A pneumatic muscle actuator driven manipulator for nuclear waste retrieval

A pneumatic muscle actuator driven manipulator for nuclear waste retrieval

Control Engineering Practice 9 (2001) 23}36 A pneumatic muscle actuator driven manipulator for nuclear waste retrieval D.G. Caldwell *, N. Tsagaraki...

1MB Sizes 0 Downloads 32 Views

Control Engineering Practice 9 (2001) 23}36

A pneumatic muscle actuator driven manipulator for nuclear waste retrieval D.G. Caldwell *, N. Tsagarakis , G.A. Medrano-Cerda , J. Scho"eld, S. Brown Department of Electronic and Electrical Engineering, University of Salford, Manchester, M5 4WT, UK BNFL, Product Development Centre, Capenhurst, Cheshire, UK Received 18 October 1999; accepted 4 May 2000

Abstract This paper describes the development of a prototype tele-operated rig for the retrieval of radioactive material (spent Magnox fuel) from underwater storage ponds. The &robotic' mechanism developed for the task combines the established and accepted manual manipulation system with new pneumatic muscle actuators (pMA), that provide the controlled, #exible and robust energy input replacing direct human operation. This paper covers all aspects of the tele-operational system but focuses on the bene"ts of using pMA drives. Results show the capacity of the actuators to manoeuvre heavy loads accurately through large work volumes.  2001 Elsevier Science Ltd. All rights reserved. Keywords: Robotics; Actuators; Nuclear applications; Tele-operation

1. Introduction Sella"eld in North-West England is the main UK site for the reprocessing of used fuel from nuclear reactors. It is also the location of Calder Hall, the world's "rst industrial-scale nuclear power station, which became operational in 1956. This power station is still operational over 40 years later and authorisation has been granted for the reactors to operate for further 40 years. Fuel used in these "rst-generation nuclear reactors is natural, un-enriched, uranium metal encased in a can made of magnesium alloy, known as Magnox. After several years in the reactor, the fuel becomes less e$cient and has to be removed to be reprocessed and make way for fresh fuel. It is then stored in cooling ponds at the reactor sites, for a minimum of 100 days, before being transported to a reprocessing facility (Alzira & Salama, 1995). During the reprocessing operation, radioactive waste is separated from the unused original Uranium, and the Plutonium by-product. This Uranium and Plutonium can subsequently be recycled to make new fuel. The waste recovered is treated before being disposed of, or stored safely to await future disposal (Lee, 1997). * Corresponding author. Tel.: #44-0-161-295-4010; fax: #44-0161-295-5145. E-mail address: [email protected] (D.G. Caldwell).

Within Sella"eld one of the oldest facilities associated with the original nuclear power developments is designated B30 (Magnox storage ponds and decanning facility). This building was originally designed for decanning operations for spent Magnox reactor fuel. Storage of the spent fuel material during the operational period was in underwater ponds, which acted as an initial level of radiation screening. Over the course of operation, in addition to the spent fuel, these storage ponds have become "lled with debris from the decanning process and general miscellaneous debris collected over the years of operation. As part of an on-going clean-up and decontamination process all materials within the ponds are to be retrieved and moved to a new storage facility for grading, reprocessing, and long-term storage. As with many plants designed many years ago, there is little in the form of automation to assist with this clean-up process and the unstructured nature of this environment and the waste has meant that direct human intervention in the retrieval process has been necessary up to this point. Some automation of the process using robotic techniques has been considered, but this has generally been felt to be unacceptable because of the unpredictable nature of the environment both above and under water, the complexity of the systems, the cost, the relatively poor performance and the reliability.

0967-0661/01/$ - see front matter  2001 Elsevier Science Ltd. All rights reserved. PII: S 0 9 6 7 - 0 6 6 1 ( 0 0 ) 0 0 0 7 3 - 3

24

D.G. Caldwell et al. / Control Engineering Practice 9 (2001) 23}36

Although the exact dimensions on each storage pond are slightly di!erent, the basic structure is fairly uniform with a signi"cant proportion following a consistent pattern and layout. Typically, ponds are 6 m deep with a 5 m layer of water within the pond. The surface outline pro"le does vary but typically the &open' area is 2}4 m wide and 2}4 m long. A signi"cant number of the ponds are approximately 2;2 m and it is these ponds for which this system has been designed. Waste has accumulated in these storage ponds and in many places this may be up to 1.75 m high. In addition to the accumulated debris within the ponds, there is also an underwater track system used to transport the original fuel container between buildings. This system now forms the method of removal from the building to a new plant for reprocessing or future storage. Around each storage pond there are guardrails approximately 1.3 m high. Operators working over these guardrails manipulate 6.5 m long poles. The poles, which have a diameter of 41 mm, are made from domestic grade stainless steel and weigh 15 kg. At the remote end of the poles there are options on mechanical attachments including pneumatic tongs to grasp the debris and hydraulic clamps to cut, crush and size-reduce the waste, before it is loaded into a &skip' for underwater transport to the nearby reprocessing facility. At present, the operators are required to manoeuvre the pole, with attached tongs, to a position just above the fuel or other material, before grasping the target using the foot-operated pneumatic tongs. At this stage, the waste (which may weigh up to 20 kg in addition to the weight of the pole) is lifted and placed in the transport skip. Other debris on the #oor of the pond may require that the retrieved material must be lifted vertically up to 2 m to be clear of entanglement. To prevent jamming of the waste in the removal skip, sections must not be larger than 1 m. If the size is greater than 1 m the waste must be manipulated into the hydraulic clamp where it is reduced, before being located in the skip. During the movement of the fuel rods silt within the pond is disturbed reducing the visibility. Reductions in the visibility may eventually mean that work must stop until the silt has resettled. When lifting very heavy loads that need to be size-reduced, material is often dragged to the cutter, a process that is particularly prone to displacing silt and causing loss of visibility. To help improve the visual quality, a hot-water layer is maintained in the top 1 m of the pond. The retrieval and clean-up process requires that the workers wear respirators and full PVC suiting. For worker operation under these conditions there are a number of hazards including: (i) Radiation levels are relatively high. Operators working in full respirator suits are typically permitted to work 1 h/day to remain within permissible dosage  levels. A maximum radiation dosage for annual ex-

posure is set at 100 microsiverts/day. This leads to relatively low-e$ciency levels due to the short shifts that can be worked. (ii) The task involves heavy manual work which is very tiring. This is exacerbated by the need to wear protective clothing and the heat of the operation. (iii) The heavy nature of the work can lead to injury in the form of back strain. (iv) The heavy load and high inertia of the system mean that operational speeds are low. In addition, these loads mean that accuracy is also low, should it be needed. The whole process of the clean-up of B30 is part of a 20-year programme. This paper describes the development of a prototype tele-operated rig for the control of manipulation poles, permitting remote retrieval of radioactive material (spent Magnox fuel and other debris) from underwater storage ponds. In particular, the work will show how a &robotic' mechanism has been developed that combines the established and accepted manual manipulation system with new pneumatic muscle actuators (pMA), that provide the controlled, #exible and robust energy input needed to replace direct human muscle power. The performance requirements for the system are outlined in Section 2, while Section 3 describes the system in terms of the mechanical handling structure, the system controller and the user interface. The design process for the sizing of the actuators is also considered in detail here, as is the performance of the actuators in test experiments. The control strategy adopted and the system components are developed in Section 4, with results from experimental trials in dry test in Section 5. The conclusions covered in Section 6 also include a discussion of the potential of these actuators for other industrial sectors.

2. Project speci5cation Previously, robotic options had been explored for the automation of the retrieval tasks, but these have been dropped due to cost and low user acceptance. To overcome this user resistance and ensure optimum user acceptability and simplicity, the aim was to use a modi"ed version of the present pole-based system, but with the operators removed from the active site and manipulation by way of tele-operation. To ful"l this requirement, the initial design speci"cation called for a robust system consisting of a long pole similar to the manually operated devices currently used in the plant but driven and controlled by recently developed pMA. To enhance the safety of the system, the workers can now be removed from the immediate area of the pond, reducing the radiation risk and the

D.G. Caldwell et al. / Control Engineering Practice 9 (2001) 23}36

Fig. 1. Graphical overview of system layout.

&manipulator' will be tele-operated from a remote and shielded site. Speci"c system requirements included (Scho"eld, 1998) (see Fig. 1): (i) A tip work volume of 3 m;3 m. (ii) A payload capacity up to 30 kg (10 kg greater than for manual operation) and a pole mass of 40 kg (25 kg greater than for manual operation). (iii) Position resolution of 3 mm at the tip of the 6.5 m long pole. (iv) A cycle time for transfer (without return to pick-up) of 10 s. (v) Stable operation in a non-aqueous test environment. (vi) Vertical lift capacity was not to be explored at this stage. It was expected that vertical lift would be obtained using a traditional pneumatic cylinder.

3. System description The overall project was divided into three major sections as shown in Fig. 2. These are: (i) The user interface comprising the input device (joystick), the video feedback display and the computer display; (ii) The remote manipulation system comprising the mechanical structure of the rig, the actuators, the sensors and the video camera; and (iii) The control system comprising the control PC with associated ADC/DAC systems. 3.1. User interface The initial design of the system called for a user interface that resembled in some scaled form the present pole

25

operation. It was felt from previous, in company, experience that this would aid operator acceptance, by forming a layout that could be natural in its application. To achieve this a conventional two axes joystick was modi"ed and inverted. The outputs from this joystick were linked directly to the control PC through an ADC card. Since it was required that the tip of the main pole should be moved and controlled in steps as small as 3 mm, this constrained the input design, as the operator motions were a scale of this. For design purposes it was assumed that the user could not reasonabaly (without hand and arm support) control the input over a prolonged period with greater accuracy than 0.1 mm (30 times smaller than speci"ed for the tip control). This means that the joystick lever arm should be at least 1/30 of the pole length. A lever arm length of 30 cm was selected. The direct view of the motion of the main pole was through a conventional 53 cm monitor positioned to ensure intuitive viewing and operation. In addition to the direct visual feedback display on the TV monitor, control data was displayed on the computer monitor. This screen (Fig. 3), shows data on the current status of the system. The central window shows the current position of the tip of the pole in the pMA (actuator) reference frame. North}South}East}West (N}S}E}W) forms a reference location, aligned with the structure on which the actuators are mounted. Details of the actuator mounting are considered later in the paper. The central square shows the limits of motion within the work volume. Below this, central position display window are the demand pressures in each of the four actuating muscles. To the right of this display the actual internal pressures are recorded both as a bar graph and as actual pressures (in bars). The window on the upper right quadrant displays the controller setting for each of the antagonistic actuator pairs. The lower left quadrant displays, in a numerical format, the present pole tip position, and the present input position. The input method is selected for either joystick or keyboard mode. Joystick input is as described above while keyboard inputs are in pre-de"ned steps. Finally, the upper left quadrant shows in graphical form the error between the present input demand and the current pole tip location. 3.2. Pole manipulation system 3.2.1. Mechanical structure A schematic representation of the mechanical structure design is shown in Fig. 4, while Fig. 5 shows the actual system structure. The mechanical superstructure of the system is basically pyramidal in shape with a spherical bearing at the peak, X, designed to support the pole and provide free motion through the work area. As there are many tanks requiring clean-up it was initially envisaged that this

26

D.G. Caldwell et al. / Control Engineering Practice 9 (2001) 23}36

Fig. 2. Hardware diagram of the system.

guardrails at location AP and DM with attachment to the pole at Y. These sensors are used to measure the position of the pole and form a closed-loop feedback path. This control path will be considered later in this paper. Two antagonistic pairs of pMA are mounted between the superstructure and the pole (AY/CY and BY/DY) and form the motive force for the manipulator pole.

Fig. 3. Computer display.

structure should be portable, however, in view of later information that the projected period of use would be several years this is no longer a primary objective and a permanent or semi-permanent structure is now also acceptable. In this prototype, the support superstructure, (Figs. 1 and 5) to which the actuators will be attached, is constructed from 50 mm stainless-steel sca!olding rods and brackets. The &manipulation' pole is 6.5 m in length and is of a higher loading than for manual operation, with a weight of 40 kg. The system superstructure was mounted over a mock tank and around the guardrails ABCD. Draw-wire position sensors are mounted on the

3.2.2. Actuation system The actuators used for this system are a newly developed range of devices termed pMA, constructed as a #exible two-layered cylinder. This design has an inner rubber liner, an outer containment layer of braided nylon and endcaps that seal the open ends of the muscle. The detailed construction and operation can be found in Caldwell, Medrano-Cerda and Goodwin, (1994, 1995). The structure of the muscles gives the actuator a number of desirable characteristics (Caldwell et al., 1994, 1995): (i) This muscle can be made in a range of lengths and diameters with increases in sizes producing increased contractile force. Operating pressures can be up to 800 kPa (8 bar), but for this operation only 600 kPa could be guaranteed in the plant on a continuous basis, and the system was therefore designed for this supply limit. (ii) Actuators have exceptionally high power and force to weight/volume ratios. The contractile force for a given cross-sectional area of actuator can be over 300 N/cm for the pMA.

D.G. Caldwell et al. / Control Engineering Practice 9 (2001) 23}36

27

Fig. 4. (a) Schematic of the rig construction, (b) Muscle/sensor arrangements.

(vi) The actuators can operate safely in aquatic or other liquid environments and are safe in explosive/gaseous states.

Fig. 5. Manipulator rig.

(iii) The actual achievable displacement (contraction) is dependent on the construction and loading, but is typically 30}35% of the dilated length. (iv) Being pneumatic in nature the muscles are highly #exible, soft on contact and have excellent safety potential. (v) Controllers developed for the muscle systems have shown them to be controllable to an accuracy of better than 1% of displacement. Bandwidths for antagonistic pairs of muscles of up to 5 Hz can be achieved, although this is dependent on the internal volume of the actuator and the air-supply #ow rate. For this particular application, bandwidths as low as 0.05 Hz were considered to be acceptable (Caldwell et al., 1995). Force control using antagonistic pairs of muscle is also possible with an accuracy better than 0.1 N.

The workspace of the tip of the pole is a square 3 m;3 m, with the muscles mounted at a distance from the pivot of 15 m, which gives a muscle workspace requirement for an 6.5 m pole of 0.7 m;0.7 m. From the dimensions of the structure this requires muscles which are capable of a motion range of $0.35 m. In addition, in the corners, the range must be greater to allow for the diagonal motion. At the same time, the muscles must be able to support all the masses and inertias of the pole and waste fuel load over the full work volume (Scho"eld, 1998). To perform this motion, pMAs were constructed with an overall length of 3.5 m and an in#ated diameter of 70 mm. 3.3. Actuation system experimental model From the initial stages of the development of this system it was obvious that the actuation and its modelling would be a key factor for the overall system performance (Caldwell et al., 1994; Tsagarakis & Caldwell, 2000). As no pMAs of this size had ever been constructed before and smaller version had never been tested with such high loading and accuracy requirements a new series of performance models was required. A number of experiments were conducted to identify the basic parameters of a large system actuator and accurately model its performance for future design requirements. These included force capacity, force/displacement relationship, bandwidth, etc:

28

D.G. Caldwell et al. / Control Engineering Practice 9 (2001) 23}36

Fig. 6. Rubber elasticity experiment. Fig. 8. Tension/pressure, displacement experiment.

Fig. 7. Force developed due to the rubber sti!ness.

(i) The "rst experiment was to estimate the actuator's inner rubber axial elasticity which in e!ect forms the spring constant of the inner liner. One endcap of the pMA was mounted at a rigid aluminium plate, while the other end was left free to move, Fig. 6. Starting with the pMA fully extended but not tensioned the rubber liner was stretched and the increases in force were recorded by a strain gauge mounted on the aluminium plate. Fig. 7 shows the force versus displacement pro"le obtained from a muscle with an unstrained length of 1.6 m. The force developed due to the inner rubber gives a linear relationship for the 17 cm of extension rising rapidly after this point. During the initial stages of extension the outer braiding has no e!ect on the forces induced and a typical rubber spring response is observed. As the extension increases the braiding has an increasingly restrictive effect on the elongation and this manifests itself as a rapid increase in the force at extensions beyond 17 cm. From this experiment, the rubber spring constant has been calculated to be 120 N/m. (ii) The second series of experiments aimed to determine the force capacity and the force/displacement relationship with pressure for the actuator. To obtain this data, the pneumatic actuator was mounted in a rigid structure as shown in Fig. 8. A closed-loop control

scheme (considered in the control section later) regulated the pressure inside the pMA while the force developed was recorded by a strain gauge sensor mounted on one of the plates. The experiment was repeated for di!erent actuator lengths and in each case the force versus pressure was stored. As in the previous experiment, the actuator had an unstretched length of 1.6 m, a fully stretched length of 1.78 m and a fully in#ated diameter of 50 mm. Fig. 9 shows the force as a function of the input pressure for six di!erent actuator start-up lengths. The recorded length is given as percentage shortening relative to the fully stretched length of 1.78 m. It can be noted that the force/pressure pro"le shows good linearity across the pressure range from 0 to 500 kPa (0 to 5 bar). A small dead band can be observed for low pressures. This is due to the pressure required to overcome the radial rubber elasticity, i.e. to partially in#ate the rubber liner to make contact with the outer braid. The force/pressure and force/displacement pro"les show a fairly linear relationship between force and pressure/displacement, and this means that the pMA can reasonable be treated as a spring-type module. In fact, the pMA should be considered to be a variable sti!ness spring module with its sti!ness being proportional to the input pressure (Fig. 10). The above results suggest that the force exerted by the pMA can be experimentally modelled by the following equation: F"F #F , (1) E?Q [email protected]@CP F"K (¸!¸ )#K (¸!¸ ) if ¸'¸ , E?Q

 [email protected]@CP [email protected]@CP [email protected]@CP (2) F"K (¸!¸ ) if ¸(¸ , (3) E?Q

 [email protected]@CP K "K P, (4) E?Q N where K is the sti!ness due to the pressurized gas E?Q inside the pMA and is proportional to the input pressure, K the sti!ness per unit pressure, K the elasticity of N [email protected]@CP the inner rubber, ¸ the actuator length, ¸ the actuator

 length when is fully contracted and L , is the rubber [email protected]@CP length. F, F and F are the total contractile force, E?Q [email protected]@CP

D.G. Caldwell et al. / Control Engineering Practice 9 (2001) 23}36

29

Fig. 9. (a) Force as a function of input pressure for di!erent actuator lengths, (b) Force /displacement pro"le for di!erent input pressures.

Fig. 10. Spring model of the pneumatic muscle actuator. Fig. 11. Inner rubber diameter as a function of the input pressure.

the contractile force due to muscle pressurisation and the contractile force to rubber elasticity, respectively. At this stage, the model has not considered the lost pressure due to radial expansion of the rubber as observed in Fig. 9a. To take this rubber radial elasticity loss into account, a third experiment was performed to determine the pressure needed to expand the rubber up to the maximum actuator diameter. Fig. 11 shows the diameter of the rubber as a function of the input pressure. The "gures suggest that a pressure level P 0.3 bar is U required to expand the inner rubber to its maximum 50 mm diameter. Using this information, the gas sti!ness parameter K can now be modi"ed to include losses E?Q due to pressure expended to in#ate the rubber K "K (P!P ). (5) E?Q N U The sti!ness per unit pressure coe$cient K can be N experimentally obtained from the force/(pressure, displacement) pro"les using the following equation: F!F [email protected]@CP K " . N (P!P )(¸!¸ ) U



(6)

Here K was calculated to be 750 N/mbar with small N variation as the actuator length changes. (iii) The "nal experiment was the determination of the actuator bandwidth and how this information could be used in future design processes. The actuator bandwidth primarily depends on two factors; the actuator volume and the pneumatic supply #ow rate. For this experiment, the 1.78 m actuator tested previously was used, with air supplied through a 6 mm pipe. The input reservoir was held at a constant pressure of 800 kPa (8 bar). The actuator bandwidth was measured by applying as a series of sine wave inputs of di!erent frequency and recording the force response. Fig. 12 illustrates the actuator frequency response showing a force bandwidth of about 0.9 Hz. 3.4. Selection of actuator The "nal selection of the actuator to drive the system was dependent on the system model and the forces and displacements that had to be supported. From Fig. 4a, it can be seen that the force that is derived from the muscles

30

D.G. Caldwell et al. / Control Engineering Practice 9 (2001) 23}36

3.5. Visual feedback system For remote operation, observation of the pole manipulator was a requirement for e!ective operation. The visual feedback system closes this loop around the pole tip and the operator (Fig. 2). The visual system consists of a CCD camera initially located at the pole site on the superstructure at location D. On full wide angle operation this permitted a complete view of the workspace. Consideration given to the exact location of the camera suggested that it should be closely or completely align with the joystick inputs. However, a di$culty was noted in that the muscle DY was within the "eld of view of the camera and caused partial obscuring of the pole when in a mid location. Options investigated to remove this di$culty included: Fig. 12. pMA force frequency response.

is given by F cos h¸ , (7) + ? where L is the distance of the muscle attachment point ? from the pivot * in this instance 1.5 m and F is the + muscle force generated. This muscle action is opposed by the mass of the pole which is assumed to be a lumped mass located at the centre point of the pole and the mass to be moved located at the tip of the pole. The forces are given as ¸ M g¸ sin h#M g sin h N , * N N 2



The camera was "nally mounted 0.5 m to the right of the initial point, D, and at the same height as the muscles. At this point muscle collision is not a problem, the "eld of view still covers the whole work area and the misalignment is minimised. The visual signal from the pole site is transmitted to the operator site where it is displayed on a monitor as previously mentioned.

(8)

where M the mass of the load (up to 30 kg), g is the * gravity, ¸ is the length of the manipulation pole (6.5 m), . and M is the mass of the pole (40 kg). . h is the angle at the pivot point. At maximum displacement this is given as tan(d/¸ )+133, where d is the N displacement at the tip of the manipulator pole (Fig. 4a). Therefore, the force required by the muscle is



E Putting the camera below the muscle * this reduced the "eld of view and may potentially have been putting the camera in a site with a high collision risk. E Putting the camera at the corner of the cell * this caused some misalignment problems since the view and the joystick are not perfectly aligned and the motion is less intuitive.

¸ M F " N M # N g tan h+500 N. (9) + ¸ * 2 ? This must be achieved with a displacement at the muscle attachment point of 0.7 m (ratio of ¸ : ¸ ) for 3 m . ? tip motion. From Fig. 9, it can be seen that at an operational pressure of 5 bar the muscles can generate forces greater than 500 N for displacements less than 25% of the stretched initial state, i.e. 0.7 m must be (25% of the expanded muscle length giving a minimum value of muscle length of 2.8 m for each muscle. To allow for system losses due to friction, pressure loss, inertial e!ects and motion to the corners of the workspace muscles were constructed with a length of 3.5 m.

4. Control system 4.1. Control PC The computational requirements of the process are not rigorous and a Pentium 100 MHz was adequate for all control and display tasks. The PC was "tted with a dedicated ADC/DAC board (PC30) to provide data on the sensor inputs from the joystick, the draw-wire sensors on the rig, and the actuator pressure sensors. This control board also provided output drive signals for the valves controlling the actuator activation (Fig. 2). 4.2. Valve operations E!ective control of the air #ow into and from the pMA is the determining factor in the overall operation of the system. For this application four pressure regulating servo valves (Joucomatic Series 602) have been used. These valves are driven by a 150 Hz pulse width modulated signal with a mark/space ratio set by an external analogue voltage generated by the control PC.

D.G. Caldwell et al. / Control Engineering Practice 9 (2001) 23}36

31

ler settings and values with options available ranging from open-loop control to full PID, with di!erent options independently available for each antagonistic pair. The controller dialog box is shown in Fig. 14. The *CONTROLLER+ class implements the control scheme of the system, (Fig. 14) using as inputs the values from the sensors, the input device (joystick or keyboard) and the controller settings from the controller dialog box. The class *POLE+ provides the structure for storing all the pole data including the position of the pole tip and the pressure at the pneumatic muscles. The &POLE' class also communicates with the hardware of the system to execute the commands coming from the controller class while the *DISPLAY+ class, shown previously in Fig. 3, is responsible for displaying this information to the user.

Fig. 13. Class hierarchy.

4.4. Control description 4.3. Software description The software for the system was developed in C##. The class hierarchy and the interaction between the objects are shown below in Fig. 13. The generic class &SENSOR+ implements a sensor object. The classes &Pressure Sensor+ and *Position Sensor+ hold the readings from the pressure/position sensors on the actuators/pole and provide any calibration needed. The &DAC+ class provides access to the data acquisition board (PC30), while the *INPUT DEVICE+ provides access to the system input device (joystick). The *USER INTERFACE+ class implements the interface between the user and the system providing a wide range of options for initialising the system, tuning the position/pressure controllers, and calibrating the input device. The user options are displayed as a series of pull down menus. These menus allow the operator to calibrate the motions of the pole, and joystick. The operator can also set the control-

The control of each antagonistic pMA pair consists of two loops: an inner PID loop to control the pressure in the muscles and an outer PID loop to control the position of the pole tip in Cartesian space. The structure of the control scheme for each antagonistic pair is shown in Fig. 15. Within the initial control system the valves introduced unexpected control anomalies since there was no internal closed loop. These anomalies are due to a hysteresis of up to 20% in the valve design. This manifests itself in nonsymmetrical venting of the muscles. In practise, this means that muscle relaxation does not occur until the valve drive signal is reduced from 95 to 45% and this is therefore not equal to the increase in the antagonist muscle. The response pro"le for unsymmetrical action is shown in Fig. 16 where a rapid increase in pressure at a step input is not matched by the decrease, which in fact decays slowly over many seconds.

Fig. 14. Controller dialog box.

32

D.G. Caldwell et al. / Control Engineering Practice 9 (2001) 23}36

Fig. 15. Control diagram for each antagonistic pair.

Fig. 16. Open-loop valve "lling/venting response. Fig. 17. Fill/vent cycle with closed-loop pressure control.

To overcome this e!ect and compensate for the nonsymmetrical "lling and venting response times, the controller was modi"ed to include a muscle pressure controller. Solid-state pressure sensors have been placed at the pMA inlet down stream from the proportional valves in order to place an independent, high bandwidth, pressure loop around each valve. In this con"guration each valve now behaves as a pure air pressure source responding to pressure commands from the outer position loop and providing enhanced performance. Fig. 17 shows the response of the scheme to pressure commands. This implementation is contained in the dashed section in Fig. 15. The pressure commands for the above inner pressure control loop are provided by the outer position control loop using the error between the position command ( joystick), x and the pole position, x . The command B R

pressure for the muscles at each cycle are given by P P "  !DP,  2

(10)

P P "  #DP,  2

(11)

where



1 DP"K e# e#¹ &Q , N B ¹ G and e"x !x B R is the position error.

(12)

(13)

D.G. Caldwell et al. / Control Engineering Practice 9 (2001) 23}36

33

The position of the manipulator tip x is calculated R from the information provided by two draw wire sensors located on muscles 1(FY) and 4(EY) (Fig. 4b). In Fig. 4b, (A}D) are the muscle attachment points on the superstructure while (F,E) are sensor mounting points. b is the length of the draw wire sensor at start up which is constant, and $a is the range of displacement achievable by each pMA antagonistic pair. d and d are   the lengths of the draw wire sensors. These sensors are attached between the pole >(x, y) and the guard rail frame at the points F(!b, 0) and E(0, !b). The a;a square de"nes the range of motion of the pole connection point >(x, y) located below the spherical bearing. The following equations give the distances d and d as   a function of the co-ordinates (x, y) at the plane of the muscle attachments, point > in Fig. 4b: d "(x#b)#y, (14)  d "(y#b)#x. (15)  The above equations are used to determine the Cartesian position of the pole at the level of the muscle attachments for given distances d and d .   Subtracting Eq. (15) from Eq. (14) yields x"y#k,

(16)

where d !d . k"  2b

(17)

Substituting x into Eq. (6) "nally gives (y#k)#(y#b)"d . (18)  The y co-ordinate of the pole is determined from (18) using the condition !a(y(a. The x co-ordinate is given by x "y #k, (19)     where k is given by Eq. (17). The Cartesian position of the pole tip is calculated from the position of the pole at the plane of the muscle attachments using the following equations: ¸ ¸ x "x N , y "y N , (20) R R ¸ ¸ ? ? where ¸ is the pole length and ¸ is the length of the N ? lever (Fig. 4a). The desired pole position (x , y ) Eq. (13) is calculated B B from the joystick angles using the relatively simple kinematic relationship shown in Fig. 18. The homogeneous transform for the above manipulator is shown below: ¹"Rot(y, 0 )Rot(x, 0 )Trans(z,!l )   N

(21)

Fig. 18. Joystick/pole kinematics.

which gives



c

 0

¹"

!s 0



s )s   c  c )s   0

s

 !s

 c )c   0



!s ) c ) ¸   N s )¸  N . !c ) c ) ¸   N 1

(22)

Therefore, the desired pole parameters (x , y ) are given B B by x "!s c ¸ , B   N (23) y "s ¸ . B  N A low-pass "lter has been incorporated into the user input section of the system to prevent fast position transients from entering the position demand loop. Since the operator can move the input joystick many times faster than the bandwidth of the pole mechanism this could easily make the system unstable. A "nal feature of the design was in the air supply to the muscles. To provide a rapid response and rapid air #ow rate, two valve regulators were used. Due to the layout of the rig these were initially fed to muscles 1/4 and 2/3. Unfortunately, one of the regulators was further downstream and when initially energised, muscles 1/4 "lled before 2/3 causing the pole to induce an unpredictable swing away from the centre point at start up. This also meant that there could be uneven pressure drops in the antagonistic pairs introducing new control problems. To overcome this antagonistic pairs of muscles were supplied from the same regulator, which largely eliminates the problem (apart from slight #ow rate di!erences due to the length of the air line). These di$culties were minimised by ensuring that the line lengths were as close to equal as possible.

34

D.G. Caldwell et al. / Control Engineering Practice 9 (2001) 23}36

Fig. 19. (a) System response to "ne motions (motion step"3 mm), (b) muscle pressures.

Fig. 20. (a) System step response (1 m), (b) muscle pressures during the system response.

5. System testing and results To test the e!ectiveness of the system to execute the tasks for which it has been designed the following experiments were performed. (i) Basic actuator performance testing (work volume and loading): This simple test was required to verify that the designed pMA system could move the pole and a load (combined mass 70 kg) through a work volume measuring 3 m;3 m at the pole tip. The tests revealed that this could easily be achieved satisfying the power and motion range requirements. (ii) Fine motion testing: The initial design speci"cation required a pole tip resolution and "ne motion control of 3 mm. A series of "ne step motions (3 mm) were introduced to test the ability of the system to respond to small motions and also to locate the pole tip at 3 mm steps. The testing results are illustrated in Fig. 19. From these results it can be seen that the controller can successfully move the tip of the pole in 3 mm steps according to the initial design speci"cations of the system. A small delay was noticed when the direction of motion was reversed. This delay occurred only during the two "rst reversal steps, subsequently the system followed the desired step path without noticeable delay. This step delay on revers-

ing the direction of motion is caused by some wind-up in the system integrator. In later operator-based trials it was determined that a resolution around 10 cm would be adequate for this application. (iii) Stability/performance testing: An important speci"cation of the system was for stability across the operational environment to both large and small motion inputs. Although all initial tests have been conducted in a non-aqueous environment it was felt that this would form a more stringent test, as there would be none of the natural damping available in water and this would place higher constraints on the design. To test the stability and the quality of the system response a series of step inputs were introduced. Motions for a small, 3 mm step, have already be shown, while typical results for a 1 m step are shown in Fig. 20. Analysis of these results permitted tuning of the system's PID controllers. The "nal system controller values were set from these results based on a MATLAB-developed control scheme. (iv) Bandwidth testing: To test the bandwidth of the system a series of sinewave inputs of varying frequency were introduced from the control computer. For the speci"ed work volume (3 m;3 m) the bandwidth was

D.G. Caldwell et al. / Control Engineering Practice 9 (2001) 23}36

35

Fig. 21. (a) Response to a sinusodial input (frequency &0.05 Hz), (b) muscle pressures for a sinusoidal input.

Fig. 22. (a) Grid layout, (b) operator camera view.

found to be approximately 0.05 Hz (Fig. 21). When compared with the present operational period for human manipulation this period is found to be good and is better than the initial speci"cation parameters. (v) Operator performance testing: The previous series of tests were used to analyse the stability and performance for de"ned step inputs, but for operational use the systems required that the performance be assessed when used by the workers. To test the e!ectiveness of the system, a grid was prepared covering the work volume of the pole tip. Within the work volume a 15 kg test object was placed at a random location. The operator was asked to start from an arbitrary position, move to the load location and then try to locate the load at speci"ed points within the envelope following the sequence (A}D, A) (Fig. 22). Since the system had no gripper at the pole-tip, the operators had to continuously steer the load using the pole-tip, a task much harder than the end goal locatepick-place requirement, but designed to show accurate, rapid control and #exibility. It is obvious that this is an over-speci"ed and more di$cult task for the system as the "nal unit will be able to grasp the load and maintain a stable contact with it without achieving a controlled push.

In spite of the increased di$culty of the task, the operators reported that the ability to move and locate the load within the work-volume was good. A small lag between the operator motion and the system response was reported but fortunately this was small ((0.5 s) and the operators easily compensated for this lag and continuously steered the load while moving from one point to the other. In comparison, when the pole-load was directly manipulated by hand (from the muscle attachment location), it was found that the times and accuracy achieved were comparable, but this direct operation required two human operators.

6. Conclusions Nuclear clean-up, dismantling and decontamination operations are an inevitable legacy of work in the nuclear industry. These processes currently often involve the exposure of workers to raised radiation levels and there is a continual demand within the industry for automation to remove the workers from the harshest environments and reduce the dosage levels. This work has considered one such scheme, where the aim is to

36

D.G. Caldwell et al. / Control Engineering Practice 9 (2001) 23}36

automate for remote operation, currently a manually powered procedure for retrieval of spent fuel. A new form of actuator (pMA) has been developed for this operation and it has been shown that these pMAs, due to their #exibility and power can be incorporated into a rig, not unlike, the present manual installation. This close correspondence to the previous system has improved worker acceptance of the tool as an assistive device. This paper has shown how these pMAs can be used and suggested a possible new area of technology for some nuclear applications requiring power and #exibility, but human-like accuracy levels. This is particularly well suited to tele-operation were visual or other feedback can be used to close the control loop. Testing of the system has been demonstrated and proven to the satisfaction of the speci"cation and the site operators. The next stage will involve full grasp trials in an aqueous environment before "nal preparation for installation. Expanding the potential of this actuation system shows that it may provide a valuable drive mechanism for many industrial applications. At present, pMAs can outperform conventional pneumatic cylinder in all the areas that the cylinders have traditionally been promoted. The pMA has a higher power/weight ratio, more #exibility, lower weight, operates in aquatic environments, an easier implementation and set-up and lower cost. In addition, the conventionally cited drawbacks of pneumatics, i.e. poor control and compliance are not the di$culties they have been seen to have been, as witnessed by the controllability of the system described in this paper.

However, it should be noted that ultra-high accuracy will probably never be possible with these actuators. However, where accuracy comparable with human performance is adequate pMAs may form a totally new and improved option. In addition, in un-conventional areas, the controlled compliance features are being explored for systems that will provide variable sti!ness and relative safety in close contact with humans. Clearly, at this stage the options for industrial and non-industrial usage of pMAs are only starting to be recognised and explored but the potential could be enormous.

References Alzira, I., & Salama, A. (1995). British Nuclear Fuels Ltd. (BNFL) case study. Privatisation: Implications for corporate culture change (pp. 41}54). Avebury, England: Ashgate Publishing Ltd. Caldwell, D. G., Medrano-Cerda, G. A., & Goodwin, M. J. (1994). Characteristics and adaptive control of pneumatic muscle actuators for a robotic elbow. Proceedings of the 1994 IEEE international conference on robotics and automation (pp. 3558}3563), San Diego, CA, May 8}13. Caldwell, D. G., Medrano-Cerda, G. A., & Goodwin, M. J. (1995). Control of pneumatic muscle actuators. IEEE Control Systems Journal, 15(1), 40}48. Lee, J. (1997). UK Nuclear Fuel Reprocessing, http://gurukul.ucc. american.edu/TED/UKNUKE.HTM. Scho"eld, J. (1998). Retrieval system using air muscles} speci"cation. BNFL contract, April 1998, private communication. Tsagarakis, N., & Caldwell, D. G. (2000). Improved modelling and assessment of pneumatic muscle actuators. Proceedings of the 2000 IEEE international conference on robotics and automation conference, San Francisco, USA, April.