You are on page 1of 14

Applied Mathematical Modelling 57 (2018) 192–205

Contents lists available at ScienceDirect

Applied Mathematical Modelling


journal homepage: www.elsevier.com/locate/apm

Measuring the closeness to singularities of a planar parallel


manipulator using geometric algebra
Huijing Yao a, Qinchuan Li a,∗, Qiaohong Chen b, Xinxue Chai a
a
Mechatronic Institute, Zhejiang Sci-Tech University, Hangzhou, Zhejiang Province 310018, PR China
b
School of Information, Zhejiang Sci-Tech University, Hangzhou, Zhejiang Province 310018, PR China

a r t i c l e i n f o a b s t r a c t

Article history: A new index for measuring the closeness to the singularities of parallel manipulators us-
Received 19 June 2017 ing geometric algebra is proposed in this paper. Constraint wrenches acting on the moving
Revised 23 December 2017
platform of a parallel manipulator are derived using the outer product and dual operations.
Accepted 8 January 2018
Removing the redundant constraint wrenches, a singularity polynomial is obtained when
Available online 31 January 2018
the coefficient of the outer product of all the non-redundant constraint wrenches equals
Keywords: zero. A singularity surface can be drawn using the singularity polynomial. Similarly, an
Geometric algebra approximate singularity polynomial and approximate singularity surface can be obtained
Singularity analysis by imposing a threshold to the singular polynomial. Then the singularity volume is cal-
Parallel manipulator culated as the space between singularity surface and approximate singularity surface. The
Closeness to singularities new index is derived by calculating the ratio of the non-singularity workspace volume (the
workspace volume minus the singularity volume) to the workspace volume. The proposed
index is coordinate-free and has a clear geometrical and physical interpretation. This index
can be a basis for selecting structural parameters, path planning and mechanism design.
© 2018 Elsevier Inc. All rights reserved.

1. Introduction

Singularity analysis is a crucial issue for parallel manipulators (PMs) [1] and has always been a research hotspot. Many
researchers have used different methods to analyze the singularities of PMs [2–8]. Geometric algebra (GA), which was es-
tablished by Clifford [9], is a powerful mathematic tool that has been diffusely used in engineering fields [10–12]. Tanev
[13,14] first used geometric algebra to detect the singularities of a limited-DOF (degrees of freedom) PM. Then, Li et al.
[15] studied the singularities of a 3-RPS manipulator using Tanev’s method. Kim et al. [16] performed geometric singularity
analysis of a redundant PM using conformal geometric algebra (CGA). Not only can all the singularities of a PM be obtained
using geometric algebra but also the geometric meaning of the singularities can be detected.
In the singular configurations, parallel manipulators will lose its stiffness and become uncontrollable. Therefore, PMs
should work with a safe distance from singular configurations. However, the effective measurement of the distance between
the current configuration and the singular configuration is still a challenging problem.
Some methods, for example, the Jacobian determinant and the condition number [17,18], are put forward to detect the
closeness to singularities. However, they lack physical or geometric meaning [19]. Afterwards, many researchers studied
this issue and proposed metrics to measure the closeness to singularities. Hubert and Merlet [20] proposed a measure by
having the joint forces/torques lower than a given threshold to detect the closeness to singularities of a planar PM. In [21],


Corresponding author.
E-mail addresses: yao_hj@163.com (H. Yao), lqchuan@zstu.edu.cn, lqczstu@163.com (Q. Li), chen_lisa@zstu.edu.cn (Q. Chen), jxcxx88@163.com (X. Chai).

https://doi.org/10.1016/j.apm.2018.01.006
0307-904X/© 2018 Elsevier Inc. All rights reserved.
H. Yao et al. / Applied Mathematical Modelling 57 (2018) 192–205 193

the authors unite many different measures to evaluate closeness to singularities by adopting constrained optimization in
velocity and force. Hartley and Kerr [22] used reciprocal screw theory to detect the closeness to the singularity of a six-
screw system. However, for screw systems with less than six screws, the method in [22] is not applicable [23]. Wu et al.
[24] presented a new method considering motion/force transmissibility to measure the closeness to singularities of PMs.
This method has a physical meaning. For different types of non-redundant PM, it is a uniform metric for representing the
closeness to singularities. However, Bu [25] noted that for a limb with less than six DOF, the method in [24] was not
translation invariant when determining the transmission wrench. Bu used characteristic angles to measure the closeness to
singularities for similar mechanisms [26]. In addition, for the condition in which screws were not uniquely definite, he used
the normalized volume of weighted screws to detect the closeness to singularities [25].
To the best of the authors’ knowledge, one of the shortfalls in most of the existing measures is that they generally result
in a different value for each configuration. This leads to a problem of how best to combine those values throughout the
workspace to obtain a single number for mechanism design. Although the condition number is a single number for describ-
ing the overall kinematic behavior of a PM, however, Merlet [18] pointed the serious inconsistencies when the condition
number is applied to PMs with both rotational and translational DOFs.
Another shortfall in the existing measures is that they have the disadvantage of being non-invariant with respect to the
choice of coordinate frame, the choice of the position of the origin. For example, the manipulability index for mechanisms
and other measures based upon the singular values of the Jacobian.
Hence, a new index based on geometric algebra is presented in this work to measure the closeness to singularities. One
of the merits of the index is that the new index is a global index that describes the overall kinematic behavior of a PM with
respect to the whole workspace. Furthermore, the metric is independent of the selection of coordinate frame. Because the
singularity loci will not change with the coordinate frame change, so does the value of the index. Besides, this index has
a clear geometrical meaning and a physical meaning that it indicates the ratio of poor performance of a PM to the whole
workspace.
Moreover, this work presented here is an extension of our early work [27] with the novelties that a new index for mea-
suring the closeness to singularities is proposed and subsequently, it is applied to a 3-PRR PM to demonstrate its effective-
ness. The concepts of approximate singularity polynomial, approximate singularity surface singularity space and singularity
volume are proposed during the formulation of the index. The advantage of this work is not only to present all the singu-
larities distributed in the whole workspace but also to construct an index for measuring the closeness to singularities.
This paper is organized as follows: Section 2 gives a preliminary for some basic knowledge of geometric algebra and
the constraint screw system is introduced in this section. Section 3 introduces some information about singularities and the
definition of the new index for evaluating the closeness to singularities. In addition, a general procedure for obtaining the
index for measuring closeness to singularities is formulated. In Section 4, a 3-PRR PM is selected as an example to verify
the effectiveness of the proposed index. Finally, conclusions are summarized in Section 5.

2. Preliminary

2.1. Basics of GA

In this work relevant issues are analyzed in 6-dimensional vector space (G6 ). For more details about geometric algebra,
readers are referred to [10,28–30]. We introduce some foundations briefly.
The basis {e1 ,e2 ,e3 ,e4 ,e5 ,e6 } of vector space G6 satisfies the following relations:

ei · e j = 1 i= j
ei e j = ei ∧ e j = ei j i = j (1)
−e j ei

where “.” is the inner product operation, “∧” is the outer product operation.
The operation used in this work is mainly the outer product. It has the following properties

J ∧ (K + L ) = J ∧ K + J ∧ L
(cJ ) ∧ (dK ) = cd (J ∧ K ) (2)

where scalars c, d ∈ R, vectors J, K, L ∈ G6 .


The outer product of the same two vectors is zero J∧J = 0. The result is of significant importance in this paper. From a
geometric point of view, the two vectors are parallel if their outer product is zero.
The dual of a blade means the complementary of the space spanned by the blade. The dual of blade Et is:

E∗t  = Et  I6−1 = Et  (−I6 ) (3)

where Et ∈ G6 , I6 is the pseudoscalar of G6 , it is spanned by the basis vector {e1 ,e2 ,e3 ,e4 ,e5 ,e6 }. I6 = e1 e2 e3 e4 e5 e6 , I6−1 =
−I6 = e6 e5 e4 e3 e2 e1 .
194 H. Yao et al. / Applied Mathematical Modelling 57 (2018) 192–205

2.2. Screws in geometric algebra

A line vector in screw theory can be written as:

$; = [s; s0 ]T = [s; r × s + hs]T = [v1 , v2 , v3 ; b1 , b2 , b3 ]T (4)


where s is the direction vector of the line, r is the position vector of a point on the line, h is the pitch which can be 0, ∞
or other real numbers representing a pure revolution, a helical motion and a pure prismatic motion, respectively. vi (i = 1, 2,
3) and bi (i = 1, 2, 3) are the Plücker coordinates of the screw.
The Plücker coordinates of a screw S can be written as a vector in G6 with basis {e1 ,e2 ,e3 ,e4 ,e5 ,e6 } [13]:

S = v1 e1 + v2 e2 + v3 e3 + b1 e4 + b2 e5 + b3 e6 (5)
where v1 , v2 , v3 , b1 , b2 , b3 are the same coefficients as Eq. (4).
The elliptic polar screw of Eq. (5) is:

S˜ = S = b1 e1 + b2 e2 + b3 e3 + v1 e4 + v2 e5 + v3 e6 (6)
where  denotes an elliptic polar operation [31] which exchange the first three coefficients and the last three coefficients
of Eq. (5).
A prismatic joint with an infinite pitch can be presented in G6 as:

S = v1 e4 + v2 e5 + v3 e6 (7)
where v1 , v2 , v3 are the same coefficients as Eq. (5).

2.3. Blades of PMs

The outer product of mi joint twists in the ith limb of a PM can be written as follows:

Qi = Si1 ∧ Si2 ∧ ... ∧ Simi (8)


Qi represents the motion subspace that is spanned by the mi twists of the ith limb and is called a blade of limb motion
(BLM).
Assume that Sir is the actuator of the ith limb of a PM. When the actuator of the ith limb is locked, a reduced blade of
limb motion (RBLM) Qi can be calculated by removing Sir from Qi :

Qi = Si1 ∧ Si2 ∧ ... ∧ Si(r−1) ∧ Si(r+1) ∧ ... ∧ Simi (9)

Qi denotes the motion subspace spanned by the remaining mi − 1 twists of the ith limb.
The blade of limb constraint (BLC) consists of the constraint screws imposed on the moving platform by the ith limb. The
constraint space is the dual space or the orthogonal complement of the motion subspace of the ith limb. The dual space SCi
of a BLM is given by:

SCi = (Qi I6−1 ) (10)

where Qi is the limb motion space of the ith limb, I6−1 is the inverse of I6 .
Similarly, the dual or orthogonal complement space SCi of the RBLM of the ith limb is called the expanded blade of limb
constraint (EBLC), which can be obtained by:

SCi = (Qi I6−1 ) (11)

where Qi is the reduced limb motion space of the ith limb, I6−1 is the inverse of I6 .
For a general method, for example of solving nonlinear equations, solving the complex reciprocal screws of a screw
system is somewhat difficult. However, geometric algebra can solve this conveniently and elegantly.

3. Formulation of the index measuring the closeness to singularities

3.1. Singularity of a PM

For a PM, the differential relationship between the input θ (θ 1 ,...,θ n ) and output x(x1 ,...,xn ) coordinates with respect to
time is [4]
• •
A x +B θ = 0 (12)
where A and B are both n × n matrices. Eq. (12) can also be written as [32]
 
• v
θ =J w (13)
H. Yao et al. / Applied Mathematical Modelling 57 (2018) 192–205 195

Fig. 1. Singularity surface and the approximate singularity surfaces.

where the Jacobian matrix J = B − 1 A, [v, w]T = $ is n × 1 twist vector of output linear velocity v and angular velocity w.
The singularity of a PM occurs when the Jacobian matrix J degenerates. Consequently, kinematic properties of PMs change
dramatically at the singular configuration and its neighborhood. Thus in a context of design, it is crucial to detect and
avoid singular configurations in order to improve the kinematic performance for a given structure parameter. Then how to
measure the distance between a current configuration to a singular configuration is a challenge problem. Besides, note that
the concept of singularity distance means how far a configuration from a singular configuration. There is no mathematical
metric defining the distance to a singular configuration if the PM has both rotational and translational DOFs [33].
Many measures, for example, condition number κ , see Eq. (14), and manipulability index, see Eq. (15) [18] are proposed
for measuring closeness to singularities of PMs.
 
κ = J −1  J (14)


ω= det J (θ )J T (θ ) (15)

However, they are both based on the singular values of the Jacobian matrix and thus have the drawbacks of being non-
invariant with respect to the choice of coordinate frame. In addition, they are not geometrically meaningful.

3.2. Definition of the new index

When J = 0, the mechanism is said to be singular. The symbolic expression of J is a polynomial and is also referred
as singularity polynomial. In this paper, the singularity polynomial is denoted as D(X). For a 3-DOF PM, X contains three
independent parameters including position and orientation parameters. D(X) has the following properties:

(1) D(X) = 0 if X is a singular point.


(2) |D(X)| > 0 if X is not a singular point.

In this paper, assume that when the value of |D(X)| is close to 0 enough, the point X can also be considered as a singular
point. Therefore, we have

|D(X )| = ε (16)
where ε is called singularity threshold, which is a number close to 0. The selection of the threshold will be explained clearly
by an example in Section 4. In addition, A new concept of approximate singularity polynomial DAS is derived from Eq. (16).
This can be written as DAS =D(X) ± ε .
Note that in property (1), the singular points which satisfy D(X) = 0 form a surface, called the singularity surface, see the
surface 2 in Fig. 1. In addition, all points that satisfy |D(X)| = ε form a surface, called the approximate singularity surface.
1 , 3 in Fig. 1 are approximate singularity surfaces. Besides, the points that satisfy |D(X)| < ε form a space area that is
called the singularity space.
Note that X mentioned in the above discussion is a single configuration of a point. In order to deal with all the singular-
ities throughout the workspace to obtain a single number for manipulator design, we define a new index λ as
VW − VS VS
λ= =1− (17)
VW VW
where VW is the volume of the workspace, VS is the singularity volume (the volume of the space when |D(X)| < ε ).
It is worth noting that this index has a clear geometrical and physical meaning. It is also a global index that describes
the overall kinematic behavior of a PM in the whole workspace.
196 H. Yao et al. / Applied Mathematical Modelling 57 (2018) 192–205

Fig. 2. Procedure for calculating the new index.

3.3. Procedure for calculating the index

A flow chart that describes the procedure for calculating this index is presented in Fig. 2.
Step 1. Express the jth twist of the ith limb Sij in G6 . Calculate the RBLM (Qi ) of the ith limb and the EBLC (SCi ) of the
ith limb (i = 1…n, n is the number of limbs of the PM).
Note that there will be redundant and common constraints during the calculation process of EBLC (SCi ). Identify and
remove the redundant or common constraints from SCi by the method of [34]. The new EBLC (SCi ) is denoted by SCi
.
Step 2. Calculate the outer product of all SCi
, derive the singularity polynomial DS by assigning the vector to zero.
The singularity polynomial can be used to obtain the singularity surface. For an n-DOF (n≤ 6) parallel manipulator with
n actuators, the moving platform bears 6-n linearly independent constraint wrenches that compose all of the BLCs when the
mechanism is not singular. At this time, the moving platform cannot move finitely or instantaneously if the n actuators are
locked. In contrast, the moving platform can move finitely or instantaneously since the six constraint wrenches are linearly
dependent in singular configuration. These six constraint wrenches form all the EBLCs. According to the introduction of a
property of geometric algebra, the outer product can be used to detect the singular configurations since the six constraint
wrenches are linearly dependent when the mechanism is singular.
Step 3. Give a threshold ε to the singularity polynomial to obtain the approximate singularity polynomial DAS .
Assign a sufficiently small value ε close to zero to the singularity polynomial DS , and an approximate singularity poly-
nomial is obtained as DAS = DS ± ε . N can be any integer from 1 to 9. Approximate singularity surfaces can be drawn with
the approximate singularity polynomial. The volume of the space between the singularity surface and the approximate sin-
gularity surface is called the singularity volume.
The selection criteria of the threshold ε is introduced as follows. Firstly, determine the worst accuracy of the value of DS
by replacing the solutions into the polynomial. Secondly, illustrate the distribution of DS , select the appropriate value. Note
that the selection of the threshold depends on the kind of PM. For the same kind of PM, a same threshold is selected.
Step 4. Calculate the volumes of the singularity space and the workspace of the PM. Finally, the value of the new index
can be obtained using Eq. (17).
H. Yao et al. / Applied Mathematical Modelling 57 (2018) 192–205 197

Fig. 3. A diagram of 3-PRR PM.

Table 1
Eight working modes of 3-PRR PM.

WM1 WM2 WM3 WM4 WM5 WM6 WM7 WM8

σ1 + + – + – – + –
σ2 + + + – – – – +
σ3 + – + + – + – –

4. Application

In this section, for detecting the efficiency of the new index, the measurement of the closeness to singularities of the
3-PRR PM is presented.

4.1. Blades of the 3-PRR PM

Fig. 3 shows a 3-PRR [35] PM in which both the moving and the fixed platform are equilateral triangles with circum-
scribed circle radii R and r. The moving platform is connected to the base by three links in which prismatic joints are
actuated. A fixed coordinate frame XOY is established with the origin located at the center of the fixed platform. A moving
coordinate frame xoy is established with the origin located at the center of the moving platform. ρ i (i = 1,2,3) represents the
length of the prismatic joint from the initial position. α i (i = 1,2,3) is defined as the angle between OX and the three linear
guides on the fixed platform. φ is the angle between C2 C3 and OX. The position and orientation of the moving platform can
be determined by the coordinates of point o(x, y, 0) together with √ the angle φ . B1 C1 , B2 C2 , B3 C3 have √
the same length of l.
√ The coordinates of points Ai , Bi , Ci (i = 1, 2, 3) are A1 = ( 3R/2, −R/2, 0 ), A2 = √ (0, R, 0), A3 = (− 3R/2, −R/2, 0 ), B1 =
( 3R/2+ √ρ1 cos(α1 ), −R/2 + ρ1 √ sin(α1 ), 0 ), B2 = (ρ 2 cos(α 2 ), R + ρ 2 sin(α 2 ),0), B3 = (− 3R/2 + ρ3 , −R/2, 0 ) and C1 = (0, r, 0),
C2 = (− 3r /2, −r /2, 0 ), C3 = ( 3r /2, −r /2, 0 ), respectively.
Among them, ρi = (qi1 + σi i )/2, qi2 = q2i3 + q2i4 − l 2 , qi1 = 2cαi qi3 + 2sαi qi4 , qi3 = x − xAi + xCi cαi − yCi sαi , qi4 = y − yAi +
xCi sαi + yCi cαi .
i = q2i1 − 4qi2 , i ≥ 0(i = 1, 2, 3) defines the workspace which is the intersection of three pairs of parallel sur-
faces. σ i = ±1(i = 1, 2, 3) correspond to the eight inverse kinematics of the 3-PRR PM. There are eight working modes
(WMs) as Table 1 shows: WM1(+ + +), WM2(+ + −), WM3(− + +), WM4(+ − +), WM5(− − −), WM6(− − +), WM7(+ − −),
WM8(− + −).
The twist-related direction vectors sij (i = 1,2,3, j = 1,2,3) are s11 = [−cα1 , sα1 , 0]T , s21 = [−cα2 , −sα2 , 0]T , s31 = [1, 0, 0]T ,
si2 = [0, 0, 1]T , si3 = [0, 0, 1]T , where ca and sa denote the cosine and sine functions of angle α .
The screws of limb 1, 2, 3 are written as:

S11 = −cα1 e4 + sα1 e5


√ √
S12 = e3 + (( 3ρ1 )/2 − R/2 )e4 + (ρ1 /2 − ( 3R )/2 )e5
S13 = e3 + (y + r cφ )e4 + (r sφ − x )e5 (18)

S21 = −cα2 e4 − sα2 e5



S22 = e3 + (R − ( 3ρ2 )/2 )e4 + (ρ2 /2 )e5
√ √
S23 = e3 + (y − r cφ /2 − 3r sφ /2 )e4 + ( 3r cφ /2 − r sφ /2 − x )e5 (19)
198 H. Yao et al. / Applied Mathematical Modelling 57 (2018) 192–205

S31 = e4

S32 = e3 + (R/2 )e4 + ( 3R/2 − ρ3 )e5
√ √
S33 = e3 + (y − r cφ /2 + 3r sφ /2 )e4 + (−x − r sφ /2 − 3r cφ /2 )e5 (20)
The RBLM of limb 1 with the actuated prismatic joint locked is the outer product of two passive screws:
Q1 = S12 ∧ S13
√ √
= (R/2 + y − 3ρ1 /2 + r cφ )e3 ∧ e4 + ( 3R/2 − x − ρ1 /2 + r sφ )e3 ∧ e5
√ √
+ ((R/2 − 3ρ1 /2 )(x − r sφ ) − (ρ1 /2 − 3R/2 )(y + r cφ ))e4 ∧ e5 (21)
The EBLC of limb 1 is derived as:

SC1 = (Q1 I6−1 ) = (ρ1 /2 + x − 3R/2 − r sφ )e1 ∧ e3 ∧ e4 ∧ e5

+ (R/2 + y − ( 3ρ1 )/2 + r cφ )e2 ∧ e3 ∧ e4 ∧ e5
√ √
−((R/2 − 3ρ1 /2 )(x − r sφ ) − (ρ1 /2 − 3R/2 )(y + r cφ ))e3 ∧ e4 ∧ e5 ∧ e6
= SC11 ∧ SC12 ∧ SC13 ∧ SC14 (22)
where
√ √
SC11 = ((ρ1 /2 + x − 3R/2 − r sφ )e1 + (R/2 + y − ( 3ρ1 )/2 + r cφ )e2
√ √
+ ((R/2 − 3ρ1 /2 )(x − r sφ ) − (ρ1 /2 − 3R/2 )(y + r cφ ))e6
SC12 = e3 , SC13 = e4 , SC14 = e5 (23)
Similarly, the EBLC of limb 2 and limb 3 can be obtained as:

SC2 = (Q2 I6−1 ) = (ρ2 /2 + x + r sφ /2 − 3r cφ /2 )e1 ∧ e3 ∧ e4 ∧ e5
√ √
+ (y − R + 3ρ2 /2 − r cφ /2 − 3r sφ /2 )e2 ∧ e3 ∧ e4 ∧ e5
√ √ √
−((ρ2 (r cφ /2 − y + 3r sφ /2 ))/2 − (R − 3ρ2 /2 )(x + r sφ /2 − 3r cφ /2 ))e3 ∧ e4 ∧ e5 ∧ e6
= SC21 ∧ SC22 ∧ SC23 ∧ SC24 (24)
where
√ √ √
SC21 = ((ρ2 /2 + x + r sφ /2 − 3r cφ /2 )e1 + (y − R + 3ρ2 /2 − r cφ /2 − 3r sφ /2 )e2
√ √ √
+ ((ρ2 (r cφ /2 − y + 3r sφ /2 ))/2 − (R − 3ρ2 /2 )(x + r sφ /2 − 3r cφ /2 ))e6
SC22 = e3 , SC23 = e4 , SC24 = e5 (25)

√ √
SC3 = (Q3 I6−1 ) = (x − ρ3 + 3R/2 + r sφ /2 + 3r cφ /2 )e1 ∧ e3 ∧ e4 ∧ e5

+ (R/2 + y − r cφ /2 + 3r sφ /2 )e2 ∧ e3 ∧ e4 ∧ e5
√ √ √
−((R(x + r sφ /2 + 3r cφ /2 ))/2 + (ρ3 − 3R/2 )(y − r cφ /2 + 3r sφ /2 ))e3 ∧ e4 ∧ e5 ∧ e6
= SC31 ∧ SC32 ∧ SC33 ∧ SC34 (26)
where
√ √ √
SC31 = ((x − ρ3 + 3R/2 + r sφ /2 + 3r cφ /2 )e1 + (R/2 + y − r cφ /2 + 3r sφ /2 )e2
√ √ √
+ ((R(x + r sφ /2 + 3r cφ /2 ))/2 + (ρ3 − 3R/2 )(y − r cφ /2 + 3r sφ /2 ))e6
SC32 = e3 , SC33 = e4 , SC34 = e5 (27)
SC1 , SC2 , SC3 are three 4-blade vectors containing the constraints e3 , e4 , e5 . The vector e3 denotes one constraint force in
the z-axis. Vectors e4 , e5 represent two constraint couples in the xy plane. Therefore, there are six constraint couples in the
xy plane and three constraint forces in the z-axis totally. Obviously, they are linearly dependent.
Using the method for identifying the redundant constraints [34], new EBLCs are derived:

= S
SC1 C11 ∧ SC12 ∧ SC13 ∧ SC14

= S
SC2 C21

= S
SC3 (28)
C31

and S
are two 1-blades, S
is a 4-blade. Note that the three-blade e e e can join with any one of the other three
SC2 C3 C1 3 4 5

, S
and S
indicate six constraint screws with the
constraint wrenches and they do not affect the singularity analysis. SC1 C2 C3
actuated joints locked.
H. Yao et al. / Applied Mathematical Modelling 57 (2018) 192–205 199

4.2. Singularity polynomial and singularities of 3-PRR PM

When one or two limbs of the mechanism are perpendicular to the translation joint, the mechanism reaches the bound-
ary of the workspace. These singular conditions are a type I [36] singularity.

, S
and S
is calculated as:
To detect type II [36] singularities, the outer product of SC1 C2 C3


∧ S
∧ S

D = SC1 C2 C3
= DS e1 ∧ e2 ∧ e3 ∧ e4 ∧ e5 ∧ e6 (29)

DS → D(x, y, φ ) (30)

or simply,

D
= SC11 ∧ SC21 ∧ SC31
= DS e1 ∧ e2 ∧ e6 (31)
where DS is the same as that in Eq. (30), SC11 ,SC21 ,SC31 are three constraints along B1 C1 ,B2 C2 ,B3 C3 , respectively.
D
is a 3-blade with a long coefficient DS and is called the singularity polynomial. The singularity polynomial contains
the structural parameters and the output variants x, y, φ of the mechanism. The outer product is a dimension increasing
operation and the coefficients of the vectors are multiplied. Thus, the coefficient DS is usually a long one with the degree
3 in y and 3 in x. Due to space limitations, the polynomial is not given here. When the EBLCs are linearly dependent, the
value of the output variants in some configurations makes the polynomial DS equal to zero. These configurations are the
singularity configurations. The structural parameters of the 3-PRR PM are given as l = 0.13, α 1 = 2π /3, α 2 = 4π /3, α 3 = 0,
r = 0.064, R = 0.1408. When the singularity polynomial is equal to zero, the singularity surfaces of the eight working modes
of the 3-PRR PM are shown in Fig. 4(a)∼(h). Note that each singular surface corresponds to one working mode. In addition,
the workspace of the 3-PRR PM is shown in Fig. 4(i)
To identify the properties of these singularity surfaces, a cross section in φ = π /35 is made in the singularity surfaces of
the eight working modes (see Fig. 5). It can be seen that the singularity loci in the WMs[ + + −], [− + + ], [ + − + ] exhibit
a 120° rotational symmetry. In addition, the singularity loci in the WMs[ + − −], [− + −] are antisymmetric with respect
to the y axis. Afterwards, four singular points p1 ,p2 ,p3 ,p4 are selected on the singularity loci of WMs[ + + + ], [ + + −], [ + –
+ ], [− − −], respectively. The coordinates (x, y, φ ) of the four singular points are p1 (−0.01, 0.05878784721620087, π /35),
p2 (−0.08, 0.026334283509461667, π /35), p3 (−0.04, −0.05610285499306627, π /35), p4 (0.015, −0.115700940222628, π /35),
respectively. Note that these singular points are obtained by solving the equation DS = 0. This equation only contains the
variable y because the other two variables x and φ are given. However, the solutions y to this equation are not exact
values. So the number of digits after the decimal point of the value y will affect the orders of magnitude of DS . The four
corresponding singular configurations of the 3-PRR PM are illustrated in Fig. 6. It turns out that three constraints along three
limbs in the four configurations satisfy the following equation

⎪SC11 ∧ SC21 ∧ SC31 = 7.4539 · 10−20 e1 e2 e6 ≈ 0


⎨ −19
SC11 ∧ SC21 ∧ SC31 = 2.4395 · 10 e1 e2 e6 ≈ 0
(32)

⎪SC11 ∧ SC21 ∧ SC31 = 2.1684 · 10−19 e1 e2 e6 ≈ 0

⎩ −20
SC11 ∧ SC21 ∧ SC31 = 5.4210 · 10 e1 e2 e6 ≈ 0
In the four configurations, the geometric conditions are that two constraints along the limb are parallel to each other
and they intersect with the third constraint simultaneously.
Fig. 7 shows the distribution of all the eight singularity loci in the whole workspace when φ = π /35. As can be seen in
Fig. 7, the singularity loci are distributed inside the workspace and they change the working modes where they intersect
with the workspace.

4.3. Approximate singularities of the 3-PRR PM

A new definition, namely the approximate singularity polynomial is proposed for obtaining the approximate singularity
surface based on the singularity polynomial. The key to get the approximate singularity polynomial is the selection of a
threshold ε .
The first step in choosing a threshold is to determine the largest order of magnitude of DS . As we can see in Eq. (32),
the order of magnitude of the singularity polynomial is −19∼−20 power under the singular points p1 ∼ p4 . When the values
of y in the singular points p1 ∼ p4 are accurate to one after the decimal point, that are p1 (−0.01, 0.1, π /35), p2 (−0.08, 0.0,
π /35), p3 (−0.04, −0.1, π /35), p4 (0.015, −0.1,π /35). Consequently, this will affect the order of magnitude of the singularity
polynomial value. Considering this, we take these singular points into the singularity polynomial DS again. It is proved that
the maximum order of magnitude for the singular polynomial is −4 power. At this time, three constraints along the limbs
200 H. Yao et al. / Applied Mathematical Modelling 57 (2018) 192–205

Fig. 4. (a)∼(h) Singularity surfaces of eight WMs (i) Workspace of 3-PRR PM.

satisfy the following relationship



⎪SC11 ∧ SC21 ∧ SC31 = −1.2606 · 10−4 e1 e2 e6


⎨ −5
SC11 ∧ SC21 ∧ SC31 = 2.9951 · 10 e1 e2 e6
(33)

⎪SC11 ∧ SC21 ∧ SC31 = 1.5456 · 10−4 e1 e2 e6

⎩ −5
SC11 ∧ SC21 ∧ SC31 = −6.8094 · 10 e1 e2 e6

The other step is to determine the suitable order of magnitude of DS . In order to explain clearly this process, we take
WM2[ + + −] as an example. Fig. 8(a) shows the singularity loci of WM2[ + + −] when φ = π /35, every point on the loci
makes DS = 0. When the range of DS is given in [− 10 − 4 , 10 − 4 ] according to Eq. (33). However, it can be observed in
Fig. 8(b) that the surface distributes irregularly, which means the range [− 10 − 4 , 10 − 4 ] is large and also the error is big.
Thus, the value of DS is reduced by an order of magnitude. Assume that the range of DS is [− ξ , ξ ] = [− N • 10 − 5 , N •
10 − 5 ]. We found that when N ≤ 2 the distribution of DS is regular, and this can be seen form Fig. 8(c). However, when N
≥ 3, DS needs to be reduced by an order of magnitude again. It worth noting that the value ε ≤ ξ can be selected as the
threshold value. Therefore, the value in [− ξ , ξ ] = [− N • 10 − 5 , N • 10 − 5 ] is called optional value, and the value ε is known
as the selected threshold value.
In this work, the threshold is given as ε = 5 • 10 − 6 . Consequently, the approximate singularity polynomial is obtained
as DAS = DS ± ε . Fig. 8(d) presents the distribution of DAS which has the similar tendency as DS = 0. Therefore, by utilizing
DAS , the approximate singularity surfaces can be derived. Note that each singularity surface in Fig. 4(a)∼(h) has two sym-
metrical approximate singularity surfaces. For example, the singularity surface and approximate singularity surfaces of the
WM8[− + −] is provided in Fig. 9. In order to reveal the overall behavior of the PM, the approximate singularity surfaces for
the eight WMs in the workspace are illustrated in Fig. 10(a). Moreover, Fig. 10(b) is the top view of Fig. 10(a). In addition,
the explicit presentation of the distribution of approximate singularity loci in the workspace for different orientation angles
H. Yao et al. / Applied Mathematical Modelling 57 (2018) 192–205 201

Fig. 5. 2d-Singularity loci of eight WMs when φ = π /35.

Fig. 6. Configurations of 3-PRR PM in singular points p1 ,p2 ,p3 ,p4 .

is shown in the form of Fig. 11. Every point in the space between two approximate singularity surfaces of each WM is con-
sidered as a singular point. As a result, the mechanism will have a poor performance in these singular configurations. Next
section an index is proposed to quantify this statement.

4.4. Measuring the closeness to the singularities

As mentioned in the previous section, the points in the singularity space are considered as singular points. Therefore,
the volume of this space is defined as singularity volume. We then use the singularity volume and the workspace volume
to formulate a metric for measuring the closeness to singularities of PMs.
In order to calculate the volumes of singularity and workspace of the 3-PRR PM, the Monte Carlo method [37], which
solves mathematical problems by means of random sampling, is utilized. In addition, the numbers of random points have
an effect on the accuracy of the calculation of the Monte Carlo method. As a comparison, the workspace volume of the
mechanism is determined to be 0.01548 using numerical integration. Then the volume of the workspace of the 3-PRR PM is
calculated using Monte Carlo method with different numbers of random points. The calculation is conducted 100 times for
8 sets of random points. Table 2 provides the maximum and minimum errors of the results. From Table 2 we can see that
the maximum error is lower with the random points increasing.
202 H. Yao et al. / Applied Mathematical Modelling 57 (2018) 192–205

Fig. 7. Distribution of all the 2-d singularity loci of 3-PRR PM in the workspace when φ = π /35.

Fig. 8. (a) 2d-Singularity loci of WM[ + + −] for φ = π /35 (b) Distribution of DS when φ = π /35 and DS ∈ [ − 10 − 4 , 10 − 4 ] (c). Distribution of DS when
φ = π /35 and DS ∈ [ − 2 • 10 − 5 , 2 • 10 − 5 ] (d). Distribution of DS when φ = π /35 and DS ∈ [ − 5 • 10 − 6 , 5 • 10 − 6 ].

Table 2
Workspace volume computational error with different numbers of random points.

Number random points errormax errormin

1 10 0 0 0.00347 0.0 0 0 04
2 50 0 0 0.00107 0.0 0 0 0 0
3 80 0 0 0.0 0 081 0.0 0 0 0 0
4 15,0 0 0 0.0 0 070 0.0 0 0 0 0
5 20,0 0 0 0.0 0 059 0.0 0 0 0 0
6 50,0 0 0 0.0 0 033 0.0 0 0 0 0
7 10 0,0 0 0 0.0 0 024 0.0 0 0 0 0
8 20 0,0 0 0 0.0 0 022 0.0 0 0 0 0
H. Yao et al. / Applied Mathematical Modelling 57 (2018) 192–205 203

Fig. 9. Singularity surface and approximate singularity surface of WM [− + −].

Fig. 10. (a) Approximate singularity surfaces of eight WMs in the workspace (b) Top view of (a).

Fig. 11. Distribution of approximate singularity in the workspace with orientation angle range from 0 rad to 1 rad.
204 H. Yao et al. / Applied Mathematical Modelling 57 (2018) 192–205

Table 3
Index values of four sets of structure parameters.

l R r λ
0.13 0.1408 0.064 0.725
0.15 0.1408 0.064 0.862
0.13 0.1408 0.075 0.783
0.13 0.1650 0.064 0.631

Fig. 12. The effect of limb length l, fixed platform radius R and moving platform radius r on the value of the index λ. (For interpretation of the references
to color in this figure, the reader is referred to the web version of this article.)

In this work, the volume calculations are performed with 2 • 105 random points. Note that the volume of the singularities
of the 3-PRR PM is the sum of the volumes of the eight working modes VASi (i = 1…8). Based on Eq. (17), λ is calculated as:


8
VASi
VW − VAS i=1
λ= =1− (34)
VW VW
Assume that the fixed platform and the moving platform of the 3-PRR PM in this paper are both equilateral triangles,
that is α 1 = 2π /3, α 2 = 4π /3, α 3 = 0. Therefore, the effect of changes in limb length l, radii of the fixed moving platform
R and moving platform r on the index λ are considered. For example, Table 3 presents four index values of four sets of
structure parameters. The index λ of the original structure parameter l = 0.13, r = 0.064, R = 0.1408 is calculated as 0.725. We
found that the index λ increases as the limb length l and the moving platform radius r increase. However, the increase of
the fixed platform radius R leads to the decrease of the index λ. Fig. 12 verifies the above statement. The green line (the
effect of fixed platform radius R on the value of the index λ) has an opposite trend with the red line (the effect of limb
length l on the value of the index λ) and the blue line (the effect of moving platform radius r on the value of the index
λ). In Fig. 12, λ = 0 means the PM is always singular throughout the whole workspace. Note that the above conclusions are
drawn by keeping the two of the three parameters unchanged. The value of the index λ may become larger or smaller if
the three structure parameters are given arbitrarily.
A larger λ value indicates that there are less configurations of the PM that are close to singularities. Moreover, this index
can be used to identify the rationality of the structure parameters of a PM. For example, if 0.7 is the defined limit value,
the numbers greater than 0.7 are considered usable.
It worth noting that the value of the index λ is a global index which illustrates the overall performance of the mechanism
with the given structure parameter. This is important to the design, optimization and structure parameter selecting of the
mechanism. In addition, this index has a clear geometrical meaning and a physical meaning that it indicates the ratio of
poor performance of a PM to the whole workspace.

5. Conclusions

In this work, we proposed a new index based on geometric algebra to measure the closeness to singularities of PMs.
This index was defined as the ratio of the non-singularity workspace volume (the workspace volume minus the singularity
volume) to the workspace volume of a parallel manipulator, which made the index have clear geometrical and physical
H. Yao et al. / Applied Mathematical Modelling 57 (2018) 192–205 205

interpretation. One of the merits of the new index is that it dealt well with the problem of mixed dimensions without
the normalized operation because it was constructed with space volumes. Furthermore, importantly, the value of this new
index is a single number which has the advantage of being a general metric for evaluating the closeness to singularities of
a parallel mechanism with the given link parameter. The index was applied to a planar 3-PRR PM to testify its effectiveness.
It is shown that the index, obtained by the Monte Carlo computing after the geometric algebra calculation, was proved to
be suitable to measure the closeness to singularities with its value distributed within [0,1]. Notably, the index is also useful
in the design and optimization of the mechanism. Future works include the extension of this methodology to the further
studies (i.e. path planning and optimal design) and the analyses of more complex mechanisms.

Acknowledgments

The work is supported by the National Natural Science Foundation of China(NSFC) under Grant 51525504 and Natural
Science Foundation of Zhejiang Province under Grant LY17E050028.

References

[1] K.H. Hunt, Structural kinematics of in-parallel-actuated robot-arms, J. Mech. Des. 105 (1983) 705–712.
[2] J.P. Merlet, Singular configurations of parallel manipulators and Grassmann geometry, Int. J. Robot. Res. 8 (1989) 45–56.
[3] H. Fang, Y. Fang, K. Zhang, Reciprocal screw theory based singularity analysis of a novel 3-DOF parallel manipulator, Chin. J. Mech. Eng. 25 (2012)
647–653.
[4] C. Gosselin, J. Angeles, Singularity analysis of closed-loop kinematic chains, IEEE Trans. Robot. Autom. 6 (1990) 281–290.
[5] D.S. Zlatanov, R.G. Fenton, B. Benhabib, Classification and interpretation of the singularities of redundant mechanisms, in: Proceedings of ASME Design
Engineering Technical Conference, Atlanta, GA, 1998, pp. 13–16. Sept.
[6] F. Park, J.W. Kim, Singularity analysis of closed kinematic chains, J. Mech. Des. 121 (1999) 32–38.
[7] P. Ben-Horin, M. Shoham, Singularity analysis of a class of parallel robots based on Grassmann–Cayley algebra, Mech. Mach. Theory 41 (2006) 958–970.
[8] D. Kanaan, P. Wenger, S. Caro, D. Chablat, Singularity analysis of lower mobility parallel manipulators using Grassmann–Cayley algebra, IEEE Trans.
Robot. 25 (2009) 995–1004.
[9] P. Clifford, Applications of Grassmann’s extensive algebra, Am. J. Math. 1 (1878) 350–358.
[10] C. Perwass, H. Edelsbrunner, L. Kobbelt, K. Polthier, Geometric Algebra with Applications in Engineering, Springer, 2009.
[11] E. Hitzer, T. Nitta, Y. Kuroe, Applications of Clifford’s geometric algebra, Adv. Appl. Clifford Algeb. 23 (2013) 377–404.
[12] J. Lasenby, E. Bayro-Corrochano, A.N. Lasenby, G. Sommer, A new methodology for computing invariants in computer vision, Proceedings of Thirteenth
International Conference on Pattern Recognition, ICPR, IEEE, 1996, p. 393.
[13] T.K. Tanev, Singularity analysis of a 4-DOF parallel manipulator using geometric algebra, Advances in Robot Kinematics, Springer, 2006, pp. 275–284.
[14] T.K. Tanev, Geometric algebra approach to singularity of parallel manipulators with limited mobility, in: J. Lenarčič, P. Wenger (Eds.), Advances in
Robot Kinematics: Analysis and Design, Springer Netherlands, Dordrecht, 2008, pp. 39–48.
[15] Q. Li, J.N. Xiang, X. Chai, C. Wu, Singularity analysis of a 3-RPS parallel manipulator using geometric algebra, Chin. J. Mech. Eng. 28 (2015) 1204–1212.
[16] J.S. Kim, J.H. Jeong, J.H. Park, Inverse kinematics and geometric singularity analysis of a 3-SPS/S redundant motion mechanism using conformal geo-
metric algebra, Mech. Mach. Theory 90 (2015) 23–36.
[17] T. Yoshikawa, Manipulability of robotic mechanisms, Int. J. Robot. Res. 4 (1985) 3–9.
[18] J.P. Merlet, Jacobian, manipulability, condition number and accuracy of parallel robots, J. Mech. Des. 28 (2007) 175–184.
[19] J. Duffy, The fallacy of modern hybrid control theory that is based on “orthogonal complements” of twist and wrench spaces, J. Robot. Syst. 7 (1990)
139–144.
[20] J. Hubert, J.P. Merlet, Static of parallel manipulators and closeness to singularity, J. Mech. Robot. Trans. ASME 1 (2009).
[21] P.A. Voglewede, I. Ebert-Uphoff, Overarching framework for measuring closeness to singularities of parallel manipulators, IEEE Trans. Robot. 21 (2005)
1037–1045.
[22] D. Hartley, D. Kerr, Invariant measures of the closeness to linear dependence of six lines or screws, Proc. Inst. Mech. Eng. Part C J. Mech. Eng. Sci. 215
(2001) 1145–1151.
[23] D. Kerr, D. Hartley, Invariant measures of closeness to linear dependency of screw systems, Proc. Inst. Mech. Eng. Part C J. Mech. Eng. Sci. 220 (2006)
1033–1043.
[24] C. Wu, X.J. Liu, F. Xie, J. Wang, New measure for ‘Closeness’ to singularities of parallel robots, in: Proceedings of the IEEE International Conference on
Robotics and Automation, 2011, pp. 5135–5140.
[25] W. Bu, Closeness to singularities of manipulators based on geometric average normalized volume spanned by weighted screws, Robotica 35 (2017)
1616–1626.
[26] W. Bu, Closeness to singularities of robotic manipulators measured by characteristic angles, Robotica 34 (2016) 2105–2115.
[27] H. Yao, Q. Chen, X. Chai, Q. Li, Singularity analysis of 3-RPR parallel manipulators using geometric algebra, Adv. Appl. Clifford Algeb. 27 (2017)
2097–2113.
[28] D. Hestenes, New Foundations for Classical Mechanics, Springer Science & Business Media, 2012.
[29] D. Hildenbrand, Foundations of Geometric Algebra Computing, Springer, 2013.
[30] E. Bayro-Corrochano, Geometric Computing: For Wavelet Transforms, Robot Vision, Learning, Control and Action, Springer Publishing Company, Incor-
porated, 2010.
[31] H. Lipkin, J. Duffy, The elliptic polarity of screws, J. Mech. Des. 107 (1985) 377–387.
[32] H.M. Daniali, P. Zsombor-Murray, J. Angeles, Singularity analysis of planar parallel manipulators, Mech. Mach. Theory 30 (1995) 665–678.
[33] X.-J. Liu, C. Wu, J. Wang, A new approach for singularity analysis and closeness measurement to singularities of parallel manipulators, J. Mech. Robot.
4 (2012) 041001–1–10.
[34] Q. Li, X. Chai, Mobility analysis of limited-DOF parallel mechanisms in the framework of geometric algebra, J. Mech. Robot. 8 (4) (2016) 041005–1–9.
[35] C.M. Gosselin, S. Lemieux, J.P. Merlet, A new architecture of planar three-degree-of-freedom parallel manipulator, in: Proceedings of the IEEE Interna-
tional Conference on Robotics and Automation, Vol. 3734, 1996, pp. 3738–3743.
[36] I.A. Bonev, D. Zlatanov, C.m.M. Gosselin, Singularity analysis of 3-DOF planar parallel mechanisms via screw theory, J. Mech. Des. 125 (2003) 573.
[37] J. Rastegar, B. Fardanesh, Manipulation workspace analysis using the Monte Carlo Method, Mech. Mach. Theory 25 (1990) 233–239.

You might also like