12th IFToMM World Congress, Besanon (France), June18-21, 2007

Workspace optimization of 3 RSS+CP parallel mechanisms

T.A.Hess-Coelho* University of So Paulo So Paulo, Brazil
Abstract The paper presents1 a novel architecture of a three-degree-of-freedom parallel mechanism, purposely conceived as a robotic manipulator for pick-and-place operations. The kinematic structure of the proposed mechanism contains one central passive limb that constrains the endeffector motion to two translations and one rotation. This work mainly focuses on the workspace optimization, stating the employed design variables, objective function and considered constraints as well. In addition, important issues such as position kinematics and singular configurations are also discussed in detail. Keywords: parallel robot, topology, workspace

F. Malvezzi Escola de Engenharia Mau So Caetano do Sul, Brazil

of their rotating axis, in order to constrain the endeffectors motion to only three translations. More recently, novel 3-dof architectures such as the Universal Cartesian Robot [6], Tripteron [7] and 3 PCC [8] present some convenient features: their set of nonlinear position equations become linear and fully decoupled which is not only valid for the inverse but also for the direct kinematics. This work introduces a three-degree-of-freedom parallel mechanism, purposely conceived as a robotic manipulator for pick-and-place operations. Initially, the mechanism topology is described not only by enumerating the employed joints in the active and passive limbs, but also by defining the input and output motions. Then, important issues such as the position kinematics and singular configurations are analysed in detail. Finally, the paper focuses on the workspace optimization, stating the employed design variables, objective function and considered constraints as well. II. Topology description There are many different approaches for topology generation of parallel kinematic structures. Herv [9] introduced the algebraic group structure of the displacement set, which is based on the mathematics of the motion group. The method of enumeration of active limbs was first proposed by Hunt [10] and expanded later by Tsai [5]. Other alternative methods are either based on the screw theory [11] or even on the addition of a passive limb [12]. The method of addition of a passive limb, which we apply in this paper, considers that the moving platform motion is constrained by a passive limb connected to it. In fact, the passive limb is carefully chosen in such a way that the number of degrees of freedom, F, and type of available motions for the end-effector correspond to the desired ones. Besides, partial connectivities of remaining m active limbs must be equal to , the space dimension where the mechanism is supposed to function. In our case, we select a passive limb composed by two joints: a cylindrical (C) and a prismatic (P) ones. As a consequence, the moving platform has three degrees of freedom and is constrained to perform two translations and one rotation.

I. Introduction Most of commercially available industrial robots are based on serial kinematic structures, i.e., their actuators and moving links are assembled serially, one after the other, resulting only one open-loop kinematic chain to position and orient a gripper or welding electrode. However, both academic and industrial communities have demonstrated a growing research interest on using another kind of kinematic structure, known as parallel, which is characterized by the presence of many independent limbs (kinematic chains), actuating in-parallel or simultaneously on end-effector. This nonconventional architecture becomes attractive due to some potential advantages over its traditional serial counterpart. Among them, one can mention: high rigidity, lightness, fast dynamic response, precision and high load capacity [1]. Different types of parallel architectures have been proposed to operate as robotic manipulators. The Neos Tricept [2] represents a tetrapod structure that contains one central passive limb to constrain the spatial endeffectors motion. Later, this architecture was applied as a milling machine-tool. Clavel [3] conceived the Delta robot, a 3-dof parallel mechanism based on spatial pantograph linkages. The H4 robot [4] has a similar architecture of Delta but employs an intermediate module to provide three translations and one rotation. Tsai [5] modified the Delta robot by replacing the spherical joints by using universal joints with special relative orientation
*E-mail: [email protected] E-mail: [email protected]

12th IFToMM World Congress, Besanon (France), June18-21, 2007

After describing the location of point P, we can also determine the center of each spherical joints Cj. For the sake of simplicity, 1 is chosen zero.
H c j + h H c j 0 C j 0 H s j H s j c 7 v s 7 = = 7T 0 H s j s 7 + v c 7 1 1 1 j = 1,2 ,3

(a)

(b)

(c)

Fig. 1. Skeleton kinematic diagram of the members of the family: (a) 3 UPS+CP; (b) 3 PUS+CP; (c) 3 RUS+CP

( (

) )

(3)

The connectivities of the other three active limbs must be equal to six. If we assume that each active limb has only two links and three joints, then the sum of degrees of freedom of the three joints is equal to six. By employing different joints in each limb, we can use prismatic (P), revolute (R), universal (U) and spherical (S) joints. Then, a family of parallel mechanisms (fig.1) is formed by the following architectures: 3 UPS + CP, 3 PUS + CP and 3 RUS + CP. An underlined letter is an active joint, which states the presence of an actuator. One can observe that all actuators are at or near the base. Constructively, all actuators can be rotary electrical motors even for members (a) and (b), because their output shafts might be coupled to ball-screws devices. Moreover, in order to simplify the assembly of the mechanism, the universal joints may be replaced by spherical ones. The following sections will deal with the position kinematics, singularities and workspace evaluation, developed for one member of the family: the 3 RSS + CP. III. Position Kinematics The objective of position kinematics is to determine the mathematical transformation between the coordinates x, y and z of a point P that belongs to the platform, and actuator displacements 1, 2, and 3. After obtaining such relations, it is possible to perform inverse and direct kinematics. A. Inverse Kinematics In this analysis, the moving platform coordinates, variables x, y and z of point P, are known, while the actuator coordinates, the angles j (j=1,2,3) are unknown. We begin by calculating the relation between the point P coordinates and the auxiliary variables h, v and 7, that correspond to the displacement along X0-axis, Z7-axis and rotation around X0-axis, respectively. The location of this point can be described by the homogeneous transformation [13] of frame 7 with respect to the fixed frame 0, 0T . 7
0 P 0 7 P = 7T 1 1

(a)

(b)

Fig.2 - Employed notation for position analysis: (a) the parallel mechanism; (b) the constraining passive limb.

(1)

(a)

(b)

where
0 7T

Fig.3 - Parameters in (a) the base platform and (b) the moving platform.

= Trans(X0,h) Rot(X0, 7) Trans(Z7,v)

(2)

The point Bj is the center of the other spherical joints in each limb j (fig.4)

12th IFToMM World Congress, Besanon (France), June18-21, 2007

0Bj 0 j +3 B j j = jT j + 3 T 1 1

j = 1,2 ,3

(4)

where
c j s j s 0 j c j jT = 0 0 0 0
c j 0 j T = j +3 s j 0

0 Lc j 0 Ls j , 1 0 0 1
0 j +3 Bj = 0 L1

mechanism itself may have up to eight assembly modes. These assembly modes are only theoretically possible and, consequently, due to constructive reasons, one among the others is preferrable. The chosen assembly mode represented in fig.2 is adequate because it avoids interference between each peripheral active limb with the central passive one. B. Direct Kinematics In this analysis, the angles j (j=1,2,3) are known while the unknowns correspond to the variables h, v and 7. After rearranging eq. (7), we obtain
h 2 + H 2 HL + L2 + v 2 HLc 2 j + HLs 2 j Hs j L1 + 2 Ls j L1 Hc 2 j s j L1 + Hs 2 j s j L1 + L1 2 + h( 2( H L )c j 2c j s j L1 )

0 s j 0 1 0 0 , 0 c j 0 0 0 1

Therefore,
L1 c j s j + Lc j 0 B j = L1 s j s j + Ls j L1 c j

( (

) )

j = 1,2 ,3

(5)

+ s7 ( 2 Lvs j 2 Hc j s j L1 + 2vs j s j L1 ) + c7 ( HL + HLc 2 j HLs 2 j 2vc j L1 + Hc 2 j s j L1 H ( 1 + s 2 j )s j L1 ) L2 2 = 0

The distance between points Bj and Cj is equal to L2. Therefore,

(j=1,2,3)

(9)

( B C ) ( B C ) L
0 0 T 0 0 j j j j

2 2

=0

j = 1,2 ,3

(6)

Rewriting eqs.(9) in such a way to consider only variable h as unknown,

h 2 + K j1 h + K j 2 = 0

(j=1,2,3)

(10)

where
K j 1 = 2( H L )c j 2c j s j L1
K j 2 = D j 3 + v 2 + s 7 ( K j 6 v + K j7 ) + c 7 ( K j 4 v + K j 5 )

D j 3 = H 2 HL + L2 + v 2 HLc 2 j + HLs 2 j Hs j L1 + 2 Ls j L1 Hc 2 j s j L1 + Hs 2 j s j L1 + L1 2 L2 2 K j 4 = 2 L 1 c j
Fig.4 - Sketch of limb j indicating used reference frames.

K j 5 = HL + HLc 2 j HLs 2 j + Hc 2 j s j L1 H ( 1 + s 2 j )s j L1

Then,
K j c j + M j s j + N j = 0 j = 1,2 ,3

(7)

where

K j 6 = 2 s j ( L + L1 s j ) K j7 = 2 HL1 s j c j

( ) M j = 2 L1 (c j ((L H ) c j h ) + s j ((L Hc 7 ) s j v s 7 ))
K j = 2 L1 Hs j s 7 vc 7
N j = L2 + (L H ) c j h 1

By applying the dyalitic elimination [14] in eqs. (10), for index j equals 1 and 2, we obtain
0 1 0 1 1 K 11 1 K 21 K 11 K 12 K 21 K 22 =0 K 12 0 K 22 0

)2 + ((L Hc7 ) s j v s7 )2 + ( Hs j s7 vc7 )2

Eq. (7) can be modified into a 2th-order polynomial equation

(11)

(N j K j ) u 2j + (2 M j ) u j + K j + N j = 0

j = 1,2 ,3

(8)

Solving the previous determinant, we obtain eq. (12) presenting only the unknowns v and 7.
( K 11 2 + K 11 K 21 K 22 )K 22 = K 12 2 + K 12 ( K 11 K 21 + K 21 2 2 K 22 )

where uj = tan (j /2). As one can notice, eq. (8) may have up to two different solutions and, as a consequence the

(12)

12th IFToMM World Congress, Besanon (France), June18-21, 2007

Similarly, applying the same technique for eq. (10) and considering only the values 1 and 3 for index j,
( K 11 2 + K 11 K 31 K 32 )K 32 = K 12 + K 12 ( K 11 K 31 + K 31 2 K 32 )
2 2

(13)

J q( 1,1 ) 0 0 Jq = 0 J q( 2 ,2 ) 0 0 0 J q( 3 ,3 ) J x( 1,1 ) J x( 1,2 ) J x( 1,3 ) J x = J x( 2 ,1 ) J x( 2 ,2 ) J x( 2 ,3 ) J x( 3 ,1 ) J x( 3 ,2 ) J x( 3 ,3 )

Eqs. (12) and (13) can be rewritten assuming that only the variable v is unkown. Hence,
e12 v 2 + e11 v + e10 = 0

(14) (15)

e 22 v 2 + e 21 v + e 20 = 0

where
e10 = D13 + 2 D13 D 23 D 23 D 23 K 11
2 2 2

( ) J x( j ,2 ) = 2v + 2(L + L1 s j )s j s 7 2 L1 c j c 7
J x( j ,1 ) = 2h + 2 H L s j L1 c j

2c 7 D13 K 15 + 2c 7 D 23 K 15 c 2 7 K 15 2 2 s 7 D13 K 17 + 2 s 7 D 23 K 17 2c 7 s 7 K 15 K 17 s 2 7 K 17 2 + D13 K 11 K 21 + D 23 K 11 K 21 + c 7 K 11 K 15 K 21 + s 7 K 11 K 17 K 21 D13 K 21 c 7 K 15 K 21 2 s 7 K 17 K 21 2 + 2c 7 D13 K 25 2c 7 D 23 K 25 c 7 K 11 K 25 + 2c 2 7 K 15 K 25

2 2

( HL + HLc
2

J x( j ,3 ) = 2 s j Lv HL1 c j + vL1 s j c 7
j

HLs j 2vc j L1 +
2

Hc 2 j s j L1 H ( 1 + s 2 j )s j L1 s 7

J q( j , j ) = HL1 c j + 1 + c 2 j s 2 j 2 LL1 c j + 2hL1 c j c j 2 HL1 s j s j + 2vL1 c j s j + 2vL1 s j c 7 Hc 2 j c j L1 c 7 + H ( 1 + s 2 j )c j L1 c 7

(j=1,2,3)

+ 2c 7 s 7 K 17 K 25 + c 7 K 11 K 21 K 25 c 2 7 K 25 2 + 2 s 7 D13 K 27 2 s 7 D 23 K 27 s 7 K 11 K 27
2

+ 2c 7 s 7 K 15 K 27 + 2 s 2 7 K 17 K 27 + s 7 K 11 K 21 K 27 2c 7 s 7 K 25 K 27 s 2 7 K 27
2

The other coefficients e12, e11, e22, e21 and e20 can be easily obtained. Repeating the dyalitic elimination technique for eqs. (14) and (15), this leads to eq. (16) that finally has one unknown, the angle 7.
e12 2 e 20 2 e11 e12 e 20 e 21 + e10 e12 e 21 2 + e11 2 e 20 e 22 2e10 e12 e 20 e 22 e10 e11 e 21 e 22 + e10 2 e 22 2 = 0

(16)

Developing the eq. (16), we obtained a 16-th order polynomial equation.

m =0 m G16 m u7 = 0 16

(a)

(17)

where u7 = tan (7 /2)

IV. Singular configurations In singular configurations, a parallel mechanism either reaches a locking position or even becomes uncontrolable. The conditions for occurrence of singularities can be investigated by inspection of the determinants of jacobian matrices, Jq and Jx [5] . Fig.5 shows two examples of theses configurations. & & (18) Jx x = Jq q where
& & & x = h v &7

& q = &1 &2 &3

(b)
Fig.5 Two examples of singular configurations.

12th IFToMM World Congress, Besanon (France), June18-21, 2007

V. Workspace The available workspace of the 3 RSS+CP represents a 3D-region where point P, that belongs to the moving platform, can move. To determine this workspace, the discretization method [15] is employed. This method considers that the workspace is determined from a solid, assumed larger than the feasible workspace, discretized by a regular mesh. Then, a procedure checks whether or not each mesh node violates the physical and kinematic constraints. Consequently, workspace boundaries are composed by a set of nodes that have at least one node neighbor that does not belong to the workspace. The physical constraints are represented by both the lengths of moving links and the strokes of passive joints. In addition, another important factor refers to the kinematic constraints. Our analysis calculates the determinants of jacobian matrices Jq and Jx and verifies if their values are null, which correspond to singular configurations. For the workspace optimization, the chosen design variables KL and KS correspond to the ratio between the lengths of moving links and the quotient of the sides of the moving platform and base triangles, respectively. KL = L2 / L1 KS = H / L (19) (20)

Fig. 6 Workspace optimization : distribution of the objective function in the feasible domain of design variables.

Spherical joint Strokes (deg.) 60 140

Initial Workspace [dm3] 0.9223 7.2523

Optimum Workspace [dm3] 1.2084 9.3932

Increase [%] 31 29.5

Table 2 - Comparison between the workspace volume before and after the optimization

The objective function f, which represents the workspace volume, is calculated by the sum of small volumes associated to each feasible node. Once there are only two design variables, the nummerical procedure comprises an exhaustive search for the maximum value of the objective function in its feasible domain. f (KL, KS) = V = vi
i =1 n

(21)

Table 1 presents the initial parameters of the parallel mechanism. The angles 1, 2, 3 remain constant even after optimization. Fig.6 shows the distribution of the objective function in the feasible domain of design variables. For 60-degree stroke of spherical joints, the maximum value of the objective function is 1.2084 dm3, while KS equals 0.4 and KL is 1.625.
H [mm] 85 L [mm] 119 L1 [mm] 90 L2 [mm] 210

The workspace volume for this mechanism is strongly influenced by the strokes of spherical joints, which can be observed in fig.7. Fig.8 presents the workspace shape after performing the optimization for 60-degree stroke of spherical joints. The volume of the available workspace is aproximately 9.4 dm3 for 140-degree of spherical joints. The ratio between the workspace and the projected area on the mechanism base is 1.5 dm . In a comparison with other parallel robots, by using this performance index as a valid criterium, Tripteron [7] and Delta robot [16] reach 0.9 dm and 2.5 dm, respectively.

1
[0 ] 0

2 [0 ]
120

3 [0 ]
240

Table 1 Parameters of parallel mechanism

Table 2 presents the values for the workspace volume before and after the optimization for 60-degree and 140degree strokes of the spherical joints. For both cases, notice an increase of aproximately 30 % of the workspace volume after performing the optimization.

Fig.7 Influence of the spherical joint strokes on the size of the workspace.

12th IFToMM World Congress, Besanon (France), June18-21, 2007

Fig.8 Workspace shape after performing the optimization for 60degree stroke of spherical joints.

VI. Conclusion The main contributions of this paper include the description of a novel 3-dof parallel mechanism, the development of its position kinematics, singularity analysis and the optimization of its available workspace. Regarding the optimization procedure, the chosen design variables are related to the size of the base with respect to the size of the moving platform, and the ratio between the lenghts of the connecting links. The adopted objective function considered the workspace volume, calculated by the discretization method, subjected to physical and kinematic constraints. After the optimization, the obtained results indicated an increase of 30 % of the workspace volume compared to the initial estimate. This work also discussed the influence of spherical joint strokes on the size of the available workspace. The feasibility of more than 60-degree stroke for spherical joints rely on the replacement of the conventional ones by, for instance, a sequence of three revolute joints, whose axes intersect at a single point. The proposed mechanism is a parallel version of the serial PRP, because it also provides a cylindrical workspace. However, differently from its serial counterpart, the actuated 3 RSS + CP will be able to reach higher speeds, in spite of a smaller workspace. Therefore, we foresee a promissing future for this parallel mechanism as a pick-and-place robot, suitable to execute the required tasks in pharmaceutical, electronics and food industries.

[4] Pierrot, F. and Company, O., 1998, H4: a new family of 4-dof parallel robots, LIRMM UMR 5506 CNRS / U.M. 2 Montpellier, France. [5] Tsai, L.-W. , 1999, Robot analysis: the mechanics of serial and parallel manipulators, John Wiley & Sons, New York [6] Kim, H.S. and Tsai, L.-W., 2002, "Design Optimization of a Cartesian Parallel Manipulator, ASME DETC Confer., 29-1 Sep. 2002, Montreal, Canada. [7] Gosselin, C.M.; Kong, X.; Foucault, S. and Bonev, I.A., 2004, A fully-decoupled 3-dof translational parallel mechanism, Parallel Kinematic Machines in research and practice, Proceedings of the 4th Chemnitz Parallel Kinematics Seminar, PKS2004, Verlag Wissenschaftliche Scripten, Reports from IWU, vol. 24, pp.595-610 [8] Di Gregorio and Parenti-Castelli, V., 2004, Design of 3-dof parallel manipulators based on dynamic performances, Parallel Kinematic Machines in research and practice, In Proc. of the 4th Chemnitz Parallel Kinematics Seminar, PKS2004, Verlag Wissenschaftliche Scripten, Reports from IWU, vol. 24, pp.385-397. [9] Herv, J. M., 1978, Analyse structurelle des mcanismes par groupe des dplacements, Mechanism and Machine Theory, 13, pp. 437450. [10] Hunt, K. H., 1983, Structural kinematics of in-parallel-actuated robot arms. Journal of Mechanisms, Transmission and Automation in Design. Transactions of the ASME, vol. 105, pp 705-712. [11] Kong, X. and Gosselin, C. M. 2004, Type synthesis of Threedegree-of-freedom spherical parallel manipulators, The International Journal of Robotics Research, 23(3), pp 237-245. [12] Brogrdh, T. , 2002, PKM Research - Important Issues, as seen from a Product Development Perspective at ABB Robotics. In Proc. of the WORKSHOP on Fundamental Issues and Future Research Directions for Parallel Mechanisms and Manipulators October 34, 2002, Quebec City, Quebec, Canada, pp 68-82. [13] Craig, J. J., 1989, Introduction to robotics: mechanics and control, Pearson Education, 2nd edition. [14] Raghavan, M.; Roth, B., 1995 Solving polynomial systems for the kinematic analysis and synthesis of mechanisms and robot manipulators. Journal of Mechanical Design. Vol.117, pp.71-79, [15] Hess-Coelho, T.A. and Raszl, G., 2004, Characterization of a prototype pf a robotic parallel structure considering its potential application as a machine-tool, Parallel Kinematic Machines in research and practice, Proceedings of the 4th Chemnitz Parallel Kinematics Seminar, PKS2004, Verlag Wissenschaftliche Scripten, Reports from IWU, vol. 24, pp 421-436. [16] ABB, 2005, Robotics-Parallel robots IRB340, Flexpicker, http://www.abb.com

Acknowledgement The authors would like to acknowledge the financial support of the Fundao de Amparo Pesquisa do Estado de So Paulo (FAPESP), Grant No. 03/13862-9. References
[1] Merlet, J.P., 2002, Still a long way to go on the road for parallel mechanisms, ASME DETC Confer., Montreal, Canada. [2] Neumann, K.E., 1988, Robot, US Patent No. 4,732,525. [3] Clavel, R., 1990, Device for the movement and positioning of an element in space, US Patent no. 4,976,582.