Fusion Engineering and Design 84 (2009) 698–702

Authors' copy - Article presented at the 25th Symposium on Fusion Technology (SOFT-25), Sept. 2008, Rostock, Germany Contents lists available at ScienceDirect

Fusion Engineering and Design journal homepage: www.elsevier.com/locate/fusengdes

PREFIT project: Integration of man-in-the-loop and automation for manipulation of heavy loads and forces in ITER G. Dubus ∗ , O. David, Y. Measson, J.-P. Friconneau CEA, LIST, Interactive Robotics Unit, 18 route du Panorama, BP6, Fontenay-aux-Roses F-92265, France

a r t i c l e

i n f o

Article history: Available online 20 January 2009 Keywords: Heavy load Flexible manipulator Control scheme

a b s t r a c t Operating within the framework of the Euratom Fusion Training Scheme (EFTS), the PREFIT partnership prepares remote handling (RH) engineers for ITER with an integrated programme of training and research over 3 years. One of the research topics aims at improving effectiveness and safety of remote handling manipulation of heavy loads, a significant challenge associated with ITER’s maintenance scheme. In all areas needing scheduled maintenance work – in-vessel components, neutral beam injector, hot cells – introduction of man-in-the-loop technologies would provide very useful help in case of unexpected events. The focus of this work is the problem of vibrations experienced when handling heavy loads using flexible robotic structures. Vibrations are induced during start and stop movements, high accelerations or in the case of collision. It is therefore necessary to develop strategies to remove undesired effects of stimulating dynamics or critical trajectories. Analysis of previous works on flexible arms proves that model-based control of light structures gives good performance providing that dynamic terms take into account the compliance of the joints and the structural deformation.

1. Introduction and context of this study Within the framework of the Euratom Fusion Training Scheme (EFTS), the PREFIT partnership prepares remote handling (RH) engineers for ITER with an integrated programme of training and research over 3 years [1]. During this period six engineers are familiarised with the technological and operational key issues of RH for ITER. Throughout the programme each engineer conducts his/her own research project. These projects have been designed to support the ITER RH effort and provide useful input to anticipated problems without interferring with the current developments in that field. Very heavy loads, very poor access for operating and viewing, but also high dependency on accurate remote positioning are considered by fusion experts as some of the numerous key technological challenges to be taken up by the ITER RH team. Whether for in-vessel components replacement, neutral beam injector, hot cells or any other task of the planned maintenance of ITER, the ratio between payloads and the constrained size of the manipulators will lead to design rather flexible structures needing integration of high level compensation schemes to complete the tasks within the requirements.

The introduction of a man in the loop would provide very useful help in case of unexpected events, but it has to be optimized. Vibrations due to the structure high flexibility are probably the main identified problem during coarse positioning. Critical trajectories or stimulating dynamics have to be detected before their occurrence. The goal of this project is to develop strategies for manipulation of heavy loads in constrained environments in order to improve safety and positioning accuracy, in both automated procedures and man-in-the-loop tasks. Several leads are assessed, from the integration of structure dependant mechanical models of robots in control schemes to the development of algorithms to smooth trajectories and avoid stimulating dynamics. Load transfer issues are also studied. This paper presents industrial and academic state-of-the-art studies on the control of flexible arms respectively in Section 2 and 3. In Section 4 theoretical bases are given on the developments currently carried out at CEA LIST on the modelling of flexible manipulators. Section 5 describes the single joint flexible mock-up that has been built to validate our models. 2. Industrial state-of-the-art 2.1. COGEMA/AREVA La Hague hot-cells

∗ Corresponding author. Tel.: +33 1 46 54 89 66; fax: +33 1 46 54 89 80. E-mail address: [email protected] (G. Dubus).

Authors' copy © 2008 Authors. All rights reserved. doi:10.1016/j.fusengdes.2008.11.061

Since 2005 the telerobotic system MT200-TAO has been under evaluation in COGEMA-La Hague reprocessing plant hot-cells [2].

G. Dubus et al. / Fusion Engineering and Design 84 (2009) 698–702

699

Fig. 2. JET ICRH antenna.

Fig. 1. MT200 in La Hague reprocessing plant.

It is the result of several attempts to replace the conventional through-the-wall telescopic telemanipulator. The main goals of these developments were to allow automatization of some repetitive tasks, as well as to improve security for difficult tasks or enhance the ergonomics of the workstation. The slave arm is a 6-DOF (5 rotary joints and a telescopic mechanism) 4 m-long manipulator (see Fig. 1). Its payload capacity is up to 20 kg for 40 kg own weight (maximum torques of 1540 Nm and 385 Nm for the two main rotations). Its transmission technology is based on gears and screws for the upper part, and cables and chains for the lower joints which makes it excessively flexible. The software TAO 2000 (CEA LIST) performed the force controlled master–slave mode between the two architecturally different machines and brought new functions to the operator such as active balancing, accurate force surveillance, tool weight compensation, adjustable velocity and effort scaling ratios, virtual mechanism modes, and automatic pursuit of the gripper by telesurveillance camera. These features helped the operator optimizing his working conditions inside a working volume that was triple the size of the original. It is interesting to notice that play-back robotic modes were also available for some repetitive tasks that did not require force feedback. In spite of the quality features available thanks to TAO2000 no particular scheme was implemented to limit vibrations. 2.2. JET’s ICRH antenna installation JET is at the present time the only fusion device whose maintenance can be fully remotely performed [3]. Over the past 20 years a considerable experience has been built up by the JET RH team. In spite of the differences that will necessarily exist between the ITER and the JET RH environments, RH engineers can undoubtedly take advantage of this unique knowledge. From April to November 2007 the JET tokamak was in a phase of shutdown for enhancement and refurbishment. This shutdown was mainly devoted to the installation of a number of new systems. The most extensive of these systems was the

ITER-like Ion Cyclotron Resonance Heating (ICRH) antenna (see Fig. 2). This 300 kg antenna was transported into the vessel on a subframe connected to the 10 m-long boom via the ICRH Antenna End Effector. For the rough approach the tip was moved along smooth trajectories using pre-defined moves, or “teach files”, and a virtual rail, avoiding this way any stimulating dynamics (except in case of emergency stop at full speed). The final approach was done manually on the basis of the contact forces exerted on the antenna and measured thanks to a 6-axis force and torque sensor equipping the end effector. Once the antenna was fully engaged with the front face of the box assembly fixing bolts were fastened from ex-vessel. Considering the load transfer issue, disengaging the boom from the bolted antenna was a critical task since the fastening of the fixing bolts had induced constraints on the sub-frame and the boom. On the basis of the displayed forces the operators used the boom compliance to minimize this load distribution before disconnecting. It is been a very efficient way to avoid any undesirable vibration of the arm during the disengagement. 3. Academic state-of-the-art Over the last decade the control problems of flexible manipulators have been intensively studied due to the challenging demand of fast and precise robots in various industrial applications. In most of the literature the strategy is not only to control the motion of the rigid mode with reasonable accuracy, but also to control the vibrations of the highest modes to achieve high speed and precise tip positioning. A considerable amount of theoretical and experimental research has been carried out in this field from very different perspectives: input shaping techniques [4], feedforward compensation [5], frequency domain techniques [6], linear control [7], adaptive control [8], sliding mode control [9], time-delay methods [10], output redefinition methods [11]. 3.1. Modelling of flexible arms Most control algorithms rely on models of the robots’ dynamic behavior. In these model-based approaches, the key to successful control is the model accuracy. An arbitrarily selected model may exclude some significant parameters or on the contrary complicate the identification procedure by including useless parameters.

700

G. Dubus et al. / Fusion Engineering and Design 84 (2009) 698–702

But if modelling a rigid arm is not trivial [12], modelling a flexible system is even more difficult. Indeed whereas rigid robots’ dynamics are defined by ordinary second order differential equations, flexible robots’ dynamics are defined by nonlinear fourth order partial differential equations with corresponding dynamic boundary conditions [8]. To simplify the modelling task, flexible links have frequently been modeled analytically, using Euler-Bernoulli [7] or Timoshenko [13] beam models with uniform mass and stiffness distributions leading to originally infinite-dimensional models of the system. However it is almost impossible to practically design controllers based on such infinite models. A popular approach is then to obtain a finite dimensional dynamic system by truncation and employ this reduced order model to design the controller. The higher order modes are then treated as disturbances that must be rejected. An alternative to the analytical modal analysis is the finite element method [14] which possibly gives more accurate results and leads ideally to real-time controller implementations. Each approach has its own characteristics from the view of complexity and accuracy. Most of the work previously mentioned has been centered on the flexibility of the links rather than the joints which are considered to be rigid. Some developments consider on the contrary that the main contribution to overall flexibility in a manipulator comes from the compliance of its joints and transmission system [15]. 3.2. Force control In specific applications such as surgical robotics, wheelchair attachable robotic arms, or fire-rescue turnable ladders, control of the interaction between the manipulator and the environment is crucial to obtain a desired force or to avoid damages on the contact surfaces. Most works on force control are based on lumped parameter models [16] or distributed parameter models [17]. Designing a controller based on distributed parameters is quite hard since its dimension increases with the number of vibrational modes. In [16] a simpler force control scheme is developed assuming that the whole mass is concentrated at the tip. When a contact occurs between the arm and the environment, only its static deflection is considered. This work also details a switching mechanism, which eases the transition between position control and force control at the collision instant. Many dynamicists have investigated the effects of the impact itself for simply supported beams impacted in the center [13]. Ref. [14] investigated the dynamics induced by an impact at the tip of a flexible arm moved at its extremity by a joint that has friction and inertia. In this paper a finite-element approach is used because it simplifies the inclusion of initial conditions such as collisions, hub-applied torques, and frictional torques. 3.3. Master–slave systems Very few papers deal with flexible cooperative systems such as master–slave systems that are however widely used in dangerous areas such as nuclear power plants or outer space [18]. When the operator performs a positioning task with a flexible master–slave system, the remote operation becomes difficult owing to the elastic vibration of the slave arm. To avoid the excitation of this vibration a skilled operator reduces the speed immediately after starting moving the master arm. Ref. [19] proposes a control method which introduces a strain feedback to the traditional force reflecting bilateral control. The strain of the arm can easily be measured by a strain gauge. When the elastic deflection occurs, the strain feedback exerted on the master side generates a reaction force at the master motor and acts like a

brake. Then the excitation of the vibration is reduced and the operator instantly feels a slight reaction force therefore imitating what he would have done himself on a conventional bilateral controlled system. An attempt was made to use in parallel a local strain feedback on the slave side of the controller in order to autonomically suppress the vibration. The main drawback of this control scheme was the degradation of the manoeuverability since the slave arm was moved by the local strain feedback regardless of the operator’s wish. 4. Analytical modal analysis Let us now consider a fixed-free beam along the x axis (length L, uniform cross-section area S, constant moment of inertia I, density  and Young’s modulus E). If its free end is subjected to a point load P the beam deflects into a curve y(x). Assuming the beam undergoes small deflections in the linearly elastic region, the following equations can be used. Taken into account that y(0) = 0 and (0) ≈ (dy/dx)(0) = 0 the static deflection expression can be computed by integrating the bending moment twice: y(x) = −

PL EI



x3 x2 − 2 6L



(1)

When the force P is removed from the displaced beam, it returns to its original shape. However inertia causes the beam to vibrate. Internal and external dampings are neglected at this point. Applying the fundamental principles of dynamics on a piece of the beam of length dx, the differential equation for the vibrations is obtained:  · S ∂2 y(x, t) ∂4 y(x, t) · + =0 4 EI ∂x ∂t 2

(2)

Eq. (2) is best solved by the separation of variables technique, i.e. by disjoining the deflection in two parts, one depending on position and the other on time: y(x, t) = Y (x) · f (t)

(3)

From (2) we deduce two uncoupled differential equations that describe the movement of the beam: ∂4 Y − kn4 · Y = 0 ∂x4

(4)

∂2 f + ωn2 · f = 0 ∂t 2

(5)

with kn4 =

ωn2  S EI

(6)

4.0.1. Spatial equation The general solution of Eq. (4) can be written: Y (x) = a sin(kx) + b cos(kx) + c sinh(kx) + d cosh(kx)

(7)

The boundary conditions come from the supports of the beam: the fixed end must have zero displacement and zero slope due to the clamp, the free end cannot have any bending moment or shearing force. From (7) we can deduce a 4-equation linear system which admits solutions for a, b, c, d if its determinant is null:

  0 1 0 1     1 0 1 0    =0  − sin(kL) − cos(kL) sinh(kL) cosh(kL)    − cos(kL)

sin(kL)

cosh(kL)

sinh(kL)

(8)

G. Dubus et al. / Fusion Engineering and Design 84 (2009) 698–702

701

It leads to the frequency equation for the beam: cos(kL) cosh(kL) = −1

(9)

The solutions of this equation represent a discrete set of values ˛n = kn L. As a consequence only a discrete set of natural frequencies ωn is permitted:



ωn = ˛2n

E 1  L2



I S

(10)

For each of these natural frequencies a deflection shape Y (ωn , x) = Yn (x) is associated and called natural mode of vibration. The roots of (9) can be numerically computed: ˛1 = 1.875, ˛2 = 4.695, ˛3 = 7.85, . . . For this set of ˛n only, Eq. (7) can be determined from the boundary conditions up to a multiplicative constant. If we choose this constant so that Yn (0) = 0 and Yn (L) = 1, we get for the deflection expression: Yn (x) = − −

Fig. 3. Displacements for modes 1, 2, 3 (left axis) and 4, 5 (right axis).

sin(kn L) − sinh(kn L) (sin(kn x) − sinh(kn x)) 2 sin(kn L) sinh(kn L) cos(kn L) + cosh(kn L) (cos(kn x) − cosh(kn x)) 2 sin(kn L) sinh(kn L)

(11)

4.0.2. Time equation The general solution for Eq. (5) can be written: f (t) = An cos(ωn t) + Bn sin(ωn t)

(12)

where An and Bn respectively depend on the initial position and velocity. In this study, the beam starts at rest. Thus Bn = 0. We finally have the general undamped harmonic solution for the time differential equation: f (t) = An cos(ωn t)

(13)

4.0.3. Vibration expression For each frequency the characteristic vibration is: yn (x, t) = Yn (x) An cos(ωn t)

(14)

The total beam motion is complex. Each characteristic mode vibrates with a different size, shape, and frequency. The expression of the whole movement is: y(x, t) =

∞ 

yn (x, t) =

∞ 

n=0

Yn (x) An cos(ωn t)

(15)

n=0

An depends on the initial deflection y(x, 0), and using the orthogonality property of Yn (x) functions we get: An = −

4PL3 EI ˛4n

=−

4P EIL kn4

Fig. 4. Vibrations of modes 1 (left axis) and 2 (right axis).

driven by a motor through an Harmonic Drive-based speed reducer, a 3 m-long circular cross-section beam with a calibrated tip mass, and measuring devices. The joint capacity is about 1000 N m. The controller runs the real-time OS VxWorks at a sampling time of 2 ms. The measuring devices are an optical encoder to measure the joint position and a laser tracker Leica LTD800 to measure the tip position. The accuracy of the laser measuring system is 5.10−5 m with a data acquisition frequency around 500 Hz. The first experimental results from this mock-up were used in Ref. [20] to assess the reliance on a finite element analysis software simulating mechanical systems. A new test campaign is planned in late 2008 and will evaluate the capabilities of different control schemes based on the analytical model given in Section 4.

(16)

The static displacement can therefore be seen as the superposition of the different modes at t = 0 (see Fig. 3). It appears that the third natural mode has already a very little effect on the whole static displacement of the beam. Nevertheless the upper modes cannot be neglected in dynamic problems because they are the ones which potentially induce high frequency vibrations. Fig. 4 shows vibrations for the first two modes. Higher modes act similarly. 5. Experimental platform This mock-up shown in Fig. 5 gives the opportunity to highlight on a basic experimental device the encountered problems during the manipulation of heavy load with a light structure (vibrations, critical trajectory, etc.). It consists of three parts: the actuated joint

Fig. 5. Flexible mock-up at Fontenay-aux-Roses site.

702

G. Dubus et al. / Fusion Engineering and Design 84 (2009) 698–702

6. Conclusion In this paper we have presented both industrial and academic state-of-the-art studies on flexible arms modelling and control. If a significant number of theoretical and experimental works have been achieved on the topic no real control strategy of the arm internal dynamics has ever been implemented in industrial remote handling systems. Then we have put the theoretical basis of the work currently carried out at CEA LIST and presented our experimental device. In forthcoming publications we will make our model more elaborate by using explicit dynamic boundary conditions and introducing internal damping in our model. Then the next step will be to consider a multi-link arm. References [1] Rolfe, A.C., Friconneau, J.-P., Mattila, J., Timperi, A., Dubus, G., Shuff, R., et al. The Preparation of Remote Handling Engineers for ITER, Proc. of SOFT’08, 2008, Rostock. [2] Garrec, Ph. and Geffard, F. and Perrot, Y. and Piolain, G. and Freudenreich, A.G., Evaluation tests of the telerobotic system MT200-TAO in AREVA NC La Hague hot cells, Proc. of ENC’07, 2007, Brussels. [3] Rolfe, A.C., A perspective on fusion relevant remote handling techniques, Proc. of SOFT’06, 82, 2007, 1917–1923. [4] Khorrami, F. and Jain, S. and Tzes, A., Experimental results on adaptive nonlinear control and input preshaping for multi-link flexible manipulators, Automatica, 31, 1, 1995, 83–97, Pergamon Press, Inc., Tarrytown, NY, USA. [5] Siciliano, B. and Prasad, J.V.R. and Calise, A.J., Output feedback two-time scale control of mutilink flexible arms, ATJDS, 114, 1, 1992, 70–77. [6] Bayo, E. and Papadopoulos, Ph. and Stubbe, J. and Serna, M.A., Inverse dynamics and kinematics of multi-link elastic robots: an iterative frequency domain approach, IJRR, 8, 6, 49–62, 1989.

[7] Cannon, R.H. and Schmitz, E., Initial experiments on the endpoint control of a flexible one-link robot, IJRR, 3, 3, 62–75, 1984. [8] Ham, W., Lee, J.-J., Adaptive nonlinear control of one-link flexible arm, Proc. of IEEE IROS’93, 1, 3–7, 1993. [9] Moallem, M. and Khorasani, K. and Patel, R.V., An inverse dynamics sliding mode technique for flexible multi-link manipulators, Proc. of ACC’97, 1407–1411, 3, 1997. [10] Kang, M.S., Yang, B., Discrete time noncollocated control of flexible mechanical system using time delay, ATJDS, 116, 2, 216–222, 1994. [11] Yang, H., Krishnan, H., Ang, M.H., Tip-trajectory tracking control of single-link flexible robots via output redefinition, Proc. of IEEE ICRA’99, 2, 1102–1107, 1999. [12] Bidard, C., Libersa, C., Arhur, D., Measson, Y., Friconneau, J.-P., Palmer, J., Dynamic identification of the hydraulic MAESTRO manipulator: relevance for monitoring, Fusion Engineering and Design, 75–79, 559–564, 2004. [13] Timoshenko, S., Young, D. H., Weaver, W. Jr., Vibration problems in engineering - fourth edition, Wiley, 1974, New York. [14] Chapnik, B. V., Heppler, G. R., Aplevich, J. D., Modeling impact on a one-link flexible robotic arm, IJRA, 7, 4, 479–488, August 1991. [15] Khorrami, F., Analysis of manipulators with flexible joints and links, Proc. of IEEE ICSENG’89, 561–564, 1989. [16] Garcia, A., Feliu, V., Force control of a single-link flexible robot based on a collision detection mechanism, IET CTA, 147, 6, November 2000. [17] Endo, T. and Matsuno, F., Force control problem and exponential stability of constrained one-link flexible arm, SICE’05, 3, 2267–2272, 2005. [18] Mori, T., Morita, Y., Ukai, H., Kando, H., Maneuverability of flexible master-slave systems, Proc. of IEEE IECON’00, 1, 201–206, 2000. [19] Morita, Y., Tsukamoto, N., Asai, K., Ukai, H., Kando, H., Matsui, N., Assist control for positioning task by flexible master-slave system, Proc. of ICIT’03, 2, 790–795, 2003. [20] Gagarina-Sasia, T., David, O., Dubus, G., Gabellini, E., Nozais, F., Perrot, Y. et al. Remote handling dynamical modelling: assessment on a new approach to enhance positioning accuracy with heavy load manipulation, Fusion Engineering and Design, 83, 10–12, 1856–1860, 2008.