365 fe. 13.11(a) the inertance element for the mass (m+ Am) is differentially implies that the two inertances M and (m+ Am)are algebraically nd graph model we can also introduce a scale for the magnitudes of the own in Fig. 13.12. -:M* SE TF mam Fig. 13.12 s graph the mass M* and the mass (m+ Am)are algebraically linked by the with modulus ())!) such that the system is equivalent to that in re the systems in Figs. 13.11 and 13.12 from their system equations. For n Fig. 13.11(b) the system equation is e . M aaa) (13.10) aon -(4 6) v0. equation for the system in Fig. 13.12 is we" m+ Am Bee Ptus(). MODELING, SIMULATION AND FAULT IDENTIFICATION (ODELING, 8 366 BOND GRAPH IN — ee Fig. 13.13 The transformer represents a lever with the lever ratio, p= 5/a, being large due to a long effort arm b))a When viewing the system in Fig. 13.13 one can see that the consequence of the large modulus is that one needs to apply large force #7) to achieve a desired trajectory of mass point (m+ Am). It may be difficult to generate such inputs in any real s stem. Let us now view the transformer in the system as given in Fig. 13.14(a). Fig. 13.14(a) Fig. 13.14(b) In the Fig. 13.146) the transformer is r effort and a flow activ Braphs in Fig. 13,14(a) effort activated tr resolved into two separate transformers with an ated path for each. One can easily show that the two bond and Fig, 13.14(b) are equivalent when the flow activated and ‘anstormers have the same modulus, Now if we propose to use split 13.12. We also propose a modification by which we brea tans a pike activated transformer path has a modulus Of unity "Mplication of this assymmetry is discussed at the end transformer model in Fig, 13,12 the symmetry wherein th as shown in F this chapter CONTROL STRATEGIES IN PHYSICAL DOMAINS 367 pel 1 sams s «ane 2 Fig. 13.16 A F mean thHFO) Pa=-M ay By nt ia Hy Be M waft a) 32) He=1 and uf))I /AULT IDENTIFICATION NG SIMULATION AND F 368. pow Graritin MODEL f output flow and input force. » the Laplace transforms of where f(s) and e(s)are Inthe above equation when pi)" and Hy =! (m+Am MyM” which leads to reclea (13.15) es(s) M's ‘The above equation implies that the dynamics of the feedback system now depends *. The variation and the presence of the plant mass, (m+ Am), only on the mass, 4 does not effect the dynamics significantly when j1y ))1 and jy = 1. We thus say that under the high gain, j1y ))1, the controller mass, M*, overwhelms the plant mass, (m+Am), dynamics. In other words, the dynamics of the plant mass, (m-+Am) becomes a small perturbation on the controller mass, M*, dynamics. We could use the above equation to obtain input force to the controller, eg(s), for a given trajectory, Ais), through an inverse dynamics mass, M’*, solution solely on the ba of the controller The controller mass, IM*, drags along the plant mass, (m+ Am), with appropriate linked by the feedback factor, py =1, which e larity of motion. In other words, the pl control mass force. The motions are res the simil lant mass, (m+ Am), tracks the trajectory of M?, exactly. We could call the control mass, Af*, as st mass which is the spirit tl a 'e control mass, M’*, as a ghost ma i i t ee Shost mass which is the spirit th wes th ae lately determined force y(t) correctly. ANY. variations in the plant inertance, . oa Plant Inertance, Am, is also overwhelmed by the presence of the ghost We can immediatel shows the bond nd graph of a second oj with compliancy and damg j Yy extend the conce; Pt to other characteristics as well, der plant and a second order ghost considered as well ping CONTROL STRATHGIES IN PHYSICAL DOMAINS 369. Ctkp | lo x mw 2 P 1-0. p Timp sa. ¥ San e S mi P—9TF Ritp Keie RRe Fel —__ Mps@atpstkp | ‘all mp LESTE 2 fue f Mes?+Res+Ke h Sr ee Fig. 13.18 em in Fig. 13.17 the signal flow graph can be obtained as shown in ‘function of the system is » FAULT IDENTIFICATION MopeLiNG, SIMULATION AND FAUL 1p GRAPHIN il ove concept. Systems with nonlinear ang above . tis ‘can be brought under the purview of « strategy. A control system designer needs, strate es aie My, Ke and R,, to Work out desired the ghos! eters, implications of the ee jant character! weakly determinable Pl robust control using the 0% therefore, to concentrate on plant performance criteria. has been applied successfully in multi-degree of degrees of nonlinearities and coupled state Mukherjee(18] and Kumar 35}. This overwhelming control scheme vith high dom robotic system wi ‘ ane ‘The reader may refer to Ghosh [17], 13.5 Impedance control In the previous sections we presented control systems with the control variables as “ther effort or flow, the factors of power. These control systems are typically designed from the point of view of the control variable being a physical signal (force velocity, current or voltage etc.). These signals are either effort or flow variable and represent certain behavioral aspects of the system and are external parameters for the system, One proposition in control systems has been the case of impedance control by Hogan [28]. In a land mark paper Neville Hogan suggested a control scheme for robotic systems wherein the impedance, the ratio of the effort and flow variables, is adjusted toa desired value to perform certain tasks which need either effort control or flow control. The concept of impedance control may be considered as a system based control where a system's intrinsic characteristic parameter is the variable of the control scheme. This is in contrast to the signal based control where the control Parameter isa signal that is an extrinsic parameter for the system. Bond graph can be used very effectively to describe and design such control schemes. In this section the impedance cor system parameter based cont describe the controller entire! ntrol perspective is presented as a physical trol scheme. Bond graph provides the capability t0 ly in physical system terms itself without having to take *ecourse toa signal based approach and mathematical rigor. The : yysical pacers taken in the design of the overwhelming control strategy in the physi raphe, Cone ri 0 the dealin GPa aitasgn eae strategy using bond arsider the system bond graph in Fig. 13.17. The controller, a mas= spring—damper based “sed system overwhelmingly cor eC Pointusing a high feed forward gain Mn i r Let us consid, er that ‘Ger that the plant is tery tht no conte by te ee See of the system is she own in Fig. 1 a by the controller as in Fig. 13.19, CONTROL STRATEGHS IN Prysical Domaine 377 yaniv robust overwhelming foundation controller 1 Do robot | ek | 1 1 A ay el hak hat Fig. 13.19 oe oH Ate s = : * x a am 7, 0 u 5 : 0 Hig an—* ; s : ° os hee RRe 7 . RoR Fig. 13.20 HMiction between the reference input, ¢, and the output flow, /,, can be Sials) _ Hy R(s) 4b) ERGO) + sry Cle)” FG) (13.18) soe aN FAURE ROENTINEA TION 372 bono Guar w MODELING sun) The feedback compensation is shown by an effort activated transformer T the bonds 18 and 19 with a modulusa . The effort act a activated transformer ser 2 flow due to the presence of the foundalien Gia et foundation can be nullified. This can be ver oud and te done for the syster vanpenaator dosgn the ens uronanent 4 the robes wil cenpenaanon ga pt at the at CONTRON STRATEGIES IN PHYSICAL DOMAINS 373 the transfer function from the usual signal flow graph and the fils) RM ny (0 @)c(s) F(s)} eqs) 1 Hae ROVCG) + iy I aCG)EG)* Yoon (s) (13.20) where Ais), C(sdaned FS) same as defined in eqn. 13.18, {et us examine the implications of the foundation and the modulatio eqn. (13.20) in mo it or the given admittance we can obtain the stiffness characteristics in the following eu (t)= £, a constant driving force clio a Reply) | fle nye (=a )e(s) £45) , es Inala clea ale: (1321) yo GRARHIN MODELING, SINE ATION AND FAULT LDENTIFICATION 374 We Mont f PAU aarti Wy GRA . 4 »» | (overwhelming eonteal), 1 foundation compel ation) and pty me 1 (OM : «1 (full foundatic Whene (199 Ayal)|qai HW Ke ony, under full foundatioy vorwhelming control strategy, i that for the fee Is very high and when the foundation int atiffroas point stiffness or compliance is modulated by the We can see compensation, the driving. p* compensation is partial the driving ‘compensation ry out the exercise for damping and inertance and see the One can similarly €@ sutcome of equation (13,24) and eqn. (13.28) 8 th effects of the comp ion gain. validate the intuitive statement on driving polnt impedance modulation t alii . nti compensator gain e The concept of impedance or admittance is immensely physical and pertains to an if of physical systems is often carried out intrinsic property of the system, The d tion with other ystems, Several at the point of int on the basis of impedan synthesis problems wherein the network and indy work J out anni design pr tances circuit consisting of resistances, capacitance re the A, Cand /elements, Ina pioneering work 1a network synthesis approach to design of na passive electric In the context of b pate (11) has shown the relevancl impedance controllers, In the example presented In Fig, 13.19 we have seen how we relate the concept of & modulation gain to that of Impedance control, The foundation ompensation Concept as well as the subsequent Impedance modulation concept has evolved from the bond graph de ription of the system, These concepts have evolved due to the sociations with physical syst concepts carried out in the bond graph domains, 13.6 Physical system transposition tin the bon Braph space The design or de shown to be in the ion of the control concept for the ayatem in Fig. 15.19 haw been resin bond #raph domain itll, We can une the power of the Plon transposition the foundation in the ayaton The foundation of Fig, 15,19 jy connects ( * physical foundation ane bond ba Yolval robot, Lot uy ‘ap the effect of "ITUCtUTe a8 ohown In Fig. He foundation to the py (8 HON by the con ih troller) to TE Lag et tiny Fig. 13.22 this the physical foundation is ma the controller. The equivalence shown in Fig. 1322. The bond gi Ped to an equivalent virtual foundation split transformer is obtained by using the raph algebra can be seen in the equivalences es shown in Fig. 13.23(a) and Fig. 13.23(b). FAULT IDENTIFICATION ING, SIMULATION AND 376 BoND GraritiN MODE s)_ firls) Inalno)= 2th = Ais wofirpyt-acoFol (13.26) as Din i a)C() FG) and the driving point stiffness as b By(t-a) (13.27) Froal@) Hu Ke Ma Ky For fy = By) and a #1 (13.28) Kos a) and for a=land py = By» Kop) ay = (1329) Equations (13.28) and eqn. (13.29) correspond to the same in equations (13.24) and eqn. (13.25) respectively. The virtual foundation in Fig. 13.22 is entirely a bond graph creation. The design of this controller has been from the bond graph concepts and drawn from equivalences which can easily be constructed in the bond graph domain. Simple transpositioning al systems concepts. Such controllers are realized can be carried out from phy: either in active electronic circuits, made from operational amplifiers, or from digital control system implementations. The virtual foundation in the controller domain need not be actual physical circuits or devices but can rather be “digital” equations representing the state of the control system equivalent and can still have the effects oF properties of the real system due to the equivalences established in the mannet described As is evident in the case of design of controllers described in sections 13.2 to 185) impedance controller can also be described in physical terms which are friendly. In the following nple we consider the case of an impedance) d for a robotic system. Example: Impedance ‘ample: Impedance controller design for a robotic system Ifa ic system were t ¢ to interact w exch tM ith another sys 9 as ange of power Be of Power between the syste CONTROL STRATEGIES IN PHYSICAL DOMAINS 377 i flow/elfort z flow. | ‘an seen by the aia oe nent robot ' environment effort flow | fort t i Fig. 13.24 Fig. 13.25 Is described in section 13.5 the system in Fig. 13.19 depicts the interaction as shown lin Fig. 1325. 1agiven interaction between the robotic system and the environment it may se be that a particular impedance characteristic is required fmay be realized by appropriately designing th nts of the foundation in domain. Once this is carried out, one can implement th g the rirtual foundation in the design. at the interaction per ern he contrchler by iniphysical system as shown in Fig. 13.26. The single degree of freedom iis passive degree of freedom internal to the controller. The bond graph of fi the same as in Fig. 13.22. Suppose we want the system ° geas shown in Fig. 13 to behave as a with a prescribed admittance given by Fs) Ma? + s+ K 1). WICATION ULATION AND FAULT IDENT sim ELING, Guarittn MOO! ¢ 378 BOND’ 4 -\-;— 9 3.30€-1-/ 4 : a ‘ 2 ; Cc a 9 = Phase \ a 8 | 10 Frequency Fig. 13.28 . assive Now we need to determine the parameters of the foundation fan eee 1 ‘ degree of freedom) from the principles mentioned so far. This can y 13.30) equating the expressions for the admittances in equation (13.26) and equation ( given by A)fl+ By (a Van R)CU)+ By (ie (1331) __ (s)F(s) os +k Using controller parameters, M, 1.Ke=100, R= 14, wy = By =200, gains 0.5 and comparing the like foundation parameters M , «5 Ky This can be verified b equation (13.26) with the aboy, robot inertance m, terms of s in equation ( 100, and Ry = 30that will y the bode plot of the transfer © parameters, This plot will be: result CONTROL STRATEGIES IN PHYSICAL DOMAINS 379. 13.28. Small variation, not so perceptible, would be due to the robot cone in Fi above impedance controller with virtual dissipative foundation has been used ta Success in cooperative manipulation of compliant object even in the presence of wil motion dependant forces [36] this chapter thus shows how the power of bond graph language can be used to develop control strategies from physical paradigm giving the designer the fullest fdvantage of using his intuition and innovative ideas. 43.7 Implication of asymmetric transformation two activated transformer structure shown in Fig. 13.15 with unequal moduli are in several realization of control strategies in this chapter. This structure has very icant physical implications. The effort activated transformer with modulus is in fact a flow sensor with no power associated with it. The flow activated with so called high gain is an effort to effort amplifier. The actuation ver at the output port of this amplifier comes from a tank system sustaining it.

Potrebbero piacerti anche