IMECE Proceedings of IMECE2002 ASME International Mechanical Engineering Congress & Exposition November 17 22, 2002, New Orleans, Louisiana

Transcription

1 Proceeings of IMECE2002 ASME International Mechanical Engineering Congress & Exposition November 17 22, 2002, New Orleans, Louisiana IMECE CONTROL OF THE RUTGERS ANKLE REHABILITATION INTERFACE Jungwon Yoon, Jeha Ryu, Dynamics an Control Laboratory, Department of Mechatronics, Kwangju Institute of Science an Technology, Bukgu, Kwangju , Korea. Grigore Burea*, Rares Boian Human-Machine Interface Laboratory, CAIP Center, Rutgers The State University of New Jersey, Piscataway, NJ, 08854, USA. ABSTRACT This paper presents the control system evelopment of the Rutgers Ankle rehabilitation evice. The Rutgers Ankle is a haptic interface with a Stewart platform structure riven by six ouble-acting pneumatic actuators. Each cyliner input is controlle by one set of on/off solenoi valves. For overall system control, an interrupt hanler loop is esigne base on harware interrupts. For precise flow rate control of the solenoi valves, a PWM scheme is propose for a set of two on/off valves. The propose PWM logic compensates the ea ban of the valve response time. A istribution metho for the esire pressure ifferential of the two air chambers of an actuator is propose. The metho is base on the minimization of the ifference between the esire an measure pressure. During experiments, one actuator achieve a rise time of 90 ms an 0.3 mm resolution in an inepenent position control loop. The Rutgers Ankle has a 7 Hz mechanical banwith an allows stiffness control for haptic interaction uring rehabilitation exercises. The measure performance is believe to be sufficient for ankle rehabilitation purposes. 1. INTRODUCTION The Rutgers Ankle is a component of the Telerehabilitation System with Virtual Force Feeback project (Popescu et al., 1999) (See Fig.1). The Rutgers Ankle was evelope to a a new rehabilitation evice to the existing telerehabilitation system. It can move an supply forces an torques in six egrees of freeom (DOF) as require by ankle rehabilitation exercise scenarios. This haptic interface has been use successfully in orthopeic rehabilitation (Girone, et al., 1999), post-stroke rehabilitation (Deutsch, et al., 2001a), an rehabilitation of musculo-skeletal injuries (Deutsch et al., 2001b). Figure 1.The Rutgers Ankle Haptic Interface (Girone, et al., 1999). ASME. Reprinte by The evice is base on a Stewart platform architecture with ouble acting pneumatic actuators. The avantages of the pneumatic cyliners are high power-toweight ratio, ease of maintenance, cleanliness, an the ability to maintain high forces without overheating. These qualities are especially valuable since the Rutgers Ankle is esigne for the at-home-exercising rehabilitation. Noritsugu an Tanaka (1997) evelope a therapy robot *Corresponing author: burea@caip.rutgers.eu 1 Copyright 2002 by ASME

2 with two egrees of freeom. Their actuators were riven by Pulse Coe Moulation (PCM) igital control valve, which is operate as an equivalent proportional valve with 8 ON OFF valves for one actuator. Therefore, expansion to a multi-axis system for 6 DOF motion an force generation is expensive an bulky. Takaiwa an Noritsugu (1999) evelope a haptic interface base on a pneumatic parallel manipulator. Their single-acting actuators are riven by a large an expensive electropneumatic proportional valve. Ben-Dove an Salcuean (1995) presente a new pneumatic actuator with voice coil flapper valves for use in teleoperation. The force was controlle through the ifferential pressure of two single acting actuators. Use of two actuators for one axis control makes the pneumatic system expensive an too large for home use. On-off valves with multi-moular type esign (such as the Matrix valve) make a multi DOF system cheaper an more compact than the servo valves mentione above. Therefore, the Rutgers Ankle is actuate by high-spee on/off solenoi valves, which are riven by Pulse With Moulation (PWM) logic. However, fine control of on/off solenoi valves is ifficult because of the large valve response time an its iscrete on/off nature. These valve characteristics must be consiere in the control esign to assure fast an reliable system response. This paper escribes the pneumatic control system of the Rutgers Ankle rehabilitation evice. The following section gives an overview of the Rutgers Ankle haptic interface with emphasis on the valve configuration for each actuator. Section 3 presents the esign of the interrupt hanler loop for overall system control. Section 4 suggests a PWM approach for a set of two on/off valves. In section 5, the inner an outer pneumatic actuator control loops are presente. Section 6 escribes the impeance control implemente for use in rehabilitation exercises. Section 7 conclues this paper. 2. THE RUTGERS ANKLE HAPTIC INTERFACE Overall system escription The Rutgers Ankle control box is shown in Figure 2 (Bouzit et al., 2002). It outputs 12 air channels that rive the six ouble-acting actuators. Each of the 12 pressures is controlle by two solenoi pneumatic valves: one for intake an the other for exhaust. Seeking to maximize the haptic banwith, the interface MATRIX valves were specially chosen for their low response time (300 Hz) an high air flow (100 Nl/min). Sensors rea the pressure in each of the actuator compartments. These signals are amplifie an sent to an embee PC (Pentium-233) via an A/D I/O car. These pressure values serve as inputs to the inner pressure control loop. The embee PC outputs through the same A/D I/O boar a control signal for each of the 24 pneumatic valves. These control signals specify the uty cycle of the PWM scheme. Figure 2. Rutgers Ankle Controller Overview (Bouzit, et al., 2002). Rutgers University. Reprinte by Intake Pressure supply Lower Exhaust Intake Upper Double acting pneumatic actuator Atmospheric pressure Exhaust Upper platform Figure 3. Valve Configuration for One Actuator. Rutgers University. Reprinte by Valve structure Each on-off valve is two way such that two sets of valves can control the flow rate at each sie of a ouble acting pneumatic actuator. Thus the on-off valves sets can have the following four states: close-close, openclose, close-open, an open-open. As shown in Figure 3, Valve L is pressurize by opening the intake valve an by closing its exhaust valve. Similarly, Valve U is epressurize by closing its intake valve an by opening its exhaust valve. The close-close state is use for holing the pressures an the open-open state is not use. 2 Copyright 2002 by ASME

3 3. DESGIN OF THE INTERRUPT HANDLER LOOP Task scheuling The controller has two simultaneously execute loops: the main program an the timer interrupt hanler. A harware interrupt generate by one of the A/D I/O boars is use to insure exact timing. The interrupt hanler executes the time-critical tasks, such as sensor reaings, filtering an control. The main program oes the rest of the tasks, mainly interfacing with the host PC over a serial port. Although the serial communication is not time -critical from the control point of view, it nees to run at a suitable spee to upate the virtual environment frequently. The communication will stop whenever the timer issues a harware interrupt. To avoi stalling, it is important that the time the hanler nees to finish its tasks be as constant as possible. Having the work balance across interrupts reuces the risk of having interrupts raise while the previous instance is still working. The work balancing is one by efining a number of sub-tasks an assigning them evenly to interrupts. Consiering that the interrupt hanler has to execute N tasks each having an occurrence perio of T k we compute T as the least common multiple of T 1,, T N. To achieve an even sprea of the tasks we assign to each of them a starting interrupt (offset). These offsets are chosen to get the best istribution over the first T interrupts. Starting from the smaller perios to the larger ones, offsets are assigne to the tasks so that the maximum number of tasks in an interrupt is minimize. Table 1 shows the scheuling for pressure control of the Rutgers Ankle platform. The columns show the tasks while the rows show the interrupts. An X symbolizes the assignment of a task to an interrupt. The uty-cycle-upate task has a perio of 1; the rea-sensors task has a perio of 7; an the cylinercontrol-input task has a perio of 28. After scheuling, every interrupt executes at most two tasks. Interrupt Interval Determination The proper interrupt frequency nees to be selecte in orer to guarantee a proper functioning of the system control. The interrupt interval can be etermine base on the frequency of the on/off valves an the resolution of the PWM signal, as well as on the computational amount of the interrupt hanler. Among the tasks that nee to be execute in the interrupt hanler are: sensor reaing, ata filtering, pressure control an position control. For goo control performance, PWM frequency shoul take into account the solenoi valve characteristic response time an airflow. Base on the technical specifications, the solenoi valves can achieve a maximum on/off frequency of 300 Hz. A higher PWM frequency may overheat the valves reucing their lifetime ue to excessive wear. Even a PWM frequency of 300Hz can cause valve overheating. However, the pressure transient Table 1. Example of Work Scheuling. Rutgers University. Reprinte by INTR Duty Rea Cyliner control input upate sensors X X 2 X X 3 X X 4 X X 5 X X 6 X X 7 X X 8 X X 9 X 10 X 11 X 12 X 13 X 14 X 15 X X 16 X 17 X 18 X 19 X 20 X 21 X 22 X X 23 X 24 X 25 X 26 X 27 X 28 X 29 X X 30 X X 31 X X Figure 4. Pressure Transient Response With Respect To The PWM Frequency. Rutgers University. Reprinte by response graph in Figure 4 shows faster response for the higher PWM frequencies. After several experiments, 250 Hz was selecte as the best choice between better 3 Copyright 2002 by ASME

4 performance an longer valve lifetime. Finer resolution of the PWM will generate finer control input. But the interrupt intervals must be long enough for the hanler to finish executing the controlling coe. Otherwise, the controller will freeze. Consiering task scheuling an PC spee, a uty cycle resolution of 1/28 was selecte. Hence the interrupt interval is 4msec/28 or ms. 4. PWM LOGIC FOR A SET OF TWO ON/OFF VALVES The ea ban ue to the finite valve response time (Figure 5) in opening an closing valves prevents a 100% PWM uty cycle. van Varsevel an Bone (1997) applie a novel PWM scheme for position control by using two stanar 3-way solenoi valves. The propose Rutgers Ankle PWM scheme for a set of two two-way on/off solenoi valves is similar. This PWM scheme is esigne to remove the nonlinearities ue to the open an close response time of the Matrix valves.. Figure 6. Achievable Duty cycle. Rutgers University. Reprinte by Figure 5. Dea ban in a PWM Perio. Rutgers University. Reprinte by When using on/off solenoi valves to control the flow rate of the pneumatic actuator, the controller output must be resolve into the iniviual pulsing of the two valves. When control input u to the valve solenoi circuit is positive, the intake valve will be open an the exhaust valve will be close. When control input u is negative, the intake valve will be close an the exhaust valve will be open. Figure 6 shows the valve uty cycle (%) that can be achieve from a set of two on/off solenoi valves with ea bans. This figure shows that we cannot get the uty cycle below minimum possible cycle an above the maximum possible cycle. The minimum possible uty cycle Dea is efine as; Dea = ( Tv / TPWM )100% (1) where T v is the valve response time (roughly 1ms) an T PWM is the PWM perio (4 ms). It shoul be note that if the uty cycle of either the intake or the exhaust valve is below minimum Dea, intake or exhaust valves will not open. Assuming that the opening an closing time of the valve are the same, the same size of minimum possible uty cycles are locate at the valve opening an closing time as shown in Figure 5. a) b) Figure 7. Dea Ban Compensate PWM: a) pressure step response; b) uty cycle; Rutgers University. Reprinte by 4 Copyright 2002 by ASME

5 In orer to overcome the ea ban, the control input can be mappe into the uty cycle as follows; = 100 (2) max Dea = Dea + Sv u ( Dea ) / 100 ; u > 0 Sv = (3) ( Dea ) / 100 ; u < 0 where u is the esire control input, is the applie control uty cycle, is maximum applicable uty max cycle, an S v is the slope shown in Figure 6. Equation (3) compensates for the opening an closing ea ban shown in Figure 5. Namely, even though the range of control input u is 0 u 100, PWM uty cycle will be locate in the range of ea max. Figure 7(a) shows the close step response of the pressure with ea ban compensate an without compensate ea ban. Figure 7(b) shows uty cycle in each case. In the case of the ea ban compensate PWM, the system is more stable, especially after 150msec. 5. CONTROLLER DESIGN The Rutgers Ankle is controlle by an inepenent joint control strategy, which is compose of an inner control loop (pressure control) an outer control loop (position or force control loop). The Inner control loop (pressure control) The inner control loop ivies the esire force ifferential that comes from the outer control loop into the esire pressure ifferential across the upper an lower cyliners. The esire pressure ifferential is generate consiering the ratio of upper an lower cyliner areas an the minimization of pressure change from the current pressure. The esire force ifferential f can be generate by inepenently controlling the pressures at each sie of ouble acting cyliner as f = Al Pl Au Pu Au = Al ( Pl rpu ) = Al P ( r = ) (4) Al where Al an Au are lower an upper sie area of ouble acting cyliner, respectively, r is the area ratio, an P is the esire pressure ifferential. In orer to generate the esire pressure ifferential as rapily as possible, two errors of Pl an rpu shoul be minimize. el = Pl Plm (5) eu = rpu rpum (6) where e l, e u, Plm, an Pum are the lower an upper pressure errors, an the measure lower an upper pressures. In orer to minimize the pressure change from the current pressure, performance inex J is efine as 1/ J = P P = e + e (7) = + m l u 2 2 ( Pl Plm ) + ( rpu rpum ) 2 2 2Pl ( 2 P + 2Plm + 2rPum Pl + P r Pum + plm = ) + 2 PP um r (8) The partial ifferential of J with respect to P l is zero at local minimum point as J Pl = 4 Pl 2( P + Plm + rpum ) = 0 (9) The esire lower an upper pressures are then given by Plm + rpum P P l= (10) 1 Plm + rpum P P u= ( ) r 2 2 (11) In aition, Pl, Pu must satisfy the bounary conitions Pcritical Pl, Pu Ps (12) where P critial an Ps are the critical an supply pressure, respectively. If P l, Pu are less than P critical, the intake pressure will be choke, which will cause flow saturation an limit controllability (Ben-Dove an Salcuean, 1995). After the esire pressures are calculate by equation (10) an (11), a PI controller is applie for each esire pressures. u k e + k e, u k e + k e (13) l = pl l il l u = pu u iu u The control input etermine by equation (13) will change the PWM iscrete uty accoring to equation (3). Figure 8 shows the block iagram of the propose inner pressure control loop. The Outer control loop (position control) The position control of each actuator is one as follows; L = ( L c Lm ) (14) P = ( K + K s) L + G (15) p where L c an Lm are the commane an measure cyliner isplacements, K, K are proportional an p erivative gains, respectively, an G p is the gravity compensation term of one actuator for moving actuator an platform mass. p 5 Copyright 2002 by ASME

6 L c + Inner Control Loop Upper input Dea ban - + Pressure 1/2 1/r PI compensate A + - Dynamics u PWM r PD + P controller 1/ Moving Mass Dynamics L m G p Dea ban Pressure 1/ PI compensate A i - Dynamics PWM Lower input Figure 8. Joint Position Control of Each Actuator. Rutgers University. Reprinte by a) Figure 9 shows the experimental results of inepenent joint position control. Figure 9(a) show that outer loop position control is successfully implemente without overshoots an with fast response about the rise time of 90 ms. The rise time of this pneumatic actuator with pressure feeback an two sets of 2-way valves at each part of the ouble acting actuator is twice as fast as the pneumatic system controlle by the 3-way solenoi valves escribe in van Varrevel an Bone (1997). Figure 9 (b) an (c) show the response of the lower pressure an upper pressures, respectively. Each pressure is changing with minimum ifference change from current pressure. By using inepenent joint control, we can control the posture of the Rutgers Ankle interface. Experimentally, it was etermine that the pitch angle of the Rutgers ankle can follow a 7 Hz sinusoial input as shown in Figure 10. The other axes have similar response. This spee is thought to be sufficiently fast enough for rehabilitation exercise purposes. b) c) Figure 9. Position Control Results: a) isplacement; b) lower pressure; c) upper pressure; Rutgers University. Reprinte by Figure 10. Rutgers Ankle Banwith (pitch angle). Rutgers University. Reprinte by 6 Copyright 2002 by ASME

7 6. REHABILITATION CONTROL USING THE RUTGERS ANKLE This section presents a rehabilitation exercise scheme using the Rutgers Ankle interface. Our platform can be applie for creating exercises training the flexibility, strength an balance of the subjects. Strength exercises are similar to conventional weighttraining exercises. Patients move their feet as the evice applies resistive forces. Flexibility exercises involve improving the patients range of motion by performing repetitive movements near their current range limits with little or no opposing forces. Balance exercises may require the simultaneous use of two mechanical evices, one for each foot. The large variety of exercises will allow patients immeiate access to many ifferent forms of rehabilitation through a single system. In orer to generate these exercise types, a position base impeance control is applie as shown in Fig. 11. The ifference between the esire foot position X for rehabilitation exercises (Cartesian Space) an the value of X m generate by filtering the measure interaction forces an torques to satisfy equation (16) will become the esire inepenent joint position input after mapping by inverse kinematics (IK). F = KX m + BX& m (16) Force an position will be relate as F = K ( X X foot ) + B( X& X& foot ) (17) where F is a vector of the forces an torques on the Rutgers Ankle, a) b) Figure 12. Plantar Flexion an Dorsiflexion Exercise with the Position Impeance Control: a) X foot (pitch axis); b) F (pitch axis); Rutgers University. Reprinte by X I K Lc Inepenant joint acutator control PD Pressure control Rutgers Ankle L m X foot FK Virtual environment X m Gp L & m 1/(K+Bs) F Force Sensor F foot Operator X & foot Manipulator Jacobian J Figure 11. Impeance Control Block Diagram for Rehabilitation Exercises. Rutgers University. Reprinte by 7 Copyright 2002 by ASME

8 _ K = iag{ K K, K, K, K, K yaw } is a stiffness x, y z pitch roll matrix, B = iag { B, B, B, B B B } is a x y amping matrix, an the X is the measure foot position of a patient attache to the Rutgers Ankle. The will be use to manipulate the object in the virtual X foot environment. If X - is ieally equal to X, the X m esire impeance in equation (17) can be achieve. Figure 12 shows the strength exercise for ankle plantarflexion an orsiflexion base on the impeance control. In this experiment, X = 0 an the stiffness K pitch was 50 NM/ra. With this impeance control, we can manipulate the level of the strength exercise by changing the stiffness an amping accoring to the patient conition. z foot pitch, roll, yaw 7. CONCLUSION This paper presente the pneumatic control system evelopment for the Rutgers Ankle rehabilitation evice. A PWM scheme for a set of two 2-way solenoi valves can precisely control pressures with flow rate control. The inepenent position control achieve a rise time of 90 msec with mass of pneumatic system an platform. This response time is faster than previous research results reporte for 3-way solenoi valves. Base on position control of pneumatic system, position base impeance control was implemente in orer to generate exercise types accoring to the patient conition. Thus cheap an compact solenoi valves allowe a 6 DOF ankle rehabilitation evice to achieve precise response at about 7Hz banwith. Future research involves the implementation of task-level haptic effects necessary in the rehabilitation simulations foot case stuy, Presence, MIT Press, Vol. 10(4), pp , August Deutsch, J. E., J. Latonio, G. Burea, an R. Boian, Rehabilitation of Musculo-skeletal Injuries Using the Rutgers Ankle Haptic Interface: Three Case Reports, EuroHaptics Conference, pp.11-16, July Girone, M. J., G. Burea, an M. Bouzit, The Rutgers Ankle Orthopeic Rehabilitation Interface, Proc. of the ASME, Dynamic Systems an Control Division,DSC VOL.67, pp , Girone, M., G. Burea, an M. Bouzit, Ankle Rehabilitation System, U.S. patent, , Dec Matrix Corporation, Noritsugu, T. an T. Tanaka, Application of Rubber Artificial Muscle Manipulator as a Rehabilitation Robot, IEEE/ASME Transactions on Mechatronics, Vol. 2, No. 4, pp , December Popescu, V., G. Burea, M. Bouzit, M. Girone, an V. Hentz, PC-base Telerehabilitation System with Force Feeback, Proceeings of Meicine Meets Virtual Reality 7 Conference, IOS Press, Amsteram, pp , Takaiwa, M. an T. Noritsugu, Application of Pneumatic parallel manipulator as Haptic Human interface, IEEE/ASME International Conference on Avance Intelligent Mechatronics, pp , Van Varsevel, R. an G. Bone, Accurate Position Control of a Pneumatic Actuator Using On/Off Solenoi Valves, IEEE/ASME Transaction on Mechatronics, Vol. 2(3), pp , September ACKNOWLEDGEMENT Research reporte here was supporte by grants from the Brain Korea 21, the CAIP Center an the New Jersey Commission on Science an Technology. REFERENCES Bouzit, M., G. Popescu, G. Burea, an R. Boian, The Rutgers Master II-ND Force Feeback Glove, Proceeings of IEEE VR 2002 Haptics Symposium, Orlano FL, pp , March Ben-Dov, D. an S. E. Salcuean, A force-controlle pneumatic actuator, IEEE Trans. Robot. Automat., vol. 11, pp , Dec Deutsch, J., J. Latonio, G. Burea an R. Boian, Post- Stroke Rehabilitation with the Rutgers Ankle System - A 8 Copyright 2002 by ASME