Design and evaluation of a serial elastic actuator for human assistance

Transcription

1 2013 XXIV International Conference on Information, Communication and Automation Technologies (ICAT) October 30 November 01, 2013, Sarajevo, Bosnia and Herzegovina Design and evaluation of a serial elastic actuator for human assistance Dipl.-Ing. Kurt Gerlach-Hahn, Christian Dahmen, Dr.-Ing. Berno J.E. Misgeld, Univ.-Prof. Dr.-Ing. Dr. med. Steffen Leonhardt RWTH Aachen University Philips Chair for Medical Information Technology (MedIT) Aachen, Germany Mail to Gerlach-Hahn@HIA.RWTH-Aachen.de Abstract in this article, the systematic development of a serial elastic actuator for human assistance is presented. First, requirements for the actuator are deduced from the specific application, as part of an assistance robots for partially paralyzed patients. A test bench was developed to evaluate the serial elastic actuator under various conditions and to produce reference sensor data. The device was tested in four different scenarios, ranging from stiff fixation of the output to interaction with the knee joint of an active or passive human test subject. A variable system inherent friction was identified and the potential of this (usually undesired) characteristic is discussed for the special case of human assistance. Keywords serial elastic actuator, humans assistance, exoskeleton I. INTRODUCTION As effect of better life conditions and medical care, people in industrial nations are in a state of demographic transition. Accordingly, the quantity of elderly people and the number of age-related diseases increase. One common age-related disease is stroke (apoplexia cerebri), which can cause a palsy of a single limb or even a palsy of half body (hemiplegia) [1]. To prevent patients from depending on a wheelchair and the related discomfort, it is necessary to restore the lost abilities. The MedIT-Exo is an active orthosis to restore locomotion abilities. This exoskeleton will replace the missing movement abilities of the patient and thus restore their ability to stand and walk. In this article, the design and evaluation of an elastically coupled electrical actuator for an assistance robot is presented. The robot is an exoskeleton for patients with disabilities in walking, called MedIT-Exo. Fig. 1 shows a photograph and a schematic of the prototype in the present pneumatically and the planned electrically actuated version. The system consists of three one-dof (degree of freedom) joints per leg. As an alternative to the pneumatic actuation for this joints, a serial elastic actuator (SEA) was designed. The motivation for the usage of SEA in exoskeleton is given in chapter II, while the requirements of the target patient group, hemiplegic patients, and their special demands for the actuation system are defined in chapter II. The realization of the mechanical design of the SEA follows in the next chapter. A mechanical test bench was used to evaluate our first SEA concept. The measurements were conducted with a passive mechanical load for the estimation of system parameters. Based on This tests, the SEA was tested in connection to a human knee joint to evaluate the system behavior with different loads. The first prototype of the MedIT-Exo was realized with pneumatic actuators. Due to the actuators inherent elasticity, patient safety was achieved, at the price of difficulties in energy supply, such as the need for big pressure tanks. To combine the system inherent safety with the mobility of electrical systems, a SEA was designed as an alternative actuation. Fig. 1 Left: photograph of the actual pneumatically actuated exoskeleton prototype as objective system for the designed SEA. Right: design drawing of exoskeleton with application point for SEA /13/$ IEEE

2 II. EXOSKELETON REQUIREMENTS The movement assistance of physically limited patients has special requirements. Half side palsy, a hemiplegia, can occur as consequence of a stroke. Effects range from little to no impairment up to complete loss of motor function, while at the same time the other side retains its normal functionality. Depending on the complexity and the intra-individual variability in patients, there are special requirements for the actuation system. On one side, the function of the inactive side has to be fully replaced. On the other hand, the movement of the functional side of the patient should not to be disturbed in its movements. Consequently, the requirements for the actuation on the disabled side are comparable with the requirements (position and torque) of healthy subjects which are described in [2]. TABLE I. shows the maximum values of angle and torque, which occur in normal gait of healthy subjects. The values are for movements in the sagittal plane, which correspond to the movements the exoskeleton can realize. Thereby the data represent the requirements for gait pattern which correlate to the requirements of healthy human. Consequently, the high torque for the movement of ankle occur from the dynamic component and is not essential for gait. For instance it is possible that patient with amputation of a foot and ankle can walk with passive artificial limb like healthy persons. Parallel to This requirements for direct actuation, it is necessary that the system s behavior is as unobtrusive as possible at the healthy side of the patient. Typically, this data is collected on healthy human and a natural gait pattern and speed. The effects of less actuation are tested indirectly in [3] by disturbing a healthy test person using a non-actuated exoskeleton as load. At similar work of the test person, the movement velocity decreases. Consequently, it is possible to underactuate the system to a certain extent, assuming a lower movement speed is tolerable. For example, [4] describe an assistance system with less power. This motor allows a permanent power of 90W and has a power density of 150W/kg. For short time, it is allowed to use higher power, as long as the average power per second is below the maximum of 90 W for permanent use, to avoid thermal damage. For the usage with overload, a monitoring of the median current or of temperature is needed. For the following tests in this article, the maximal current was restricted to protect the test person from high torque in interaction with the system. Thus, an additional monitoring for the motor was not necessary. The gearing was a CSD-2A from Harmonic Drive AG, Limburg/Lahn, Germany, with transmission ratio of 160. It is characterized by its high ratio, 123 Nm maximum allowed transmitted torque and an extremely small construction size (17 mm height). The corresponding technical data is provided in TABLE II. The combination of motor and gear box allows a maximum torque of 123 Nm, the limit for repeated peak torque of gear. This is partially below the maximal specification for the ankle joint in natural gait, but sufficient for assistance of the hip and knee joint. As elastic elements springs with 2.5 and 5.5 windings were used, with a wire diameter of 11mm and material (spring steel). Fig. 2 shows photograph of the SEA set-up. TABLE I. REQUIRED ANGLES AND TORQUES WITHIN A GAIT CYCLE [2] Max. angle [ ] Max. torque Joint Positive Negative [Nm] Ankle (140) knee hip III. FIRST EXPERIMENTAL SET-UP OF A SERIAL ELASTIC ACTUATOR The actuation for the exoskeleton is intended to be realized by a serial elastic actuator (SEA) which involves an exchangeable spring for testing different spring characteristics. According to the idea of an electrically actuated SEA, an electrical motor EC 90 flat from Maxon Motor GmbH, München, Germany was used. The corresponding technical data is provided in TABLE II. Fig. 2 Experimental SEA system with motor, mounting plate and gear on the left side and an adaptive clutch with the 5.5 winding spring TABLE II. MOTOR CHARACTERISTICS Brushless Motor EC90flat, Maxon Motor GmbH, München, Germany Characteristic Value Unit Torque constant 70.5 mnm/a Nominal torque 425 mnm Stall torque 4840 mnm Nominal speed /min No load speed /min Weight of motor 0.6 Kg

3 TABLE III. CHARACTERISTICS OF HARMONIC DRIVE CSD-2A , Harmonic Drive AG, Limburg/Lahn, Germany Characteristic Value Unit Ratio Limit for repeated peak torque 123 Nm Limit for average torque 75 Nm Weight 0.24 kg Height 17 mm IV. TEST BENCH Our test bench for the serial elastic actuator is a modification of the system described in [5]. It consists of a position sensor and a torque sensor, while the load is a variable mass at a definable lever arm length. By variation of mass and lever arm length the desired characteristics can be defined. Fig. 3 displays the schematic of the test bench including its components: The actuation system includes the entire system under test. The mountings of the bench were each realized by two mountings of type GG.ASE-05 from Schaeffler Technologies AG & Co. KG, Schweinfurt, Germany. Fig. 4 shows the exact placement of the mountings. As reference for the actuation behavior, a torque sensor DR from Lorenz Messtechnik GmbH, Alfdorf, Germany was used. Its measurement range of 200Nm guarantees reference value also for torque peaks which exceed the motor characteristics. The velocity sensor is a contactless sensor type MHAD 50-HDmag from Baumer GmbH, Friedberg, Germany which include an absolute angular encoder and a velocity sensor. The mass was between 0.5 and 60 kg at a variable lever length of maximally 0.5 meters. The actual characteristics of torque and inertia were defined by combination of mass, lever-length and angle. The human coupling was an additional link to include a human in the tests. For the measurements in this article, an orthosis was connected for coupling a human knee to the test bench. Fig. 3 Functional scheme of test bench Fig. 4 Test bench with SEA on the left side and coupled human on the right side Fig. 5 Schematic diagram of the SEA (red), test bench (green) and human (blue) The complete system, combining the three components SEA, test bench and human, was modeled, see Fig. 5. The red area describes the SEA with the components motor (E-1), gear and spring. The inherent friction of motor and gear is combined to clarify the function of the model. The green area is the test bench. J 1 and J 2 denote the axis with the friction in the mountings modeled in parallel. The spring damper element in middle models the torque sensor. For the general behavior of the system the spring characteristic of the torque sensor exists, but can neglected in comparison to the elasticity of rest of system, especially the included spring in the actuator. For this paper it is assumed to be stiff. The blue area displays the coupling to a human. Dependent on the measurement scenario it is a fixation to ground, the passive characteristics or the muscular activities of a human, combined with the elasticity of the coupling. The parameters of the components can theoretically be determined component wise. In practice, the components change the characteristics extremely while mounting, due to interaction in the over determined system. Generally dominating characteristic of the system, displayed in Fig. 5, is the elasticity k of the spring in the SEA, the mechanical damping d of the mounted system, the inertia J see (1) and the torque constant n of the motor. Hence the system performance can be approximated to (2), with τ E torque at the sensor:

4 and (1) The time constant of the electrical system in motor and the spring dynamics of the torque sensor are much higher than the time constant of included spring and are neglected during the further procedure. Even though the parameter of the system cannot be determined exactly, it is important to understand the system behavior by analyzing the points where friction occurs. (2) V. MEASUREMENT For the system characterization, four tests were conducted. At first, the spring constants were determined. The second and third measurement deal with different signal attributes as desired values with a passive human as load. Fourthly, the system was tested with an active test person connected to it. All measurements were conducted using an E-Darc motor control unit from Eckelmann AG, Wiesbaden, Germany with the controller architecture which is represented in Fig. 6. The cascaded control loops for position, speed and motor current can be used independently. Thus every parameter in the controller can be used as input. For the following measurements the current control mode is used with parameters P=30 and I=30. Fig. 7 Spring constant measurement of 2.5 winding spring Fig. 8 Spring constant measurement of 5.5 winding spring Fig. 6 Structure of cascaded control loop in the E-Darc controller box (left box). τ H is the disturbance torque form the human activity. ω E is the speed of the lever/human leg A. Experiment I In the first experiment, we tried to determine the spring constants. The motor current was trapezoidal, while the lever was fixed. The torsion of the spring, between the motor gear and the lever, was measured and compared to the data of the torque sensor. Fig. 7 and Fig. 8 display the measurements and calculated spring constants for both springs. The spring constants were identified in opening and closing direction of the spring coil to estimate potentially differences. The spring constants of both springs were determined in steady state and are presented in TABLE IV. The difference in spring characteristic for up- and downward movements occurred, due to the inherent inhomogeneity of the coil structure. The margin of the spring constant in Fig. 7 and Fig. 8 arise from measurement inaccuracy during movement, originating from friction between spring and torque sensor. Spring 1: 5.5 winding Spring 2: 2.5 winding TABLE IV. CHARACTERISTICS OF THE USED SPRINGS Spring characteristic Property Value Unit spring constant (closing) 3.09 Nm/ spring constant (opening) 2.59 Nm/ direction margin % spring constant (closing) 4.77 Nm/ spring constant (opening) 4.26 Nm/ direction margin %

5 B. Experiment II The second experiment aimed at the interaction with human test subjects and sinusoidal motor currents. The test subject was instructed to be as passive as possible. This affects no active movements of human, but the muscular characteristics and reflexes remain, due to healthy test subjects. The current was specified with frequencies variation from two to five radiant per second at current of two ampere and amplitude variation from two to four amperes at frequency of fife radiant per second. The movement of the leg was monitored and is displayed in Fig. 9 and Fig. 10. based disturbance has more effect in amplitude than phase delay. Even though this activities of human were not intended in this experiment, the data show the disturbance just for short time. Also the volunteer did not report changes in comfort while this movements. This shows the usability of the designed SEA system as actuator for movements with a periodically repeated pattern according to the human gait. C. Experiment III This experiment was similar to the second. The entrance was 5 rad/s frequency and 3 A current. In contrast to measurement II the monitored characteristics was the torque in the system, see Fig. 11. Fig. 9 System movement at different current and 5 rad/s frequency Fig. 11 Torque measured in experiment III at a stimulus of 5 rad/s frequency and 3 A current The transmitted torque in this test with periodic movement shows a plateau every time it crosses zero torque, while the entrance of the system (the motor current and hence motor torque) were normal sinus signals. This angle range with low torque transmission allow small movements of the user of the exoskeleton with little power. This is an advantage of the use of the SEA as sensor for the movement intention of the user. Fig. 10 System movement at different frequencies and 2 A current amplitude The movement data in Fig. 9 imply a linear correlation between the current at input and the movement of the knee. Unexpected high amplitudes, e.g. around 2 sec in the red line or around 7 sec in the blue line, source in the activity of human while measurement. Fig. 10 show the movement at different frequency of the input current and indicates the phase delay of the system of SEA, test bench and human. The unexpected activity of human in the 5 rad/sec graph at 7 sec are the same than Fig. 9. In comparison of both diagram the data show that the human D. Experiment IV This test series was done with spring 1 (the more flexible one) in the SEA and a human volunteer acting as a load. The test person was instructed not to assist the machine movement and to operate against the movement with little force. Meanwhile, the desired current was repeatedly increased slowly by hand till movement occurred. The measured data of spring torsion and reference torque are shown in Fig. 12.

6 The torsion of the spring can be used as indication for the activity of the human; additionally the direction dependence of spring characteristic, its friction and backlash have no or even a positive effect. The heterogeneous in spring characteristic can be used to match the system performance to the diversity in human gait. In summary elasticity, friction and backlash affect a period where the user can produce small movements with little work, which can be used as control signal. Fig. 12 Spring torsion and related torque The influence of the test subject is the staged trend in spring torsion and transmitted torque. This is due to the interaction between human muscle behavior and the system s inherent friction. The data shows a linear correlation between the spring torsion and the transmitted torque. Sole exception is the change in current direction and hence in movement. This nonlinearity is due to the backlash in the connection of the spring to the system. VI. DISCUSSION The system allows a permanent torque of 75 Nm and average speed of ~ 20 rpm. With the allowed temporary use of overload, all requirements are satisfied. Only exception is the peak in angle torque, which was excluded from the essential requirements. The needed movement angle for the human gait should not be observed, due to the allowed full rotation abilities of the electrical motor. The complete height of motor and gear is ~35 mm at weight of less than 1 kg, including a mounting plate. The clutches of the system allows the use of a direct connection from gear to point of action as well as the usage of spiral coiled spring of different length. The friction in the combined system of SEA and test bench is clearly higher than expected from the sum of the estimated friction of the used components. This arises from little difference in the alignment of the used components to the rotation axis. In addition, the ground plate of the system is not absolutely inelastic. Consequently, the placement of components can change in dependence of the tension in the system and external load. In total, this effects prevent an elimination of the friction in system. In addition, the number of sensors in the system is not sufficient to determine the friction of all elements in Fig. 5. Furthermore, the elastic characteristics of spring and damping in the system limits the transmitted torque and therefore the movement acceleration of the subject in system, which enhances safety and comfort, especially in the case of motor failure. VII. CONCLUSION A test bench for actuator was adapted to the special requirements of serial elastic actuator (SEA) systems and used to evaluate a first SEA concept. Different kind of adaptive spring will be tested in combination with advanced control strategies. The parameters of the tested SEA concept are adequate to the identified requirements for human gait. Hereby, the spring provides mechanical decoupling and allow to store energy while gait. The high friction of the system can be a positive and not a drawback of the system, because it reduces high acceleration of human and related discomfort or injury. It is intended to analyze the effect of different reason of friction in reference to higher energy requirements versus safety. The spring torsion can be used as sensor signal for the desired movement of the human, with the effect that no other sensors for human activity, e.g. electromyography sensors (EMG), are needed as reference signal. In future work, the presented test bench should be developed further to allow determining the different sources and values of friction. Then, their individual effect on movement, decoupling and sensor function can be analyzed and a more energy-efficient system can be constructed. REFERENCES [1] A. Tennant, J. M. Geddes, J. Fear, M. Hillman, and M. A. Chamberlain, Outcome following stroke, Disability and rehabilitation, vol. 19, no. 7, pp , [2] A. M. Dollar and H. Herr, Lower Extremity Exoskeletons and Active Orthoses: Challenges and State-of-the-Art, IEEE Trans. Robot (IEEE Transactions on Robotics), vol. 24, no. 1, pp , [3] K. Gerlach-Hahn, S. Santos, B. Venema, The MedIT-Exo - Influence on Human Movement, Proceedings of the 10th International Student Conference on Electrical Engineering POSTER 2012, Prague (Czech Republic) 2012 [4] M. Grun, R. Muller, U. Konigorski, Model based control of series elastic actuators, The fourth IEEE RAS/EMBS International Conference on Biomedical Robotics and Biomechatronics, Roma, Italy, June 24-27, 2012, pp [5] K. Gerlach-Hahn, B. Misgeld, S. Leonhardt, Development of a testbench for bio-inspired actuator systems in rehabilitation robotics, Proceedings BMT 2012, 46. DGBMT Jahrestagung, Jena, September 2012