外文翻譯--液壓驅(qū)動的無級變速器控制 英文版

Vehicle System DynamicsVol. 44, No. 5, May 2006, 387406Control of a hydraulically actuated continuouslyvariable transmissionMICHIEL PESGENS*, BAS VROEMEN, BART STOUTEN, FRANS VELDPAUSand MAARTEN STEINBUCHDrivetrain Innovations b.v., Horsten 1, 5612 AX, The NetherlandsTechnische Universiteit Eindhoven, PO Box 513, 5600 MB Eindhoven, The NetherlandsVehicular drivelines with hierarchical powertrain control require good component controller tracking,enabling the main controller to reach the desired goals. This paper focuses on the development ofa transmission ratio controller for a hydraulically actuated metal push-belt continuously variabletransmission (CVT), using models for the mechanical and the hydraulic part of the CVT. The controllerconsists of an anti-windup PID feedback part with linearizing weighting and a setpoint feedforward.Physical constraints on the system, especially with respect to the hydraulic pressures, are accountedfor using a feedforward part to eliminate their undesired effects on the ratio. The total ratio controllerguarantees that one clamping pressure setpoint is minimal, avoiding belt slip, while the other israised above the minimum level to enable shifting. This approach has potential for improving theefficiency of the CVT, compared to non-model based ratio controllers. Vehicle experiments show thatadequate tracking is obtained together with good robustness against actuator saturation. The largestdeviations from the ratio setpoint are caused by actuator pressure saturation. It is further revealed thatall feedforward and compensator terms in the controller have a beneficial effect on minimizing thetracking error.Keywords: Continuously variable transmission； Feedforward compensation； Feedback linearization；Hydraulic actuators； Constraints1. IntroductionThe application of a continuously variable transmission (CVT) instead of a stepped transmis-sion is not new. Already in the 50s Van Doorne introduced a rubber V-belt CVT for vehiculardrivelines. Modern, electronically controlled CVTs make it possible for any vehicle speed tooperate the combustion engine in a wide range of operating points, for instance in the fueloptimal point. For this reason, CVTs get increasingly important in hybrid vehicles, see forexample 13. Accurate control of the CVT transmission ratio is essential to achieve theintended fuel economy and, moreover, ensure good driveability.The ratio setpoint is generated by the hierarchical (coordinated) controller of figure 1. Thiscontroller uses the accelerator pedal position as the input and generates setpoints for the localcontrollers of the throttle and of the CVT.*Corresponding author. Email: pesgensdtinnovations.nlMichiel Pesgens was previously affiliated with Technische Universiteit Eindhoven.Vehicle System DynamicsISSN 0042-3114 print/ISSN 1744-5159 online 2006 Taylor & Francishttp:/www.tandf.co.uk/journalsDOI: 10.1080/00423110500244088388 M. Pesgens et al.Figure 1. Hierarchical powertrain control.The CVT and its hydraulic actuation system are depicted in figures 2, 3. The hydraulicsystem not only has to guarantee good tracking behavior of the CVT but also has to realizeclamping forces that, on the one hand, are high enough to prevent belt slip but, on the otherhand, are as low as possible to maximize the transmission efficiency and to reduce wear. Inpractice, the clamping forces levels are kept at levels that avoid belt slip at all times, whilestill maintaining an acceptable degree of transmission efficiency.The main focus of this paper is on the ratio control of the CVT, using the hydraulic actuationsystem of figure 3. The presented control concept is based on the work of 3, 4. It enablestracking of the ratio setpoint, while guaranteeing at least one of the two pulley pressuresetpoints to be equal to its lower constraint. Even though the controller effectively changesfrom controlling one of the two pressures to the other, no actual switching between differentcontrollers takes place. Among the approaches seen in the literature, some incorporate aswitching algorithm 3, 5, whereas others control only one of the two (usually the primary)pressures 6, 7. Although the former approach cannot guarantee one of the two pressures tobe equal to its lower constraint, the latter cannot explicitly prevent the uncontrolled pressureto stay above its lower constraint.The rest of this paper is organized as follows. First, a mathematical model is derived forthe mechanical part of the CVT in section 2. Next, in section 3, the hydraulic part is modeled.The physical constraints, imposed by the hydraulic system, are discussed in section 4. Theseconstraints are taken into account by the CVT ratio controller, that is developed in section 5Figure 2. Variator.Hydraulically actuated CVT 389Figure 3. Variator with hydraulic system.and is based on the earlier derived models for the mechanical and the hydraulic CVT parts.The tracking performance of this controller is experimentally evaluated in section 6. Finally,section 7 gives some concluding remarks.2. The pushbelt CVTThe CVT (figure 2) considered here is equipped with a Van Doorne metal pushbelt. Thisbelt consists of a large number (around 350) of V-shaped steel block elements, held togetherby a number (between 9 and 12) of thin steel tension rings. The belt runs on two pulleys,namely the primary pulley at the engine side and the secondary pulley at the wheel side. Eachpulley consists of one axially fixed and one moveable sheave, operated by means of a hydrauliccylinder. The cylinders can be pressurized, generating axial forces (clamping forces or thrusts)on the belt, necessary for transmission of torque (without macro-slip of the belt) and for ratiochange. Here the distinction is made between micro-slip, needed for torque transfer betweenbelt and pulleys, and macro-slip, which should be avoided at all times for its negative effecton efficiency and especially the risk of severe belt and pulley wear 8.The bounded transmission ratio rcvtrcvt,LOW,rcvt,OD is defined here as the ratio ofsecondary pulley speed sover primary pulley speed p, so:rcvt=sp(1)In deriving the variator model, it has been assumed that the pulleys are rigid and perfectlyaligned, and that the V-shaped blocks are rigid and the steel rings are inextensible. The beltis assumed to run in perfect circles on the pulleys. Further, it has been assumed that theclamping forces are large enough to prevent macro belt slip. The effects of micro-slip arerelatively small with respect to the ratio change behavior of the CVT, and are, therefore,neglected in the model. The power transmission between the belt and the pulleys is modeledas Coulomb friction (which is assumed in the majority of CVT variator research 3).390 M. Pesgens et al.Using these assumptions, the running radii Rpand Rsof the belt on the primary andsecondary pulleys are functions of the ratio rcvtonly and are related by:Rp= rcvt Rs(2)The axial position s( = p for the primary pulley, = s for the secondary one) of themoveable pulley sheave of pulley is also completely determined by the ratio rcvt. Denotingthe taper angle of the conical sheaves by (see figure 4), it is easily seen that sis given by:s= 2 tan() (R R,min) (3)Subscript max (or min) implies the maximum (or minimum) value possible, e.g. Xmax=max(X), unless stated otherwise. Differentiation with respect to time yields the axial velocitysof the moveable sheave of pulley s= (rcvt) rcvt(4)where the function follows from the geometry of the variator.Assuming that the radial friction force component between the pulley and the belt is zero,the critical pulley clamping force (equal for both pulleys, neglecting the variators efficiency)is given by references 3, 5 (for pulley ):Fcrit=cos() |T|2 R(5)where Tis the net transmitted torque between belt and pulley and is the coulomb frictioncoefficient between pulleys and belt. The factor 2 appears, as there are two friction surfacesbetween pulley and belt.Radial forces between belt and pulleys can be mainly contributed to centrifugal forces andCoriolis forces. In the detailed thrust force model of ref. 9, it is reported that even if thesliding angle (and hence the friction force angle) between the belt path and the friction forcevector changes along the pulley circumference, its value converges rapidly towards values lessthan 10. As a result, the angle is assumed zero. The friction force angle would enter intoequation (5) as a multiplication factor cos(), which rapidly converges to 1 for small angles.For the choice of , a worst-case approach is applied. It is chosen as the maximum of thetraction curve (of which several have been presented in ref. 10), which is the point of transitionfrom micro-slip to macro-slip. The lowest value of all the maxima found in ref. 10, as wellas in ref. 8 (for both very similar variators) is 0.09, the value of that has been used here.Figure 4. Pulley sheave definitions.Hydraulically actuated CVT 391The torque ratio is the ratio of transmitted torque and maximally transmittable torquewithout belt slip for pulley :=TT,max=cos() T2 R F(6)As in a practical vehicle application a good estimate of the torques acting on the secondarypulley is not available, the following modified torque ratio is introduced:primes=cos() Tp2 Rp Fs(7)The estimated primary transmitted torqueTpcan be obtained from the dynamic drivelineequations together with engine and torque converter characteristics (also see section 4). Incase of a perfect torque estimation, i.e.Tp= Tp, it is easily seen that (using equation (6):primes=PpPs s(8)with transmitted power P= T . As it has been assumed that Pp= Ps, the modifiedtorque ratio becomes equal to the torque ratio for the secondary pulley.An important part of the model for the mechanical part of the CVT is the sub-model for therate of ratio change as a function of, for instance, the clamping forces. Sub-models of this typeare proposed, among others, by Guebeli et al. 11, Ide et al. 12, 13 and Shafai et al. 14.The blackbox model of Ide is preferred here, as it reasonably describes the results of a seriesof experiments with metal V-belt CVTs 3, 4.The steady state version of Ides model yields a relation for the primary clamping force Fpthat is required to maintain a given ratio rcvtwith a given secondary clamping force Fsand agiven primary torque Tp(through the modified torque ratio primes):Fp= (rcvt,primes) Fs(9)For obvious reasons, the quantity in equation (9) is called the thrust ratio. Some experimen-tally obtained results for this highly non-linear function of the CVT ratio rcvtand the torqueratio primesare given in figure 5.For instationary situations, Ides model states that the rate of ratio change rcvtis a functionof the ratio rcvt, primary pulley speed p, clamping forces Fpand Fsand torque ratio primes:rcvt= kr(rcvt) |p|Fshift； Fshift= Fp (rcvt,primes) Fs(10)An axial force difference Fshift, weighted by the thrust ratio results in a ratio change, and istherefore called the shift force. The occurrence of pin the model (10) is plausible because anincreasing shift force is needed for decreasing pulley speeds to obtain the same rate of ratiochange. The reason is that less V-shaped blocks enter the pulleys per second when the pulleyspeed decreases. As a result the radial belt travel per revolution of the pulleys must increaseand this requires a higher shift force. However, it is far from obvious that the rate of ratiochange is proportional to both the shift force and the primary pulley speed. kris a non-linearfunction of the ratio rcvtand has been obtained experimentally. Experimental data has beenused to obtain a piecewise linear fit, which are depicted in figure 6. The estimation of krhas392 M. Pesgens et al.Figure 5. Contour plot of (rcvt,primes).Figure 6. Fit of kr(rcvt)； greyed-out dots correspond to data with reduced accuracy.Hydraulically actuated CVT 393Figure 7. Comparison of shifting speed, Ides model vs. measurement.been obtained using the inverse Ide model:kr(rcvt) =rcvt|p|Fshift(11)In the denominator Fshiftis present, the value of which can become (close to) zero. Obviously,the estimate is very sensitive for errors in Fshiftwhen its value is small. The dominant dis-turbances in Fshiftare caused by high-frequency pump generated pressure oscillations, whichdo not affect the ratio (due to the low-pass frequency behavior of unmodeled variator pulleyinertias). The standard deviation of the pressure oscillations and other high-frequency distur-bances has been determined applying a high-pass Butterworth filter to the data of Fshift.Toavoid high-frequency disturbances in Fshiftblurring the estimate of kr, estimates for values ofFshiftsmaller than at least three times the disturbances standard deviation have been ignored(these have been plotted as grey dots in figure 6), whereas the other points have been plottedas black dots. The white line is the resulting fit of this data. The few points with negative valuefor krhave been identified as local errors in the map of .To validate the quality of Ides model, the shifting speed rcvt, recorded during a road exper-iment, is compared with the same signal predicted using the model. Model inputs are thehydraulic pulley pressures (pp, ps) and pulley speeds (p, s) together with the estimatedprimary pulley torque (Tp). The result is depicted in figure 7. The model describes the shiftingspeed well, but for some upshifts it predicts too large values. This happens only for high CVTratios, i.e. rcvt 1.2, where the data of is unreliable due to extrapolation (see figure 5).3. The hydraulic systemThe hydraulic part of the CVT (see figure 3) consists of a roller vane pump directly connectedto the engine shaft, two solenoid valves and a pressure cylinder on each of the moving pulley394 M. Pesgens et al.sheaves. The volume between the pump and the two valves including the secondary pulleycylinder is referred to as the secondary circuit, the volume directly connected to and includingthe primary pulley cylinder is the primary circuit. Excessive flow in the secondary circuitbleeds off toward the accessories, whereas the primary circuit can blow off toward the drain.All pressures are gage pressures, defined relative to the atmospheric pressure. The drain is atatmospheric pressure.The clamping forces Fpand Fsare realized mainly by the hydraulic cylinders on the move-able sheaves and depend on the pressures ppand ps. As the cylinders are an integral part of thepulleys, they rotate with an often very high speed, so centrifugal effects have to be taken intoaccount and the pressure in the cylinders will not be homogeneous. Therefore, the clampingforces will also depend on the pulley speeds pand s. Furthermore, a preloaded linear elasticspring with stiffness kspris attached to the moveable secondary sheave. This spring has toguarantee a minimal clamping force when the hydraulic system fails. Together this results inthe following relations for the clamping forces:Fp= Ap pp+ cp 2p(12)Fs= As ps+ cs 2s kspr ss+ F0(13)where cpand csare constants, whereas F0is the spring force when the secondary moveablesheave is at position ss= 0. Furthermore, Apand Asare the pressurized piston surfaces. Inthe hydraulic system of figure 3, the primary pressure is smaller than the secondary pressure ifthere is an oil flow from the secondary to the primary circuit. Therefore, to guarantee that in anycase the primary clamping force can be up to twice as large as the secondary clamping force,the primary piston surface Apis approximately twice as large as the secondary surface As.It is assumed that the primary and the secondary circuit are always filled with oil of constanttemperature and a constant air fraction of 1%. The volume of circuit ( = p, s) is given by:V= V,min+ A s(14)V,minis the volume if s= 0 and Ais the pressurized piston surface.The law of mass conservation, applied to the primary circuit, combined with equation (14),results in:oil Vppp= Qsp Qpd Qp,leak Qp,V(15)Qspis the oil flow from the secondary to the primary circuit, Qpdis the oil flow from theprimary circuit to the drain, Qp,leakis the (relatively small) oil flow leaking through narrowgaps from the primary circuit and Qp,Vis the oil flow due to a change in the primary pulleycylinder volume. Furthermore, oilis the compressibility of oil. The oil flow Qspis given by:Qsp= cf Asp(xp) radicalBigg2|ps pp|sign(ps pp) (16)where cfis a constant flow coefficient and is the oil density. Asp, the equivalent valve openingarea for this flow path, depends on the primary valve stem position xp. Flow Qpdfollows from:Qpd= cf Apd(xp) radicalBigg2 pp(17)Here, Apdis the equivalent opening area of the primary valve for the flow from primary circuitto the drain. The construction of the valve implies that Asp(xp) Apd(xp) = 0 for all possible xp.Hydraulically actuated CVT 395Flow Qp,leakis assumed to be laminar with leak flow coefficient cpl, so:Qp,leak= cpl pp(18)The flow due to a change of the primary pulley cylinder volume is described by:Qp,V= Apsp(19)with spgiven by equation (4).Application of the law of mass conservation to the secondary circuit yieldsoil Vs ps= Qpump Qsp Qsa Qs,leak Qs,V(20)The flow Qpump, generated by the roller vane pump, depends on the angular speed eof theengine shaft, on the pump mode m (m = SS for single sided and m = DS for double sidedmode), and the pressure psat the pump outlet, so Qpump= Qpump(e,ps,m). Qsais the flowfrom the secondary circuit to the accessories and Qs,leakis the leakage from the secondarycircuit. Flow Qsais modeled as:Qsa= cf Asa(xs) radicalBigg2|ps pa|sign(ps pa) (21)where Asa, the equivalent valve opening of the secondary valve, depends on the valve stemposition xs. The laminar leakage flow Qs,leakis given by (with flow coefficient csl):Qs,leak= csl ps(22)The flow due to a change of the secondary pulley cylinder volume is:Qs,V= Asss(23)with ssaccording to equation (3).The accessory circuit contains several passive valves. In practice, the secondary pressurepswill always be larger than the accessory pressure pa, i.e. no backflow occurs. The relationbetween paand psis approximately linear, sopa= ca0+ ca1 ps(24)with constants ca0 0 and ca1 (0, 1).Now that a complete model of the pushbelt CVT and its hydraulics is available, the controllerand its operational constraints can be derived.4. The constraintsThe CVT ratio controller (in fact) controls the primary and secondary pressures. Severalpressure constraints have to be taken into account by this controller:1. the torque constraints p p,torqueto prevent slip on the pulleys；2. the lower pressure constraints p p,lowto keep both circuits filled with oil. Here, fairlyarbitrary, pp,low= 3 bar is chosen. To enable a sufficient oil flow Qsato the accessorycircuit, and for a proper operation of the passive valves in this circuit it is necessary that396 M. Pesgens et al.Qsais greater than a minimum flow Qsa,min. A minimum pressure ps,lowof 4 bar turnsout to be sufficient；3. the upper pressure constraints p p,max, to prevent damage to the hydraulic lines,cylinders and pistons. Hence, pp,max= 25 bar, ps,max= 50 bar；4. the hydraulic constraints p p,hydto guarantee that the primary circuit can bleed off fastenough toward the drain and that the secondary circuit can supply sufficient flow towardthe primary circuit.The pressures pp,torqueand ps,torquein constraint 1 depend on the critical clamping forceFcrit, equation (5). The estimated torqueTpis calculated using the stationary engine torquemap, torque converter characteristics and lock-up clutch mode, together with inertia effects ofthe engine, flywheel and primary gearbox shaft. A safety factor ks= 0.3 with respect to theestimated maximal primary torqueTp,maxhas been introduced to account for disturbances onthe estimated torqueTp, such as shock loads at the wheels. Then the pulley clamping force(equal for both pulleys, neglecting the variator efficiency) needed for torque transmissionbecomes:Ftorque=cos() (|Tp|+ksTp,max)2 Rp(25)Consequently, the resulting pressures can be easily derived using equations (12) and (13):pp,torque=1ApparenleftBigFtorque cp 2pparenrightBig(26)ps,torque=1AsparenleftbigFtorque cs 2s kspr ss F0parenrightbig(27)Exactly the same clamping strategy has been previously used by ref. 3 during test standefficiency measurements of this gearbox and test vehicle road tests. No slip has been reportedduring any of those experiments. As the main goal of this work is to an improved ratio trackingbehavior, the clamping strategy has remained unchanged.A further elaboration of constraints 4 is based on the law of mass conservation for theprimary circuit. First of all, it is noted that for this elaboration the leakage flow Qp,leakandthe compressibility term oil Vpppmay be neglected because they are small compared tothe other terms. Furthermore, it is mentioned again that the flows Qspand Qpdcan never beunequal to zero at the same time. Finally, it is chosen to replace the rate of ratio change rcvtby the desired rate of ratio shift rcvt,d, that is specified by the hierarchical driveline controller.If rcvt,d0 andQsp= 0. Constraint 4 with respect to the primary pulley circuit then results in the followingrelation for the pressure pp,hyd:pp,hyd=oil2parenleftbiggAp p max(0, rcvt,d)cf Apd,maxparenrightbigg2(28)where Apd,maxis the maximum opening of the primary valve in the flow path from the primarycylinder to the drain.In a similar way, a relation for the secondary pulley circuit pressure ps,hydin constraint 4can be derived. This constraint is especially relevant if rcvt 0, i.e. if the flow Qspfrom thesecondary to the primary circuit has to be positive and, as a consequence, Qpd= 0. This thenHydraulically actuated CVT 397results in:ps,hyd= pp,d+oil2parenleftbiggAp p max(0, rcvt,d)cf Asp,maxparenrightbigg2(29)where Asp,maxis the maximum opening of the primary valve in the flow path from the secondaryto the primary circuit.For the design of the CVT ratio controller it is advantageous to reformulate to constraintsin terms of clamping forces instead of pressures. Associating a clamping force F,with thepressure p,and using equations (12) and (13) this results in the requirement:F,min F F,max(30)with minimum pulley clamping forces:F,min= max(F,low,F,torque,F,hyd) (31)5. Control designIt is assumed in this section that at each point of time t, the primary speed p(t), the ratio rcvt(t),the primary pressure pp(t) and the secondary pressure ps(t) are known from measurements,filtering and/or reconstruction. Furthermore, it is assumed that the CVT is mounted in avehicular driveline and that the desired CVT ratio rcvt,d(t) and the desired rate of ratio changercvt,d(t) are specified by the overall hierarchical driveline controller. This implies, for instance,that at each point of time the constraint forces can be determined.The main goal of the local CVT controller is to achieve fast and accurate tracking of thedesired ratio trajectory. Furthermore, the controller should also be robust for disturbances. Animportant subgoal is to maximize the efficiency. It is quite plausible (and otherwise supportedby experiments, 3) that to realize this sub-goal the clamping forces Fpand Fshave to be assmall as possible, taking the requirements in equation (30) into account.The output of the ratio controller is subject to the constraints of equation (31). The constraintsF F,mincan effectively raise the clamping force setpoint of one pulley, resulting in anundesirable ratio change. This can be counteracted by raising the opposite pulleys clampingforce as well, using model-based compensator terms in the ratio controller. Using Ides model,i.e. using equation (10), expressions for the ratio change forces Fp,ratioand Fs,ratio(figure 8)can be easily derived:Fp,ratio= Fshift,d+ Fs,min(32)Fs,ratio=Fshift,d+ Fp,min(33)where Fshift,dis the desired shifting force, basically a weighted force difference betweenboth pulleys. As explained earlier, depends on primes, which in turn depends on Fs. This is animplicit relation (Fs,ratiodepends on Fs), which has been tackled by calculating from pressuremeasurements.It will now be shown that at each time, one of the two clamping forces is equal to F,min,whereas the other determines the ratio. Using equations (30), (32) and (33) the desired primary398 M. Pesgens et al.Figure 8. Ratio controller with constraints compensationand secondary clamping forces Fp,dand Fs,dare given by:Fp,d= Fp,ratioFs,d= Fs,minbracerightBiggif Fshift,d+ Fs,minFp,min(34)Fp,d= Fp,minFs,d= Fs,ratiobracerightBiggif Fshift,d+ Fs,minFp,minFp,min Fs,ratio= Fshift,dif Fshift,d+ Fs,minFp,max Fs,ratioFs,max(40)If either pressure saturates (pp= pp,maxor ps= ps,max), the shifting speed error inevitablybecomes large. The anti-windup algorithm ensures stability, but the tracking behavior willdeteriorate. This is a hardware limitation which can only be tackled by enhancing the variatorand hydraulics hardware. The advantage of a conditional anti-windup vs. a standard (linear)algorithm is that the linear approach requires tuning for good performance, whereas the con-ditional approach does not. Furthermore, the performance of the conditional algorithm closelyresembles that of a well-tuned linear mechanism.6. Experimental resultsAs the CVT is already implemented in a test vehicle, in-vehicle experiments on a rollerbench have been performed to tune and validate the new ratio controller. To prevent a non-synchronized operation of throttle and CVT ratio, the accelerator pedal signal (see figure 1) hasbeen used as the input for the validation experiments. The coordinated controller will track themaximum engine efficiency operating points. A semi kick-down action at a cruise-controlledspeed of 50 km/h followed by a pedal back out has been performed in a single reference exper-iment. The recorded pedal angle (see figure 9) has been applied to the coordinated controller.This approach cancels the limited human drivers repeatability.The upper plot of figure 10 shows the CVT ratio response calculated from speed measure-ments using equation (1), the plot depicts the tracking error. As this is a quite demandingexperiment, the tracking is still adequate. Much better tracking performance can be obtainedwith more smooth setpoints, but the characteristics of the responses will become less distinctas well. Figure 11 shows the primary and secondary pulley pressures. The initial main peakin the error signal (around t = 1.5 s) is caused by saturation of the secondary pressure (lowerplot of figure 11), due to a pump flow limitation. If a faster initial response were required,adaptation of the hydraulics hardware would be necessary. After the initial fast downshift, theratio reaches its setpoint (around t = 7 s) before downshifting again. All changes in shifting400 M. Pesgens et al.Figure 9. Pedal input for the CVT powertrain.direction (t = 1.3, t = 1.6 and t = 7.5 s) occur with a relatively small amount of overshoot,which shows that the integrator anti-windup algorithm performs well.Looking at the primary pressure in the vicinity of t = 1.5 s, it can be observed that thispressure peaks repeatedly above its setpoint. This behavior is caused by performance limi-tations of the primary pressure controller. The developed controller guarantees that only onepulley pressure setpoint at the time is raised above its lower constraint, and only to realizeFigure 10. CVT ratio response and tracking error, roller bench semi-kickdown.Hydraulically actuated CVT 401Figure 11. Primary and secondary pulley pressures, roller bench semi-kickdown.a desired ratio. This is visualized in figure 12. Higher clamping forces cause more losses inthe CVT 10, as long as no macro-slip occurs. The main causes are oil pump power demand(approximately linear with pressure) and losses in the belt itself, which both increase withincreasing clamping pressure, as supported by measurements 16. Hence, this controller hasa potential for improving the efficiency of a CVT, compared to non-model based controllers.Figure 12. New controllers pulley pressure setpoints minus lower constraints.402 M. Pesgens et al.Looking back to the lower plot of figure 10, the second (positive) peak (after the firstnegative peak due to actuator saturation) represents the overshoot of the ratio response due toa shifting direction change. This quantity describes the tracking performance of a controllerwell, and will be used to evaluate a controllers performance. The overshoot is computed hereas the (positive) maximum of the ratio error: max(rcvt,d rcvt). Also, the mean absolute error(1/N)summationtextN0|rcvt,d rcvt| (for the N data points in the 10 s response) will be used to compareresults.The same experiment has been performed for several variations on the controller. For eachof these variations, all constraints are still imposed, but some of the compensator terms in theratio controller have been temporarily switched off (the vertical arrows in figure 8). The resultshave been compared with the results for the total controller and are depicted in figure 13. Thecases that will be addressed are:1. All feedforwards and compensators on (total).2. No setpoint feedforward (setp ff off), rcvt,d= 0 in equation (37).3. No critical (no belt slip) torque constraint compensation (T comp off), Ftorque= 0.4. No hydraulic constraints compensation (hydr comp off), F,hyd= 0.5. No torque transmission nor hydraulic constraints compensation (T,hydr comp off),Ftorque= 0, F,hyd= 0.It is immediately clear that of all alternatives, the total controller with all feedforwards andcompensators on (total) described in the previous paragraph performs best, implying that allcontroller terms have a positive contribution towards minimizing the tracking error. Switchingoff either the hydraulic constraints compensation terms (hydr comp off) or the torque trans-mission compensator (T comp off) does not severely degrade the tracking quality. However,switching both compensators off (T,hydr comp off) does introduce large tracking errors. Thisoccurs because the maximum operator of both constraints is taken to calculate the compen-sating action, and if one constraint compensator is zero, the output of the maximum operatorFigure 13. Overshoot and mean absolute error for several controller alternatives.Hydraulically actuated CVT 403will still be non-zero due to the second constraint. Both compensators switched off simulta-neously effectively introduce a dead zone in the controller output u, the result of which isobvious. The response with the setpoint feedforward switched off (setp ff off) increases theerrors due to increased phase lag of the resulting response. The obtained results of the totaldeveloped controller show better tracking behavior (overshoot and mean absolute error) andlower transient pulley pressures (only during ratio change, as the clamping strategy is equal)compared with results obtained with a previously adopted controller, as described in ref. 3.This could be an indication for the potential for improving the CVT efficiency of the newcontroller, as described before.Vehicle tests including tip shifting (featuring stepwise ratio setpoint changes) have been per-formed on a test track, see figure 14. The stepwise changes in the ratio setpoint are trajectoriesthat cannot be realized. Hence, the measured CVT ratio will always lag behind. Hence, thisexperiment demonstrates the robustness against actuator saturation, as the pressure of the pul-ley that controls the ratio will saturate. As the errors in the feedforward terms of the controllerwill increase, the feedback controller becomes increasingly important. Also the anti-windupmechanism of the ratio controller needs to prevent overshoot. Results of an experiment drivingat a cruise-controlled speed of 50 km/h are depicted in figures 15 and 16. A new gear ratiosetpoint is generated every 2 s.At the start of the up-shift ratio responses at t = 2.1 s and t = 4.2 s, an inverse response ispresent. As the shifting speeds are indeed very high in this experiment, because of the layoutof the hydraulic system, the secondary circuit needs to supply the primary circuit with oil. Asa result, the secondary pressure rises in advance to the primary pressure and causes an initialdownshift. Around t = 3 s and t = 5 s, the ratio initially rises approximately linear, causedby the limited pump flow as the oil pump runs at engine speed, which is low. Upshifting isfurther characterized by some overshoot, which is clearly visible at t = 14 s. As the primarypressure cannot drop sufficiently quick due to a limited primary valve flow-through area towardthe drain, upshifting continues and causes overshoot. The secondary pressure only saturatesbriefly due to the limited pump flow after each ratio setpoint change. Much less overshoot ispresent during a downshift, the speed of which is not limited by pump flow. Again the primarypressure peaks above its setpoint when the secondary pressure is increased rapidly, causedFigure 14. Experimental vehicle during tip-shifts at the test track.404 M. Pesgens et al.Figure 15. CVT ratio response and tracking error, road tip shifting.by limitations in the primary pressure controller. This phenomenon lowers the maximumdownshift speed and is visible as a slight bump in the ratio at t = 6.2 s and t = 8.2s.As the main goal of the presented experiments is to demonstrate a new ratio controllerconcept, during the experiments belt slip has been avoided using a proven clamping strategyas mentioned earlier. Also, an online model-based detection algorithm was used, verifyingthat |primes|1. Two methods to detect belt slip off-line from measurement data (without directFigure 16. Primary and secondary pulley pressures, road tip shifting.Hydraulically actuated CVT 405measurements of the belts running radius on the pulleys to calculate the so-called geometricratio) have been used after the experiments. First, it has been verified if the range of CVTratios geometrically possible is not exceeded (rLOW rcvt rOD). Secondly, the maximumshifting speed of the CVT is limited due to limited clamping forces and variator speed, seeequation (10). The coefficient of friction in the excessive (macro-) slip region of a push-beltdecreases with slip speed 8. This causes unstable dynamic behavior, and hence slip speedwill increase rapidly when the torque capacity of a V-belt is exceeded. As the ratio is calculatedfrom measured pulley speeds, excessively fast ratio changes (high values of rcvt) can indicatebelt slip. The results of each measurement have been scrutinized, the result of which did notshow any traces of belt slip effects.7. ConclusionsA new ratio controller for a metal push-belt CVT with a hydraulic belt clamping system hasbeen developed. On the basis of dynamic models of the variator and hydraulics, compensatorterms of system constraints, a setpoint feedforward and a linearizing feedback controller havebeen implemented. The feedb