Development of Two Cooperative Stewart Platforms For Machining
Uploaded by
Quang BáchDevelopment of Two Cooperative Stewart Platforms For Machining
Uploaded by
Quang BáchDEVELOPMENT OF TWO COOPERATIVE
STEWART PLATFORMS FOR MACHINING
VINCENSIUS BILLY SAPUTRA
B.Eng., ITB
A THESIS SUBMITTED
FOR THE DEGREE OF DOCTOR OF
PHILOSOPHY
DEPARTMENT OF MECHANICAL ENGINEERING
NATIONAL UNIVERSITY OF SINGAPORE
2012
DECLARATION PAGE
i
ACKNOWLEDGEMENTS
I am deeply indebted to my supervisors, Prof. Andrew Nee Yeh Ching,
and Assoc. Prof. Ong Soh Khim for their suggestions, guidance, insights, and
patience that have been invaluable to this research project and this thesis, and
will long be treasured and greatly appreciated.
I would like to thank all fellow students in our research group and AR
group, especially Dr Ng Chee Chung, Dr Niu Sihong, Dr Fang Hongchao, for
their friendship, help, encouragement, research ideas and opinions. In
addition, I am very much obliged to the Advanced Manufacturing Laboratory
for their assistance throughout the research. I express my gratitude to the
Department and laboratory staff members, especially Mr Tan Choon Huat,
Mr Lim Soon Cheong, Mr Ho Yan Chee, Mr Wong Chian Long, Mr Simon
Tan Suan Beng, and Mr Lee Chiang Soon, for their administrative and
technical help with the project.
I appreciate the work done by FYP, undergraduates and JC students
that in many ways help this research project and the group. I would especially
thank Low Minyi Cindy, Chock E-Wei, V.M. Ajay, and Ye Chenhao for their
contribution. Finally, I would like to thank my family and my friends for their
support and encouragement.
ii
TABLE OF CONTENT
DECLARATION PAGE...............................................................................I
ACKNOWLEDGEMENTS........................................................................ II
TABLE OF CONTENT ............................................................................ III
SUMMARY .............................................................................................. VII
LIST OF TABLES ..................................................................................... IX
LIST OF FIGURES .................................................................................... X
LIST OF ABBREVIATIONS ................................................................XIII
LIST OF SYMBOLS .............................................................................. XIV
CHAPTER 1 INTRODUCTION ................................................................ 1
1.1 Overview ........................................................................................... 1
1.2 Background....................................................................................... 2
1.2.1 Serial Architecture .................................................................. 2
1.2.2 Parallel Architecture............................................................... 3
1.2.3 Hybrid Architecture ............................................................... 4
1.3 Organization ..................................................................................... 6
1.4 Objectives of the Study .................................................................... 8
CHAPTER 2 LITERATURE REVIEW .................................................. 10
iii
2.1 Kinematics ...................................................................................... 10
2.2 Workspace and Singularities ........................................................ 11
2.3 Calibration and Accuracy ............................................................. 14
2.4 Motion Planning and Redundancies ............................................ 16
2.5 Dynamics and Control ................................................................... 18
2.6 Stewart Platform for Machining Applications ............................ 20
CHAPTER 3 COMPUTER NUMERICALLY CONTROL MACHINE
TOOL CONCEPTS ................................................................................... 23
3.1 Part Geometry Design.................................................................... 25
3.2 PKM-based Machine Tool Advantages........................................ 27
CHAPTER 4 THE COOPERATIVE MANIPULATORS DESIGN AND
IMPLEMENTATION ............................................................................... 29
4.1 Cooperative Manipulators Structure Description ...................... 29
4.2 Coordinate Systems and Kinematics ............................................ 32
4.3 Components and Control System ................................................. 35
4.3.1 Tool Stewart Platform .......................................................... 35
4.3.2 Table Stewart Platform ........................................................ 38
4.3.3 Design Consideration ............................................................ 41
4.3.4 Joints Location ...................................................................... 45
4.3.5 Frame Design ......................................................................... 48
4.4 Single Stewart Platform Configuration ....................................... 48
iv
4.5 Extended Configuration ................................................................ 51
CHAPTER 5 SIMULATION AND CONTROL OF STEWART
PLATFORM ............................................................................................... 53
5.1 Workspace Analysis and Kinematic Constraints ........................ 53
5.2 Stewart Platform User Interface................................................... 58
5.3 Programming .................................................................................. 62
5.4 Numerical Control Post-Processor for Stewart Platform .......... 68
5.5 Stewart Platform Motion Emulation and Dynamics .................. 71
CHAPTER 6 IMPLEMENTATION OF THE COOPERATIVE
MANIPULATORS AS MACHINE TOOL ............................................. 73
6.1 Coordinate Mapping of the Cooperative Manipulators ............. 73
6.1.1 Single Stewart Platform Configuration .............................. 74
6.1.2 Extended Configuration ....................................................... 80
6.2 Extended Configuration Motion Planning .................................. 88
6.2.1 Jacobian Matrix and Condition Number ........................... 89
6.2.2 Optimization Procedure ....................................................... 92
6.2.3 Straight-line Milling.............................................................. 94
6.3 Stewart Platform Machining Framework with CAD/CAM
Software .......................................................................................... 98
6.3.1 Tool Path Post-processing .................................................. 102
6.3.2 Determining Machine Origin ............................................. 106
6.4 Machining Case Studies............................................................... 107
v
6.4.1 Machining an ‘NUS’ Pocket ............................................... 107
6.4.2 Machining a Dome .............................................................. 109
6.4.3 Machining a Test Part ........................................................ 113
6.4.4 Machining with Rotation Axes .......................................... 117
CHAPTER 7 STEWART PLATFORM MACHINING
OPTIMIZATION AND EVALUATION ............................................... 120
7.1 Machining Workspace Analysis ................................................. 120
7.2 Application of workspace data for optimal setup in
machining ...................................................................................... 124
7.3 Calibration and Accuracy Improvement ................................... 125
7.3.1 Perpendicularity of Dial Gauges ........................................ 134
7.3.2 Pose selection for Calibration ............................................ 137
7.3.3 Online Calibration for Kinematic Parameters Error
Compensation ...................................................................... 142
7.4 Machining Evaluation.................................................................. 144
7.5 Stewart Platforms Evaluation ..................................................... 151
CHAPTER 8 CONCLUSIONS AND RECOMMENDATIONS ......... 154
8.1 Conclusions ................................................................................... 154
8.2 Research Contributions ............................................................... 156
8.3 Future Work ................................................................................. 158
BIBLIOGRAPHY .................................................................................... 160
PUBLICATIONS FROM THIS RESEARCH ...................................... 177
vi
SUMMARY
While very large and heavy duty machines are still needed for high
volume mass production, there is a growing need in today’s manufacturing
for lighter production machines with smaller size and mass to increase the
efficiency in certain sectors that produce low volume customized products.
This research investigates the application of Parallel Kinematic Manipulators
(PKM), namely Stewart platforms, for such manufacturing applications
especially for machining and positioning. PKMs have inherent properties for
machining applications, but the main constraint of PKMs is the limited
workspace. In this study, cooperative manipulators comprising a
configuration of two Stewart platforms is built. The two Stewart platforms
interact with one another. One of them carries the tool and the other one holds
the object. This approach increases the flexibility of the cooperative
manipulators to handle multi-axis machining jobs and enables the cooperative
manipulators to achieve larger workspace and wider tilting ranges.
The scope of this research includes the modelling of the Stewart
platforms, design methodology for optimal geometric parameters, test of a
prototype for error compensation and an analysis of the machining results.
The motion control input is implemented with translation from standard G-
codes such that a commercial CNC software can be used. An optimization
strategy is developed to solve extra degrees of freedom with objectives related
vii
to the characteristics of the Stewart platforms. Development and results of the
cooperative manipulators is presented.
viii
LIST OF TABLES
Table 4.1 Search range for the dimensional synthesis ................................. 44
Table 4.2 Stewart platform joint locations (in mm) ..................................... 46
Table 5.1 NC codes and their functions ....................................................... 69
Table 5.2 G-codes used for Stewart platform and their meaning ................ 71
Table 6.1 Values used in the example.......................................................... 95
Table 6.2 Trajectory Planning Result Summary .......................................... 95
Table 7.1 Workspace Volume with various tilt angles of the tool axis ..... 124
Table 7.2 Real model for calibration simulation........................................ 133
Table 7.3 Error comparison of calibration simulation ............................... 133
Table 7.4 Kinematic Parameters after Calibration ..................................... 134
Table 7.5 Error comparison of calibration with measurement errors ........ 136
ix
LIST OF FIGURES
Figure 1.1 The first octahedral hexapod or the Gough Stewart platform ...... 4
Figure 1.2 The Logabex robot LX4 (courtesy of Logabex Company) .......... 6
Figure 1.3 Operational model of hybrid robotic arm ..................................... 6
Figure 4.1 Two Stewart platforms ............................................................... 31
Figure 4.2 Extended Configuration of the cooperative manipulators .......... 32
Figure 4.3 Schematic representation of the Stewart platform ...................... 33
Figure 4.4 Tool-SP (not installed in the frame) ........................................... 36
Figure 4.5 Actuator and passive joints in the tool-SP .................................. 36
Figure 4.6 The tool attached to the moving platform .................................. 37
Figure 4.7 The PC controller card for the tool-SP ....................................... 38
Figure 4.8 The table-SP (outside the frame) ................................................ 40
Figure 4.9 The passive joints of the table-SP............................................... 41
Figure 4.10 Relation ship between joints positions ..................................... 43
Figure 4.11 Dimensional parameter synthesis of Stewart Platform ............. 45
Figure 4.12 Joints location of the tool-SP .................................................... 46
Figure 4.14 Single Tool Stewart platform Configuration ............................ 50
Figure 4.15 The machining table installed in the frame............................... 50
x
Figure 4.16 Machining a clamped work-piece on the table-SP with the tool-
SP ................................................................................................................. 52
Figure 5.1 The flowchart to plot the workspace of the Stewart platform .... 57
Figure 5.2 The tool Stewart platform workspace ......................................... 58
Figure 5.3 User Interface for Stewart platform control................................ 60
Figure 5.4 Point-to-point motion control flowchart ..................................... 65
Figure 5.5 Spline motion control flowchart ................................................. 66
Figure 5.6 Jog motion control flowchart ...................................................... 67
Figure 5.7 Interpolated motion control flowchart ........................................ 68
Figure 5.8 Common NC program format ..................................................... 69
Figure 5.9 A 3D Stewart platform model..................................................... 72
Figure 5.10 Stewart platform model in MatLab Simulink ........................... 72
Figure 6.1 Coordinate system in the single configuration ........................... 75
Figure 6.2 The moving platform position to reach the tool contact point.... 79
Figure 6.3 Coordinate mapping in single configuration .............................. 80
Figure 6.4 Coordinate system in extended configuration ............................ 83
Figure 6.5 Coordination of the tool-SP and the table-SP to reach the input
coordinate (the tool contact point on the work-piece) ................................. 84
Figure 6.6 Rotation sequence in the tool-SP ................................................ 87
Figure 6.7 Rotation sequence in the table-SP .............................................. 87
xi
Figure 6.8 Coordinate mapping in extended configuration ......................... 88
Figure 6.9 Algorithm for motion planning in extended configuration......... 94
Figure 6.10 Resulting trajectory plan qz from the optimization procedure .. 96
Figure 6.11 A smooth trajectory from the algorithm with extra constraint;
showing: ....................................................................................................... 98
Figure 6.12 Steps to machining with the cooperative manipulators .......... 100
Figure 6.13 Fixture for holding work-piece on top of the tool-SP ............ 100
Figure 6.14 Information flow of part design and NC code generation ...... 102
Figure 6.15 Machining Input Parameters ................................................... 102
Figure 6.16 Additional information in a G-Code file that cannot be
processed by the MatLab post processor (blue codes) ............................... 103
Figure 6.17 Block Numbers removal in the NC code ................................ 104
Figure 6.18 Testing feasibility of the resulting trajectory, (a) Inaccessible
trajectory (b) Partially accessible trajectory (c) Accessible trajectory....... 105
Figure 6.19 Determining the machine origin ............................................. 107
xii
LIST OF ABBREVIATIONS
PKM Parallel Kinematic Manipulator
SP Stewart platform
DOF Degree Of Freedom
CNC Computer Numerically Control
LSF Least Squares Formulation
TLSF Total Least Squares Formulation
PD Proportional Derivative
PID Proportional Integral Derivative
CAD Computer-Aided Design
CAM Computer-Aided Manufacturing
FKM Forward Kinematic Model
IKM Inverse Kinematic Model
CMM Coordinate Measuring Machine
PSO Particle Swarm Optimization
CC Cutter Contact
CL Cutter Location
xiii
LIST OF SYMBOLS
XP ,YP ZP Coordinate axes of the Stewart platform moving platform
XF ,YF ZF Coordinate axes of the Stewart platform base
ai Platform attachment joints, ball-socket joints, i = 1,2…6
bi Base attachment joints, universal joints, i = 1,2…6
{F} Coordinate frame attached to the Stewart platform base
{P} Coordinate frame attached to the moving platform
{W} Coordinate frame attached to the machining table
li Leg vector
Li The ith leg length
LOi The ith leg offset
R Rotation matrix
q The position vector of {P} relative to {F}
θ The vector of orientation angle of {P} relative to {F}
S Sine
C Cosine
Xɺ Vector of Cartesian velocities of the moving platform
lɺ Vector of leg velocities
X Vector of generalized coordinates of the moving platform
XP A pose set for calibration
Y Measured variables
η A vector containing all kinematic parameters
xiv
Fi The objective function for the ith pose
Q A vector of computed nominal leg length
J The Stewart platform Jacobian matrix
JP Identification Jacobian matrix
O Observability index (Noise Amplification Index)
calculated from JP
σ Singular values matrix JP
κ Condition Number
Ra Radius of the moving platform of the SP
Rb Radius of the base of the SP
α Angle between two closest joints on the moving platform of
SP
β Angle between two closest joints on the base of SP
xv
CHAPTER 1
INTRODUCTION
1.1 Overview
Currently, robots are applied in a variety of manufacturing
applications, such as machining, welding, polishing, assembly, pick and place,
etc. This has triggered the accelerated development and applications of
robotic manipulators in manufacturing. There are two common classes of
industrial applications of robotic manipulators, namely, serial and parallel
kinematic manipulators. Serial manipulators have open kinematic chain and
parallel kinematic manipulators (PKMs) have closed structure of links and
joints. This thesis improves upon the development of a specific type of PKMs,
which are also known as Stewart platforms (SP) for positioning and
machining applications, and investigates the control and user interface aspects.
PKMs have not received much attention as compared to the serial
counterparts probably because of the complexities due to their limited
workspace, control and their singularity characteristics, which sometimes can
occur within the workspace and need to be avoided at all cost. Such problems
are seldom found in conventional industrial serial manipulators. Nevertheless,
this does not stop researchers from developing new strategies to work with
PKMs and come up with new methods that could bring PKMs to their full
potential. Therefore, this research proposes a combined structure or a
cooperative architecture that consists of two PKMs to study the effect of extra
degrees-of-freedom (DOFs) and investigate whether this cooperative
1
architecture could bring additional benefits to the applications of PKMs in
manufacturing and other industrial fields.
1.2 Background
The invention of the first robotic manipulator has triggered
development in many research and industry fields, such as satellite
positioning, underwater explorations, medical operations, flight simulators,
etc. However, there are actually three basic robot architectures, namely as
follows:
1. Serial architecture
2. Parallel architecture (PKMs)
3. Hybrid architecture
These three architectures are classified based on the basic structure of
a robot’s kinematic properties, which are closely related to the sequence and
the arrangement of the joints and links in a robot manipulator.
1.2.1 Serial Architecture
Serial manipulators usually consist of at least two and up to a
maximum of eight rigid links and joints with some prismatic and revolute
joints which can be passive and active (actuated). The main advantage is their
large workspace resulted from their first long links from the base and
followed by wrists with three or fewer DOFs connected to an end-effector or
a tool suitable for a specific task. The serial architecture also tends to have
higher dexterity. Their weaknesses are mainly their limited payload, low
2
precision due to the bending force over the long links connected in a serial
manner, and the large number of parts leading to high inertia which is
undesirable for high bandwidth motion control. The high inertia disadvantage
prevents the use of serial robots for applications requiring high accelerations
and agility, e.g., flight simulation and rapid pick and place tasks.
1.2.2 Parallel Architecture
Parallel kinematic manipulators are built from a series of closed
kinematic chains. A Stewart platform or Gough-Stewart platform, the first
PKM, is composed of six variable struts that are driven by prismatic actuators,
connected to a fixed rigid body and a moving platform which position and
orientation can be changed based on the lengths of the struts, as shown in
Figure 1.1 (Gough and Whitehall 1962). Stewart suggested that the structure
be used for high payload applications because every actuator is located to the
base, reducing the inertia (Stewart 1965). Compared to serial manipulators,
PKMs have higher payload-to-weight ratio, higher stiffness, and higher
precision due to their structure in which errors in each link do not add up to
be transferred to the end-effector. On the other hand, the disadvantages of the
PKMs include difficulty in the control strategies, complicated direct
kinematics, inconsistent performance over the workspace, and the occurrence
of singular configurations.
3
Figure 1.1 The first octahedral hexapod or the Gough Stewart platform
1.2.3 Hybrid Architecture
To overcome those problems discussed in the last section, researchers
have explored the combination of serial and parallel structures to form hybrid
structures, in order to combine the advantages of serial and parallel structures
and complement the drawbacks of each structure. Promising results have been
reported, e.g., the Logabex LX4 robot (Figure 1.2) or the robotic arm
designed at the California Institute of Technology (Figure 1.3) (Tanev 2000).
These manipulators consist of identical parallel mechanisms piled up, and
possess a large workspace and a good ratio of load capacity/manipulator mass.
Furthermore, some hybrid manipulators have already been used in
applications, e.g., deep-sea mining, machining, and medical and assembly
operations (Chai and Young 2001; Callegari and Suardi 2003; Zheng et al.
2004; Carbone and Ceccarelli 2005; Harib et al. 2007). Hybrid structures are
usually designed such that one platform performs pure translation and the
other performs orientation so as to simplify the control algorithm (Lallemand
4
et al. 1997; Tsai and Joshi 2002). However, it is still uncertain whether this
option will solve the problems and bring forth the full potential of both types
of structures into practice. Therefore, this research may contribute to the
literature on this class of hybrid manipulators.
A study on the performance comparison between serial and parallel
structures has been conducted (Geldart et al. 2003). The result showed that
the particular parallel kinematic manipulators outperformed the conventional
machining centres while cutting hard material. In addition, a comparison of
variations of Gough Stewart platform can be found in the literature (Weck
and Staimer 2002; Schwaar et al. 2002), of which some of the structures
comprise hybrid architecture. Although some conceptual and practical
industrial works have been already done for the hybrid architecture, this thesis
focuses on a cooperative scheme of two PKMs. There are several successes
in the past where variants of PKMs are developed for specific applications
(Terrier et al. 2005; Refaat et al. 2007; Neuman 2006). The Gough Stewart
platform manipulator is selected in this research, because it is simpler to build
with modular components, and therefore can lead to lower cost.
5
Figure 1.2 The Logabex robot LX4 (courtesy of Logabex Company)
Figure 1.3 Operational model of hybrid robotic arm
1.3 Organization
Stewart platform mechanisms are less intuitive to evaluate than
conventional serial mechanisms. Design and analysis must be performed
using models and simulation tools. The research on the Stewart platforms in
this thesis is addressed in several phases. First, a kinematic model and 3D
solid model are built to analyse the motion of the Stewart platforms. Second,
a post-processor to translate machining tool paths obtained from CAD/CAM
into the Stewart platform trajectory is developed. Third, basic and complex
programs are run to test the Stewart platforms, where parts are designed and
fabricated. Lastly, tools and methods for process planning, machine
simulation, operation, calibration and redesign are explored.
6
Since there have been much research on several aspects of PKMs,
there is a considerable amount of background literature on these topics.
Chapter 2 is devoted to a brief review of the concepts and results from
relevant literature. Chapter 3 introduces computer numerically control (CNC)
concepts that are used in relation to the configuration of the Stewart platform
as a machine tool. It reviews the general process by which ordinary CNC
machines are operated and the functional requirements of the Stewart
platform as a machine tool.
Chapter 4 focuses on aspects related to the hardware of the Stewart
platforms being investigated in this research work. Chapter 5 presents the
software aspect of the Stewart platforms including the simulation and
computation tools developed for controlling the Stewart platforms, providing
graphical interface and characteristic analysis that are useful for future design.
Chapter 6 explains the crux of the motion planning algorithm for the
proposed configuration with a commercial CAD/CAM system in order to
operate the Stewart platforms to execute various machining tasks. This
chapter explains how redundancy introduced in the cooperative manipulators
with two Stewart platforms can be used to plan the optimal motion path for a
given tool path trajectory. In addition, machining case studies which have
been executed with the proposed Stewart platform are presented. From these
cases, comparisons are made based on the machining results of a single
Stewart platform and the cooperative configuration consisting of two Stewart
platforms. Observations are made based on the experiments performed and
7
are used to evaluate tool path generation, work-piece setup, user experience,
and future design considerations.
In Chapter 7, several improvements in the workspace and accuracy of
the Stewart platform are reported. The error model and calibration of the
Stewart platforms to compensate the inaccuracies caused by assembly and
manufacturing errors is presented.
Chapter 8 concludes the investigation of the Stewart platforms
configuration and application. It also summarizes the results and suggests
areas where further work is recommended.
1.4 Objectives of the Study
Stewart platforms have several potential applications. With respect to
this research, one of their uses is in a flexible manufacturing environment. In
principle, the end-effector can be positioned in any way that is required for
the respective task, e.g., milling, welding, cutting and assembly. In particular,
the goals of this thesis are:
1. To investigate the integration of two Stewart platforms (six DOF and
three DOF) or (six DOF and six DOF) to form a nine or twelve DOF
system. For example, one of the SPs can be used to locate and hold a
work-piece, and another Stewart platform can be used to hold a cutting
tool or some other measuring devices.
2. To explore the development of user interfaces that can be used to
control two Stewart platforms simultaneously to plan for the
necessary machining paths and avoid any collision. The two Stewart
8
platforms can move together such that the time to reach the final
position will be shortened.
3. To obtain the work volume of the two Stewart platforms and provide
calibration and feedback control of the two coordinated Stewart
platforms to compensate for any inaccuracies in movements and final
positions.
4. To carry out case studies to study multi-axis machining operations.
Due to the restriction of the movements of the coupled SP system, it
will be necessary to explore the type of work-piece geometry that can
be handled in a single set-up. A sub-objective here is to explore, given
a particular work-piece model in 3D, the accessible and inaccessible
features in a particular set-up and optimize the orientation of the part
such that the total number of set-ups required is a minimized.
9
CHAPTER 2
LITERATURE REVIEW
Most of the research on PKMs deals with conventional robotic issues,
such as kinematics, singularities, dynamics, workspace, calibration accuracy
and structural properties (Merlet 1999; Dasgupta and Mruthyunjaya 2000).
Few research studies have been reported on motion planning, control and
robot design or synthesis. Hybrid manipulators have received attention in
these areas (Zhang et al. 2005). In this chapter, some key issues in this field
will be reviewed, although not exhaustively.
2.1 Kinematics
The kinematics of SP mechanisms, like all robotic manipulators in
general, is a study of the geometry of the motions of the end-effector and the
actuating joints, and the relationship between these two types of motions
without consideration of the torques and forces that cause these motions. The
inverse kinematics problem, i.e., to find the lengths of the links for a given
position of the moving platform, is quite straightforward for PKMs. On the
contrary, direct kinematic problem has to be solved using numerical methods.
For a general SP, 40 assembly modes (i.e., direct model solutions) can exist
(Dietmaier 1998). In practice, the use of numerical procedures has been
proposed, which assume that an estimated solution is known (Nguyen et al.
10
1991; Parikh and Lam 2005; Wang 2006). Another method is to use a larger
number of sensors than the number of DOFs so that additional information
can be used to improve the direct kinematics algorithm (Cheok et al. 1993;
Parenti-Castelli and Di Gregorio 1995; 2000; Chen and Fu 2006). It has been
shown that the computation of forward kinematics is more efficient with an
additional off-line pre-processing phase (Tarokh 2007).
2.2 Workspace and Singularities
Singular configurations are particular poses of the end-effector or
moving platform of the PKM for which the manipulator loses its inherent
rigidity, and the end-effector has uncontrollable degrees of freedom. At
singularity positions, the joint velocities may be unbounded although the
linear velocity and the angular velocity of the robot arm are bounded. The
occurrence of singular configurations is highly undesirable in PKMs since in
these configurations, the actuators cannot control the mechanisms, which gain
additional finite or infinitesimal freedom and variation in stiffness (Merlet
1992). There are two Jacobian matrices J for a PKM (Gosselin and Angeles
1990b), i.e., one for the inverse kinematic and one for the direct kinematic.
This yields three types of singularities:
1. det ( J dir .kin. ) = 0 and det ( J inv.kin. ) ≠ 0
2. det ( J inv.kin. ) = 0 and det ( J dir .kin. ) ≠ 0
3. det ( J dir .kin. ) = 0 and det ( J inv.kin. ) = 0
Depending on which of the two matrices are singular, a PKM may be
at a direct kinematic singular configuration, an inverse kinematic singular
11
configuration, or both. In addition, another type of singularity exists, namely,
architecture singularities (Ma and Angeles 1991a). This type of singularity
arises from the symmetrical architecture of the SPs causing singularity poses
over a significant portion of the entire workspace. Some research focuses on
the characterization of the singularity of SPs (Hunt 1978; Fichter 1986; St-
Onge and Gosselin 2000), such as detecting singularities in a given workspace
(Merlet 2007), and a numerical procedure for avoiding singularities of a SP
by restructuring a pre-planned path in the vicinity of a singularity
(Bhattacharya et al. 1998). In addition, a more recent method for measuring
closeness to singularity is by using physical properties, such as the stiffness
and torque transmission of the PKM, which better capture all the singularity
configurations (Voglewede and Ebert-Uphoff 2005). Furthermore, a more
complete analysis of singularity configurations can be done through
acknowledging that they are configurations of singularities that cannot be
simply detected by computing the Jacobian matrix. This type can be detected
only if proper input-output velocity analysis is used (Zlatanov et al. 2002;
Han et al. 2002).
A main drawback of PKMs is their limited workspace. There are three
main mechanical constraints that restrict the workspace of PKMs, specifically,
the actuators’ stroke, the range of the passive joints, and the link interference
(Bonev and Ryu 2001). The workspace of a manipulator is defined as the set
of all the end-effector configurations that can be reached. Various methods to
determine the workspace of a PKM have been proposed using geometric or
numerical approaches. Since the translational and orientation workspaces of
12
a SP are coupled, a first approach is to fix the values of a few of the DOFs
until only three DOF are free, so that it can be represented in a 3D plot. Most
often, the 3D constant orientation workspace, which describes the possible
location of the origin of the end-effector for a constant orientation, is of
interest. A geometrical approach (Gosselin 1990b) that has been reported
gives the best result as it provides an exact calculation with a compact storage
and easy representation. However, numerical methods (Du Plessis et al. 2001)
are also preferred as they can deal with joint limits and workspace verification
problems (Masory and Wang 1995) (i.e., to determine whether a part of the
workspace is reachable).
One application of workspace analysis lies in the field of machine
tools, where only five DOFs are required for completing a task. Workspace
analysis for PKM based machine tools has been reported (Wang et al. 2001).
Huang (Huang et al. 1999) showed that the minimum reachable yaw angle
for a given point may be calculated exactly when the constraints on the
passive joints are modelled using a cone. Another aspect of workspace
analysis for machine tools is part positioning, i.e., given a machining
operation to be performed, the problem is to determine the positioning of the
part that the machining trajectory will lie within the workspace. This problem
has been addressed using the discretization approach (Pugazhenthi et al.
2002).
13
2.3 Calibration and Accuracy
PKMs were introduced due to their higher accuracy as compared to
conventional robots and better stiffness in the same range as the machine tools.
Due to the complicated kinematic chains in a PKM, it is difficult to achieve
the required accuracy. Generally, the error sources of a SP-based machine
tool can be classified into geometric and non-geometric. Geometric errors are
errors in the parameters that define the geometric relationships. Other sources
of errors are grouped as non-geometric errors. In the accuracy or error
analysis, it is necessary to develop a valid error model through examining the
main sources of errors and investigate the relationships between the errors of
the joints and those of the end-effector. Masory (Masory et al. 1997) has
studied the influence of the sensor errors and the manufacturing tolerances on
the locations of the joint centres. A more thorough analysis has been proposed
by Ehmann et al. (Petal and Ehmann 1997; Wang and Ehmann 2002), which
includes the location errors of the passive joint centres, errors in the leg
lengths, and the imperfect motions of the ball joints. Tischler and Samuel
(Tischler and Samuel 1998) proposed a numerical approach for determining
the influence of the backlash of the joints, while Meng (Meng and Li 2005)
and Wolhart (Wohlhart 1999) proposed an analysis of the effect of the joint
clearances on the trajectories followed by serial and parallel manipulators.
Other sources of errors, such as thermal errors, gravity induced errors, and
dynamic errors (Pritschow et al. 2002, Niaritsiry et al. 2004; Clavel 2005),
have also been studied.
14
Geometric errors, sometimes called the kinematic errors, can be
reduced through kinematic calibration, which deals with the improvement of
a kinematic model of a manipulator that is attainable through substituting the
nominal values of the kinematic parameters with their actual values. In
kinematic calibration, various methods have been suggested, e.g.,
optimization methods (Zhuang and Roth 1993), linearization method (Geng
and Haynes 1994), and partial differentiation (Ropponen and Arai 1995).
Merlet (Merlet 2006) distinguished three main types of calibration methods:
1. External calibration: An external measurement device is used to
determine (completely or partially) the real pose of the platform for
different desired configurations of the moving platform. The
differences between the measured pose and the desired pose give an
error signal that is used for the calibration
2. Constrained calibration: Methods that rely on a dedicated
mechanical system that constrains the robot motion during the
calibration process.
3. Auto-calibration or self-calibration: The platform has extra sensors
and only the manipulator measurements are used for the calibration.
In this case, it is required that an n-DOF robot has m (> n) internal
sensors.
In addition, there is another group of calibration methods which uses
interesting geometrical properties. Huang (Huang et al. 2005) proposed using
specific motion characteristics, e.g., flatness and straightness which can be
measured easily using dial gauges. Takeda (Takeda et al. 2004) proposed
15
using a double-ball-bar measuring device. In his study, the robot performed
circular paths, and the deviation from circularity was measured using the
device. In the machining field, calibration can be conducted using machining
experiments (Chanal et al. 2007). Recent research shows a trend using camera
calibration that can produce good accuracy with relatively low cost (Andreff
et al. 2004; Dallej et al. 2006a; Daney 2006; Renaud et al. 2006; Tanaka
2006).
2.4 Motion Planning and Redundancies
Motion planning is a classical problem for serial manipulators to avoid
obstacles. However, for parallel manipulators, more factors have to be
considered, such as limited workspace, singularities, and other performance
requirements. Merlet (Merlet 1994) presented a method for checking whether
a trajectory lies within the workspace of a manipulator. Harris (Harris 1995)
dealt with motion planning between two poses by looking for the parameters
of the screw motion linking the two poses, and reckoning that this motion
should be able to minimize the changes in the link lengths. Gosselin and
Angeles (Gosselin and Angeles 1990c) presented an algorithm that can find
the orientation of the manipulator with the best accuracy in some specific
poses along a path based on the condition number. Recently, probabilistic
path planning has emerged as one of the most promising approaches to path
planning of manipulators with large DOFs. A most prominent research in this
field for parallel manipulators is the probabilistic roadmap approach (Cortes
and Simeon 2003). However, this approach does not consider singularity or
16
multiple solutions for the direct kinematics, which may prohibit the use of the
trajectory.
Motion planning for machine tools may present some specificities, as
fewer than the number of DOFs of the machine may be used, e.g., for a SP,
the rotation about the tool can be ignored as it does not have to be specified
for machining tasks. Therefore, it is possible to determine the ranges for the
free DOFs to ensure that a given machining trajectory lies within the
workspace and apply an optimization procedure on the free DOFs to optimize
other performance criteria for the SP (Merlet 2000). Another approach (Chen
et al. 2003) partitioned the DOFs into critical and secondary DOFs, and
synthesized a control law that ensured the tracking of the critical DOFs while
minimizing a velocity-based secondary criterion.
Redundant manipulators are of significant importance because of their
advantages when task versatility and manipulator performances are required.
Non-redundant manipulators, serial or parallel, perform well over a certain
range of task operations corresponding to the limitations of their structural
and actuation characteristics. Redundant manipulators possess ‘additional
inputs’ that offer a means to improve their performance and increase their
versatility. Pierrot (Pierrot 2002) distinguished three different types of
redundancies:
1. Kinematic redundancy: At least one of the legs is a motion generator
with a larger number of DOFs than necessary. This may be used for
enlarging the workspace (Liu et al. 2001).
17
2. Actuation redundancy: The end-effector is over constrained by the
actuators. The number of actuators is more than the number of DOFs.
Such redundancy is mostly used for singularity avoidance (Wang and
Gosselin 2004).
3. Measurement redundancy: The number of sensors is larger than the
number of actuated joints. This redundancy plays a role in solving the
forward kinematic problem to reduce the positioning errors and for
calibration (Marquet et al. 2002).
When two SPs are combined, there is one redundancy caused by the
two SPs working together simultaneously. Thus, the main problem is to
determine an optimum use of this redundancy. A machining path for
cooperative manipulators consisting of these two SPs can be generated only
when redundancy has been resolved.
2.5 Dynamics and Control
Dynamics is the determination of the relationship between the
generalized accelerations, velocities, and coordinates of the end-effector and
the joint forces. Dynamic analysis of PKMs is complicated by the existence
of multiple closed-loop chains. The earliest discussion on computing the
dynamics of SPs can be found in research work by Fichter (Fichter 1986),
which is applicable when the leg inertia and the joint friction are negligible.
As the SPs became better known, there were three major approaches of
computing the dynamics, namely, the Newton-Euler formulation (Codourey
and Burdet 1997; Dasgupta and Choudhury 1999; Harib and Srinivasan 2003),
18
the Lagrangian formulation (Nguyen and Pooran 1989; Geng et al. 1992; Liu
et al. 1993), and the principle of virtual work (Wang and Gosselin 1998; Tsai
2000; Gallardo et al. 2003). Some researchers (Reboulet and Berthomieu
1991; Kim and Lee 1992; Kock and Schumacher 2000) concluded that the
dynamics model needs to be simplified in order to be used in a real-time
control system. Different methods can be applied depending on the situation
and the requirements, i.e., the purpose of evaluating the dynamics, whether it
is for control, evaluation or simulation purposes.
Control of the SP manipulator is still an open issue and the works
reported are not rigorous. In the field of machine tools, the trend is to try to
adopt existing hardware for controlling the PKMs. However, the use of
existing hardware for controlling the manipulators will drastically penalize
the performance of the system in the long term (Merlet 2002). Some
researchers have suggested that each actuator can be controlled independently
and robustly with a control law than a simple proportional-integral-derivative
(PID) control system (Chiacchio et al. 1993). Another approach implemented
an optimization scheme on top of a proportional-derivative (PD) control (Yurt
et al. 2002). Wang (Wang et al. 1995) and Zheng (Zheng and Haynes 1993)
presented a neural network control scheme and showed its superiority over
kinematic control. A model reference adaptive control scheme has been
proposed (Li et al. 2003) to control a machine tool, and the Popov hyperstable
theory is utilized as the adaptive control law. Recently, a more advanced
tracking control scheme has been proposed (Huang et al. 2004; Huang and
Fu 2004) and feedback using a camera (visual-servoing) has been
19
implemented (Zuo et al. 2002; Dallej et al. 2006b; Andreff et al. 2007). Lastly,
the combination of more than one single control strategies that takes
advantage of multiple coordinated PKMs is another important field that is
relatively unexplored.
2.6 Stewart Platform for Machining Applications
PKMs have several advantages over conventional industrial
manipulators and machine tools. The main objective is to find applications
where the PKMs can be best utilized according to their capability. These
include applications where flexibility, accuracy and high loads are essential
for success. Most of the previous research studies were focused on issues,
such as kinematics, dynamics, singularities, workspace, etc. Relatively little
effort has been focused on investigating the implementations of PKMs in the
industry.
The trend in the manufacturing industry is towards shorter product life
cycles and a larger variance of products. For example, the automotive
industry is producing smaller batches and uses more common platforms and
components. This increases the need for greater flexibility within the
manufacturing systems and reconfigurable systems.
The cooperative manipulators proposed and developed in this research
serves as a test-bed and a vehicle for exploring the characteristics of
cooperative manipulators. The unique geometric structure of this cooperative
manipulators is expected to confer a number of important performance
advantages. Taken altogether, these advantages could revamp precision
20
manufacturing operations and stimulate new approaches to designing and
machining parts, moulds and dies.
Currently, it is not known how well this proposed cooperative
manipulators will perform as compared to conventional machine tools.
Currently, the research is focused on investigating the attributes and
limitations of this cooperative manipulators. After an initial characterization
of this cooperative manipulators, the research focus is then expanded to
techniques for enhancing the performance of the cooperative manipulators in
machining, positioning and assembly applications. A main objective of this
research is to develop the underlying measurement methods and technologies
needed to achieve high levels of positioning accuracy and resolution. Micro
actuators will be incorporated into the cooperative manipulators so that the
strut lengths can be changed in precise micrometre-scale increments.
Eventually, a system for self-calibration will be developed so that the
cooperative manipulators can check its own performance and correct any
detected inaccuracies.
Although PKMs are considered one of the most radical innovations
since Computer Numerical Control (CNC) was found, a better understanding
of the real advantages offered by PKMs versus conventional machining
centres is still an on-going research issue. Some research efforts on the
comparison of conventional machining centres with PKMs have not found
any good standardization to compare them on an equal basis (Tlusty et al.
1999; 2000; Fassi and Wiens 2000; Neugebauer et al. 2000). Therefore, one
would try not to develop a manipulator to compete with conventional machine
21
tools on accuracy and stiffness but rather on the flexibility of accomplishing
three- and five-axis machining with respect to the development of a general
PKM. The expected performances of the cooperative manipulators with
respect to flexibility are:
1. Fewer number of setups are required,
2. Larger workspace volume, especially the range of orientation, and
3. Higher speed as a shorter processing time can be achieved with two
platforms moving together.
22
CHAPTER 3
COMPUTER NUMERICALLY CONTROL
MACHINE TOOL CONCEPTS
A CNC machine tool is positioned according to a pre-programmed
path by means of special codes forming an NC program; an NC program
consists of commands represented by letters, numbers and special symbols.
These commands are used to manipulate the machine tool or work-piece to
produce a required industrial part. Nowadays, almost all machining tasks are
generated using computer-aided design (CAD) part design followed by
computer-aided manufacturing (CAM) to process the solid model so as to
obtain the tool paths required to cut the material. Modern CAD and CAM
software are able to reflect changes in a part design almost instantly in the
part program. This allows late changes to be included in the production cycle,
but without interfering with the entire design process from the beginning.
The NC program controls the machine movements following a certain
manufacturing technology and methodology. Thus, a CAD/CAM software
package often comes together with standardized machine tools libraries that
can generate certain set of instructions compatible to a particular selected type
of CNC machine tool. Moreover, they develop speeds and feeds data
automatically based on tool selection. The NC program is able to output the
required shape and size of the tool, the speed and feed rate, and the orientation
23
of the tool relative to the work-piece. These parameters are different from one
machine to another in terms of their definition and reference. The part
program also prescribes a set of cutter location (CL) points assigned to cut
the desired part. A CL point is a specific position at which an NC machine
has to move a cutter to.
Before an NC program is produced, the CL data file is processed for
a specific machine tool. This process is called post-processing. Thus, the post-
processor of each machine to be used must be present in the CAD/CAM
package. The post-processor has the ability to output the correct syntax for a
particular machine tool and transform coordinate systems with respect to the
specific arrangement within the machine tool. The standard syntax for most
machine tools is known as G-Code and M-Code. The post-processor permits
additional functions or modification to be added according to the machine
controller, as well as variation in the machine capabilities. After the entire NC
program has been generated, it is stored in a file which is then fed to the
controller of the machine tool.
The fundamental function of the controller of a machine tool is to
move the machine tool along a linear and/or circular path interpreted from the
G-Code in the NC file. The controller may employ an interpolation technique
to overcome the limitations of the machine tool drivers, which may cause tool
chatter or breakage. Similarly, positional control is used in the control system
of a SP such that many sampled points are used. The controller of the
cooperative manipulators uses a standard industrial PID control scheme to
move the actuators to certain positions with various speed settings. Fine
24
tuning is performed to obtain the optimum PID parameters to allow actuators
to move at various speeds while maintaining good dynamic performance,
such as faster settling time and lower overshoot. In a coordinative
configuration, both Stewart platforms can move together in order to move the
tool relative to the object with less travel time.
The orientation of the cutting tool or the work-piece in one or two
rotary axes can be changed in addition to spatial motions along the Z, Y and
X axes according to the values specified in the NC program. These rotations
have axes known as a combination of C, B, and/or A axes. Since a SP is
basically a six-legged structure which can perform 6-DOF motions, it has the
potential to be used as a 5-axis or 6-axis machine tool.
3.1 Part Geometry Design
CAD/CAM systems are widely used in industries for designing
products quickly and efficiently. This has resulted in seamless
communication and automated tool path generation. These systems are able
to model the parts, and perform measurements and interference checking with
high precision. They are used as a virtual prototyping tool to evaluate design
requirements and verify errors before the designed part is sent to be machined.
The output of a CAD system is a representation of a part in an
electronic format. The tool path needs to be generated using either the CAD
system or specialized CAM software. If CAM is used, a design must be
transferred to the CAM system each time when there is a design change. This
has the tendency of slowing down the production process. Currently, there is
25
no specific CAM system or post-processor for machine tools based on SP as
SP has not been widely used for machining applications. Therefore, tool path
generation is done using standard machine tool settings. Subsequently, the
tool path has to be converted in order to run the tool path in a SP-based
machine tool.
The procedure used to machine a part using a SP-based machine tool
is described as follows:
1. Generation of the tool path in the work-piece coordinate system. This
step usually requires a successive set of coordinates called the cutter
contact (CC) points and the tool orientations in the work-piece
coordinate system, which are distributed along a set of curves
following a specified pattern, such as the zigzag or the spiral curves.
2. The CL data are computed from the set of CC points. The method for
CC-CL data conversion must include the geometry of the cutting tool
and the orientation of the tool relative to the work-piece. Various
optimization techniques can be used at this step to generate optimal
cutter location data.
3. Each CL point is transformed into the machine command. Converting
coordinates of the CL points from the work-piece coordinate system
to the machine coordinate system is called post-processing. The post-
processor requires knowledge of the machine kinematics and the
machine configuration. In SPs, this is also called inverse kinematics
control.
26
4. The machine commands are converted into an NC part program in a
format that can be interpreted by the machine controller. In this case,
MatLab is used to achieve this.
The above sequence may or may not include optimization and
verification steps. The optimization step involves defining a cost function to
represent a certain type of error or a combination of several types of errors.
The tool path is modified or reconstructed entirely in such a way that the cost
function is minimized or at least decreased. The verification stage includes an
actual machining or computer modelling of the material removal process
which produces the output as a solid model. The solid model can be compared
with the desired part and the efficiency of the optimization strategies can be
evaluated.
In this research, the CL points are converted to NC codes in CAM and
further processing of the tool path is done through analysing the NC codes.
CL points data are actually more useful compared to NC codes as it is more
generic and not specific to a particular machine tool. However, because of
software limitation, the specific commercial CAM software used in this
research cannot export CL points directly to MatLab. Thus, it is necessary to
convert and process the NC codes.
3.2 PKM-based Machine Tool Advantages
A PKM-based machine tool has several potential applications. One
example is its use in a flexible manufacturing environment. Since a PKM is
generally a positioning device, it is possible to use an interchangeable end-
27
effector module to perform a variety of operations, such as drilling, milling,
grinding, welding, and assembly. In addition, PKM machine tools can be
designed to utilize high speed spindles to remove material quickly.
Conventional machine tools, which have a cantilever or gantry type
of structure, have limited payload due to their inherently large moving masses
of their components. This surfaces the needs for powerful actuator systems,
more precise control systems, and larger components for stiffness
requirement. On top of these, thermal shielding is necessary due to the large
amount of heat from the motors and bearings. Designing a PKM machine tool
could lead to lower cost (Weck and Staimer 2002).
A PKM machine tool can have greater dexterity because it minimizes
the moving mass. In operation, PKM only moves its end-effector mass. The
work-piece is clamped onto the end-effector and remains stationary. Higher
acceleration and speeds are achievable. It also minimizes the error on the final
position of the end-effector because relative errors in the actuator positions
are averaged instead of added, which is the case in serial manipulators. PKM
consists of fewer mechanical bodies and each structural linkage is built from
the same components. It is suited for modular design that would reduce the
service and maintenance costs. Finally, the operating cost is lower because of
the lower energy consumption and faster material removal rate (Fassi and
Wiens 2000).
28
CHAPTER 4
THE COOPERATIVE MANIPULATORS DESIGN
AND IMPLEMENTATION
4.1 Cooperative Manipulators Structure Description
Two Stewart platforms, namely, a tool-SP (Figure 4.1a) and a table-
SP (Figure 4.1b), have been built in this research to form cooperative
manipulators to be used as a machine tool. Both Stewart platforms have
similar structures. The spindle is attached to the tool-SP in Figure 4.1a. This
tool-SP, mounted inversely onto a frame, is the main manipulator in this
research, and it is referred to as the tool-SP. The six actuators installed
between the moving platform and the base produce the motion. Universal
joints are used to connect the actuators to the base and ball-socket joints are
used to connect the actuators to the moving platform, which holds the spindle.
The second SP is installed on the frame below the tool-SP. This SP is
named the table-SP. The table-SP works with the tool-SP, and it is thus
referred to as the table-SP in this research. Figure 4.2 presents the
configuration where the tool-SP and the table-SP are both installed on a frame.
In this configuration, the work-piece is clamped onto the moving platform of
the table-SP. The table-SP can move the work-piece relative to the tool-SP to
provide access to locations further on the work-piece which are outside the
workspace of the tool-SP. The table-SP has a larger dimension to provide
29
larger movement range for the work-piece to achieve a larger overall
workspace. This configuration is named the cooperative PKM machine tool.
Base of
Tool-SP
Universal
joint
Leg
actuated
by an
internal
motor
Moving
Platform
Tool
(a) Tool-SP (Inverted installation)
30
Moving
Platform
Ball-
socket
joint
Leg
actuated
by a
motor
Universal
joint
Base of
Table-SP
(b) Table-SP
Figure 4.1 Two Stewart platforms
31
Figure 4.2 Extended Configuration of the cooperative manipulators
4.2 Coordinate Systems and Kinematics
Figure 4.3 shows a schematic of a Stewart platform with its kinematic
designation. A ball-socket joint connects each prismatic actuator to the
moving platform. Similarly, a universal joint connects each actuator to the
base. This arrangement allows the moving platform to move based on the
lengths of the prismatic actuators or the leg lengths.
32
ZP
Moving XP YP P
ai
Platform {P}
li
ZF
Base
YF bi
XF {F}
Figure 4.3 Schematic representation of the Stewart platform
There are two coordinate frames, namely, a fixed or base coordinate
frame {F} and a moving platform coordinate frame {P}. The position vector
bi denotes the position of the centre of the universal joint of leg i in {F} and
the position vector Pai is defined in {P} pointing to the location of the centre
of the ball-socket joint of leg i. The legs are represented by vectors li defined
in {F}. If FRP and q = [qxqyqz]T are the rotation matrix and a position vector
expressing the pose (orientation and position) of {P} relative to {F}, li can be
calculated using Equation 4.1.
li = F RP P ai + q − bi (4.1)
This equation is well known as the inverse kinematic formula. The leg
length can be computed by computing the length of this vector. The values of
the leg lengths must take into consideration the initial length offsets of the
legs LOi, and they are calculated using Equation 4.2.
λi = li − LOi = Li − LOi (4.2)
33
On the other hand, the forward kinematic model that is used to
calculate the pose given the leg lengths is a difficult problem which is usually
solved using numerical methods. The pose can be denoted by a generalized
vector X=[qTθT]T , where θ=[θxθyθz]T describes a set of Euler angles of the
orientation of {P} relative to {F}. A Newton-Raphson scheme is developed
in this research to compute a solution of the forward kinematic problem. This
is an iterative method stated as follows.
1. Given an initial estimated pose Xg = [qgT θgT]T, calculate the inverse
kinematic solution of this pose, which gives the six leg lengths, Lg =
[Lg1Lg2Lg4Lg4Lg5Lg6]T
2. Compute the partial derivative 6×6 matrix J of the pose vector, dX,
with respect to a small change of each leg displacement, where Jj,i =
dXj / dLgi , i=1..6, j=1..6.
3. Update the current pose, Xg = Xg + J-1 * (La – Lg)
4. Repeat steps (2)-(4) until (La – Lg) is less than a certain pre-defined
numerical threshold, and exit with the last Xg as the solution.
The initial pose can be set as the home position of the moving platform.
The algorithm will give a robust solution of the correct pose if the pose is
within the range of the manipulator movement and not near to any singularity
configurations.
In this thesis, the subscript ‘2’ is used to denote all the kinematic
assignments of the table-SP. Therefore, Equations 4.1 and 4.2 can be
rewritten for the table-SP as the follows.
li 2 = F2
RP 2 P 2 ai 2 + q2 − bi 2 (4.3)
34
λi 2 = li 2 − LOi 2 = Li 2 − LOi 2 (4.4)
4.3 Components and Control System
Both the tool-SP and table-SP are built from commercially available
components. The links and connecting parts are fabricated in the workshop at
the NUS Advanced Manufacturing Laboratory.
4.3.1 Tool Stewart Platform
Figure 4.4 shows the tool-SP before it was installed on the frame as a
part of the cooperative manipulators. Figure 4.5 depicts the main components
of the tool-SP. Figure 4.5a shows a Physik Instrumente M-235 series actuator
which is a high resolution linear positioning actuator. It provides a motion of
50 mm with sub-micron resolution in a compact package. It has a preloaded
ultra-low friction, heavy duty ball screw. This ball screw is driven by a closed-
loop DC motor. Figure 4.5b shows the ball-socket joint. It is a HEPHAIST
SEIKO ball-socket joint, SRJ006C. The movable parts of the ball-socket joint
have a very low frictional resistance due to its rolling joint structure under
preload, achieving high precision with little clearance. Compared with other
ball-socket joints of similar characteristics, these joints have higher rigidity
and they are also much smaller in size. They are optimal for achieving high
precision, high rigidity and downsizing of the PKM. Figure 4.5c is a universal
joint with needle bearing. It is designed to maintain low backlash for critical
positioning applications required in robotics and instrumentation and is
35
excellent for continuous operation applications. The joint has rigid axial
stiffness for push/pull loads.
The aluminum
block for
attaching the
tool spindle
Figure 4.4 Tool-SP (not installed in the frame)
(a) (b) (c)
Figure 4.5 Actuator and passive joints in the tool-SP
36
The tool holder is attached to the moving platform. The tool holder is
designed so that the tool height from the moving platform can be adjusted.
There is an aluminium square block of thickness 90mmx20 mm, which is
attached on top of the moving platform using two M10 screws (Figure 4.6).
The spindle is fixed on to the aluminium block. The tool can be moved up
and down through the slot on the aluminium block. Figure 4.6 presents this
arrangement.
Adjustable
tool height
Figure 4.6 The tool attached to the moving platform
The controller for the tool-SP is the PC-based controller cards. One of
the controllers is shown in Figure 4.7. It provides closed-loop digital servo-
control and it comes with incremental encoder position feedback signals. The
37
axes can be programmed either independently or synchronously to allow
multi-axis motion. Power is drawn from the PC power supply. Therefore, the
power of the motor could be limited.
Figure 4.7 The PC controller card for the tool-SP
The PC controller card provides a wide range of functions for
controlling the actuators. Each motor is controlled independently using a
built-in PID controller, and a desired position and velocity can be specified.
The controller card comes with C/C++ library that can be used for
programming. The library can be called from a MatLab function to perform
the operations.
4.3.2 Table Stewart Platform
The table-SP has a similar configuration as the tool-SP. It has larger
actuators and passive joints. Its actuators are driven by servo motors with a
dedicated power supply. Therefore, the power rating is higher than the tool-
SP and the table-SP can move a larger payload. The actuators are also
38
controlled by a PC controller card. The main difference lies on the passive
joint selection in the tool-SP and the table-SP. The table-SP is designed for a
larger workspace and requires passive joints that have high maximum tilting
angle. However, it is always a trade-off in the market between tilting angle
and the joint clearance in the design of passive joints themselves with respect
to price. The joint clearance affects the accuracy of the moving platform
position and orientation, especially for the ball-socket joints. Higher joint
clearance leads to higher errors in the Stewart platform kinematic chains.
Thus, although the table-SP has a larger workspace, the expected accuracy is
lower compared to the tool-SP.
Figure 4.8 shows the table-SP without the frame. The kinematic
chains are the same as the tool-SP. The table-SP is designed for carrying the
work-piece which can be fixed on top of its moving platform using fixture.
The motor at each leg is driving the leg length through a pulley and a ball
screw to convert the rotary motion to linear motion. Such a design allows the
motor to be larger and hence has a higher power rating to drive heavier load.
A high precision encoder is installed on the rotor of each motor to sense the
rotary motion and provide feedback to the controller.
39
Figure 4.8 The table-SP (outside the frame)
The ball-socket joint and the universal joint, which connect the legs
to the moving platform and the base, are shown in Figure 4.9. There are
components that are fabricated in the workshop besides the joints. The passive
joints are selected according to the size and power rating of the actuator. The
ball-socket joints and the universal joints have working angle limits of 35 and
60 degrees, respectively.
40
Figure 4.9 The passive joints of the table-SP
The table-SP legs are actuated by six servo motors controlled by a
SPiiPlus PCI controller card installed on the PC. The controller can perform
various functions and generate motion profile for testing each motor. The
vendor has supplied the program manager by which the servo parameters can
be adjusted. In addition, it provides 10 internal program buffers than can be
activated simultaneously. It also has an extensive C/C++ library that can be
integrated with MatLab.
4.3.3 Design Consideration
Machine design is a complex subject. In this section, the dimensional
synthesis of the Stewart Platform is elaborated. Both SPs have undergone
similar design processes. Design is carried out with emphasis on achieving
maximum 3D workspace. The workspace of a Stewart platform is obtained
using a discretization method (Merlet 2006) and the results with respect to
41
independent variables are plotted so that optimum geometric parameters can
be determined. Mechanical kinematics constraints are also considered in the
simulation so that the limitation and singularity of the movements of the
platform can be predicted without driving the links to their limits. During the
normal operation of the Stewart Platform, the singularity configuration should
always be prevented to reduce the risk of uncertainty of the position and
orientation as well as the possibility of overloading the joints of the platform
that may cause breakdown.
Consider the tool SP which has been introduced in the previous
section. There are four geometric design parameters to be considered. For
simplicity, the main relationship between joints location attached to the base
and to the moving platform is shown in Figure 4.10.
α° Y
α°
120° X
120° 120°
Ra
Legend:
= Joint
attachment ai
α°
(a) Moving platform
42
β°
Y
120° 120°
Rb
120°
Legend:
β° β° = Joint
attachment bi
(b) Base
Figure 4.10 Relationship between joints positions
The base radius is always greater than the moving platform radius to
ensure stability (i.e., Rb ≥ Ra). Design optimization on dimensional synthesis
can achieve certain performance, such as the workspace/footprint ratio. In real
application, however, there are many factors that can affect the design
decision, thus optimization depends on multi-criterion characteristics and
other constraints. In this case, the workspace is computed with varying
parameters of α, β, Ra, and Rb. The approach is to keep a pair of parameters
constant (α, β) and (Ra, Rb) and plot the workspace characteristics with respect
to two variables. The search range is defined for α, β, Ra, and Rb. It can be
shown in Table 4.1.
43
Table 4.1 Search range for the dimensional synthesis
Minimum value Maximum value
α and β 10° 45°
Ra 25mm 75mm
Rb 50mm 100mm
A three-dimensional surface plot is used to depict the resulting
simulation. Figure 4.11 shows the 3D positional workspace volume of the SP
with each pair of the dimensional design parameters held constant. From the
results, a designer can then pick an optimum solution to the requirements,
which can include workspace and footprint ratio. This process can be repeated
for any combination of parameters of SP design.
6
x 10
2
Workspace Vol.(mm )
3
1.5
0.5
0 60
100 90 80 40
70 60 50
Rb Ra
(a) (α, β) is fixed to (30°,30°)
44
5
x 10
11
10.5
Workspace Vol.(mm3)
10
9.5
8.5
40
30 40
30
20
20
10
β
10
α
(b) (Ra,Rb) is fixed as (30mm,75mm)
Figure 4.11 Dimensional parameter synthesis of Stewart Platform
4.3.4 Joints Location
The six universal joints which connect the Stewart platform’s legs to
the base are located at point bi, where i=1, 2,… 6. These points bi are located
along the perimeter of a circle and the centre of this circle coincides with the
origin of the base frame {F} at the centroid of the base platform. Similarly,
the six ball-socket joints which connect the Stewart platform’s legs to the
moving platform are located at point ai, where i=1, 2,… 6. The points bi are
45
located along the perimeter of a circle which centre coincides with the origin
of the moving platform frame {P} at the centroid of the moving platform.
This arrangement is shown in Figure 4.12 for the tool-SP. The circle radii rF
and rP for the base platform and the moving platform are different for the tool-
SP and the table-SP according to their dimensions. Furthermore, the joints are
positioned based on angles θai and θbi with reference from the positive X axis
of the respective frame. Thus, the coordinates of the joint positions in the
respective frames are determined as listed in Table 4.2.
{F} θai XF
rF {F}
ai
{P} XP
rP θbi
{P}
bi
Figure 4.12 Joints location of the tool-SP
Table 4.2 Stewart platform joint locations (in mm)
46
Tool Stewart platform Table Stewart platform
{F} {P} {F} {P}
i ai bi ai bi
152.5 cos (180° ) 70 cos (150° ) 250 cos ( 45° ) 125 cos (15° )
152.5 sin (180° ) 70 sin (150° ) 250 sin ( 45° ) 125 sin (15° )
1 0 0 0 0
152.5 cos ( 210° ) 70 cos ( 240° ) 250 cos ( 315° ) 125 cos ( 345° )
152.5 sin ( 210° ) 70 sin ( 240° ) 250 sin ( 315° ) 125 sin ( 345° )
2 0 0 0 0
152.5 cos ( 300° ) 70 cos ( 270° ) 250 cos ( 285° ) 125 cos ( 255° )
152.5 sin ( 300° ) 70 sin ( 270° ) 250 sin ( 285° ) 125 sin ( 255° )
3 0 0 0 0
152.5 cos ( 330° ) 70 cos ( 0° ) 250 cos (195° ) 125 cos ( 225° )
152.5 sin ( 330° ) 70 sin ( 0° ) 250 sin (195° ) 125 sin ( 225° )
4 0 0 0 0
152.5 cos ( 60° ) 70 cos ( 30° ) 250 cos (165° ) 125 cos (135° )
152.5 sin ( 60° ) 70 sin ( 30° ) 250 sin (165° ) 125 sin (135° )
5 0 0 0 0
152.5 cos ( 90° ) 70 cos (120° ) 250 cos ( 75° ) 125 cos (105° )
152.5 sin ( 90° ) 70 sin (120° ) 250 sin ( 75° ) 125 sin (105° )
6 0 0 0 0
47
4.3.5 Frame Design
Before determining the dimensions of the frame structure, the
kinematic model of the cooperative manipulators is built using MatLab.
Based on mathematical analysis, the workspace can be determined (Figure
4.13a) which can be used to determine the size of the stationary frame. Based
on the volume of the working space, the dimension of the frame structure is
1300 x 840 x 840mm (Height x width x length) (Figure 4.13a).
After determining the dimensions, the frame structure model was
designed (Figure 4.13b). The main frame structure consists of aluminium bars
of cross-section 40 x 40 mm and aluminium bars of cross-section 80 x 80 mm,
which were used at the four corners to provide better rigidity and minimize
vibration during the operating period.. In addition, four pieces of aluminium
plates were assembled together to form a square base structure, surrounding
the base plate of the table-SP.
For the top frame structure, all the aluminium bars (40 x 40 mm) were
welded together and they can move up and down and can be tightened to the
aluminium bars using angle brackets with M8 bolts and nuts. For the top
aluminium cover plate, four M8 holes were drilled and tapped to hold the base
plate of the tool-SP. Another twelve holes were also drilled to hold the cover
plate onto the top of the frame structure.
4.4 Single Stewart Platform Configuration
In this configuration, the tool-SP is used alone. The work-piece is
clamped onto an external fixed plate (referred to as the machining table),
48
below the tool-SP, which can be installed onto the frame as shown in Figure
4.15. Therefore, the table-SP is not used and it can be assumed that the table-
SP is stationary holding the work-piece at one location. The home position of
the tool-SP is not the zero position in the NC code programming. In the NC
code programming, the coordinate system is selected with respect to the
work-piece. Thus, a coordinate transformation is required. In Figure 4.14, a
coordinate frame {W} is attached to the work-piece. The work-piece is
clamped onto the machining table relative to the coordinate frame {W}. By
knowing the transformation between coordinate frames {W} and {F}, the
tool-SP can be controlled to perform the machining task according to the NC
program.
Top
aluminum
cover plate
(a) (b)
Figure 4.13 Frame Design for the cooperative manipulators
49
Stewart Platform Base {F}
{P}
Moving
Platform
Tool
Work
piece {W}
External Plate (Machining table)
Figure 4.14 Single Tool Stewart platform Configuration
Figure 4.15 The machining table installed in the frame
50
4.5 Extended Configuration
In the extended configuration, the work-piece is clamped onto the
moving platform of the table-SP, as shown in Figure 4.16. This table-SP acts
as a movable table. The idea is to enable the work-piece to move beyond the
workspace of the tool-SP so that the tool can access points that are not
reachable in the single configuration. In this configuration, the bases of both
the tool-SP and table-SP are attached with the coordinate frame {F}. The
difference in this configuration is that the transformation of the work-piece
coordinate frame relative to frame {F} is no longer constant. The work-piece
coordinate frame is attached to the table-SP moving platform. However,
because both SPs base are installed to the frame, the relation between
coordinate frames of the tool-SP and the table-SP can be determined.
Therefore, the machining task can be performed using both SPs in the
extended configuration, which provides a bigger overall workspace than the
single configuration.
51
Figure 4.16 Machining a clamped work-piece on the table-SP with the tool-
SP
52
CHAPTER 5
SIMULATION AND CONTROL OF STEWART
PLATFORM
5.1 Workspace Analysis and Kinematic Constraints
It is important to develop algorithms by which the workspace of the
Stewart platform can be determined and the effect of different designs on the
workspace can be evaluated. The constraints that limit the workspace of a SP
are modelled, namely, the ranges of the leg actuator, collisions between the
legs, and the physical limitations of the passive joints (Harib and Srinivasan
2003). Additionally, singular configurations are to be avoided. In the
workspace analysis, the following constraints are implemented.
1. Constraint on the leg actuator’s length
Each leg of the Stewart platform has a minimum and maximum length
due to its mechanical limit. Therefore, each leg length must satisfy
Equation 5.1, where Li,min and Li,max are the minimum and maximum
lengths of the ith leg, respectively.
Li,min < Li < Li,max, i = 1,2…6 (5.1)
2. Constraint on the range of the passive joints angle
The ball-socket joints and the universal joints have their respective
operating angles. Let θi and φi be the current angles of the ball-socket
and universal joints, and they must satisfy Equation 5.2, where θi,min
53
and φi,max are the maximum operating angles of the ith ball-socket
joints and the ith universal joints, respectively. The value of the current
angle of the passive joints can be determined by the cosine rule to
obtain the angle between two vectors.
θi < θi,max ; φi < φi,max , i = 1,2…6 (5.2)
Let jai and jbi be the unit vector along the axis of symmetry of the ball-
socket joints and the universal joints respectively. Furthermore, let ni
be the unit vector along the leg axes, which can be computed by
dividing the leg vector by its magnitude, ni = li / Li. The current angle
of the ball-socket joints and the universal joints are calculated using
Equation 5.3.
cos-1 ( jaiT . ni ) , i = 1,2…6;
cos-1 ( jbiT . ni ) , i = 1,2…6 (5.3)
3. Constraint on the collisions between the legs
The legs can be modelled as a cylinder with a certain height and
diameter, DL. The centre of the cylinder is a line segment and
coincides with the centre of the leg axis. Hence, in order to avoid any
collision, the minimum distance between any two line segments
corresponding to the legs of the Stewart Platform should be greater
than or equal to DL.
Since a 6-DOF workspace plot is not possible, only a 3D plot is used
to represent the 3-DOF workspace graphically. The 3-DOF workspace can
include any of three generalized coordinate variables of the moving platform
54
location. However, the most commonly used 3-DOF workspace is the 3-DOF
positional workspace, which contains all the points reachable in a 3D
Cartesian space with a fixed platform orientation.
The 3-DOF positional workspace of a Stewart platform is dependent
on the orientation of its moving platform. The workspace can be defined as
the 3D Cartesian space which is reachable by the centre of the moving
platform where the origin {P} is located. The algorithm to find the workspace
is performed using a MatLab program, which algorithm can be summarized
in Figure 5.1. This algorithm is known as the discretization method for the
workspace computation of the Stewart platform. The 3D space is divided into
finite number of discrete points representing a small region (a cell) of
workspace with a certain size (a box with a length, a width, and a height). The
box size can be approximated by choosing a size that contains the workspace.
The algorithm is to check all the points whether they are reachable by the
moving platform and store the result. After all the points have been tested, the
results can be plotted.
The algorithm only checks the point of a corresponding cell. There is
always a possibility that a portion of a cell is not reachable by the moving
platform. Thus, the result of the discretization method is an approximation
and it is more accurate than the actual volume with a smaller cell size. Let
Nreach be the number of reachable points, the calculation of the volume of the
workspace can be implemented by adding the volume of all the cells that are
reachable using Equation (5.4), where ∆X x ∆Y x ∆Z is the cell size in the
Cartesian coordinates.
55
V = N reach ∗ ∆X .∆Y .∆Z , (5.4)
Figure 5.2 presents the plot of 3D positional workspace of the tool-SP
with the orientation (0°, 0°, 0°). All the points inside the envelope in the figure
are reachable. For this particular analysis, the cell size is ∆X x ∆Y x ∆Z =
1mm x 1mm x 1mm, and the total volume is 386,680 mm3. After obtaining
the result of the workspace, the limitation of the position and orientation of
the moving platform can be verified. Hence, the motion of the platform can
be operated safely within the allowance of the workspace.
56
Get kinematic parameters of
the Stewart platform
description
Collection of
points (position
coordinates) to be
tested from the test
space
Process the next point in the
collection
END
No point left in the
collection? YES
NO
Inverse kinematics, determine
leg lengths and the passive
joints angle
YES
Any constraint
violated?
Store the point as reachable
point and plot the result on the
screen
Figure 5.1 The flowchart to plot the workspace of the Stewart platform
57
-290
-300
-310
Z
-320
-330
-340
50
50
0
0
X -50 Y
-50
Figure 5.2 The tool Stewart platform workspace
5.2 Stewart Platform User Interface
The development of the Stewart platforms requires testing of the
hardware and software. A user interface has been developed for controlling
the Stewart platforms. The interface allows the user to define platform motion
relative to the various systems: base coordinate frame, moving platform
coordinate frame, and joint coordinates. In addition to these references, a
joystick control has been added as a means to perform rapid movements of
the Stewart platforms. The constraints of the passive joints that limit the
movement of the moving platform are implemented after inverse kinematics
calculation is performed. Therefore, the software can simulate the motion
prior to its execution on a Stewart platform in real time.
The velocity and position input to the Stewart platform’s controller
can be provided in the following ways: manual input to the user interface,
58
pre-programmed data file, and 6-DOF joystick or a combination of joystick
and keyboard. Figure 5.3 shows a user interface. In this interface, there are a
few selections to be set in order to control the Stewart platforms, and they are
listed as follows.
• Stewart platform to be controlled.
• Type of control: point-to-point motion, jog motion or spline motion,
manual using joystick or keyboard, or read trajectory from a file.
• Sampling time and scale factor for the joystick control.
• The target location to move the moving platform for point-to-point
control.
• Input file for the trajectory of the motion.
• Option to execute the prescribed motion in simulation or execution on
the Stewart platform.
• Constraints that have to be verified to prevent damage.
59
Figure 5.3 User Interface for Stewart platform control
The user interface is capable of guiding the user through various
options of the interface in order to avoid any damage that may be done to the
Stewart platforms. Most common errors can be avoided by checking the
system status whenever possible. The interface displays various options from
which the user can select using the PC pointing device (mouse). The user can
opt to run the motion in graphical model shown in the interface without
moving the actual Stewart platforms to check whether the execution of the
motion is according to the desired trajectory.
When the program starts by establishing communication with the
corresponding PC controller cards and homes all actuators to their zero
reference positions. If the platform is initially high above the home position,
performing homing directly is not advisable. This is because the homing
function homes one axis at a time and hence if the platform is too far away
60
from the home position, the process can damage the joints. Therefore, there
is an option to select the axis to perform the home function. The user is
expected to know the state of the Stewart platform to select the axis to move.
Two additional options are implemented for joystick control, namely,
scale factor and sampling time. The scale factor is the value of how much the
joystick movement range is multiplied to determine the actual Stewart
platform motion. The scale factor determines the range of the movement of
the actual Stewart platform corresponding to each axis. This option enables
the user to perform a rapid motion on the Stewart platform. Secondly, the
sampling time is the waiting time between two subsequent data reads from
the joystick. This is related to the response time of the Stewart platform to the
joystick movement. The smaller the value, the more responsive is the Stewart
platform and vice versa. There is a minimum sample time to prevent over
speeding the moving platform. The joystick input can be treated as a distance
relative to the current location of the moving platform or as a velocity. Both
options can be selected from the user interface.
The graphical model of the Stewart platform moves when a move
command is being executed. If the real motion option is selected, the Stewart
platform also moves accordingly. The forward kinematics is implemented in
the background such that for each movement, the values of the joint
coordinates and the location of the moving platform are updated. Based on
these values, the program is set to stop immediately and will halt all the
Stewart platform motion as soon as one of the constraints is violated. Thus,
61
the state of the Stewart platform is always known to prevent violation of the
constraints and joints over the limits during motion.
There is an option to specify the velocity of the motor of the Stewart
platform legs when manual input using keyboard or automated motion is
being selected. The input file should be specified if an automated motion is
requested. This automated motion is especially useful for executing a pre-
programmed trajectory, such as in NC machining. In summary, the user
interface provides a basic and simple way to control the Stewart platforms.
5.3 Programming
The control of the user interface is implemented in MatLab which has
built-in mathematical functions and matrix operations that make it ideal for
inverse and forward kinematics calculation. In addition, it has a graphical user
interface design environment (GUIDE) feature, which is useful for creating a
user interface. In terms of control, MatLab can communicate with the
controllers of the tool-SP and the table-SP simultaneously. This is supported
by the MatLab capability to call C/C++ library, which is provided by the
vendors.
As has been stated above, there are two input methods to the interface
for controlling the Stewart platforms, namely, manual with keyboard or
joystick, and automated by a pre-programmed data file. In addition, the
controller can generate different types of motion profiles, namely, point-to-
point motion, jog motion and spline motion. They are explained as follows,
62
1. Point-to-point motion: The response of the Stewart Platform is
proportional to the joystick movement or the time period of a key
pressed on the keyboard. Another option is to input the desired
destination in the user interface or in the pre-programmed data file. This
option uses a constant velocity for each leg. For machining, this motion
scheme is used.
2. Spline motion: The joystick movement and keyboard input are
interpreted as velocity in the corresponding direction of the moving
platform. However, the values sent to the controller are the leg length
values. Therefore, the program calculates the distance travelled per each
sampling time period. The spline function is not suitable for manual
input using keyboard or joystick because it needs to predict the next
movement to determine the current velocity. It uses bi-cubic
interpolation to calculate the required acceleration and velocity along
the specified motion profile. Hence, this motion scheme results in lower
position error.
3. Jog motion: The joystick movement and keyboard input are interpreted
as velocity in the corresponding direction of the moving platform. The
program sends the computed velocity values of each leg to the controller.
The motion is smoother compared to the spline motion because there is
no velocity interpolation. The side effect is that the resulting motion has
lower accuracy. This motion is useful for rapid homing or retracting.
4. Interpolated motion: The previous motion types do not consider the
kinematics of the Stewart platform. Since the motion control is done in
63
the joint coordinates, the resulting moving platform motion profile may
not be as desirable in the specified trajectory because of the non-linear
relationship between the location of the moving platform and the leg
lengths. In order to control the moving platform trajectory correctly, the
interpolated motion is developed. The interpolated motion uses many
via-points between the starting and final position of the moving
platform. Inverse kinematics calculation is performed in each via-point
to update the required leg lengths. This process ensures that the moving
platform follows the desired trajectory between the starting and final
position.
The actuator leg lengths obtained after performing inverse kinematics
are sent to the controllers after ensuring that the constraints are satisfied.
Figure 5.4 summarizes the algorithm for the point-to-point motion in a
flowchart. In each sampling, the variable in the selected coordinate system is
decreased or increased according to the state of the input and the magnitude
is set by the scale factor in the user interface.
64
Figure 5.4 Point-to-point motion control flowchart
For the spline motion, the program has to calculate the leg lengths in
each sampling period. The flowchart for the spline motion program is shown
in Figure 5.5. The input is interpreted as velocity reference for the moving
platform. In each sampling period, the new position is calculated. This is
followed by inverse kinematics to determine the leg lengths. After a sequence
of leg lengths has been acquired, it is sent to the controller.
65
Figure 5.5 Spline motion control flowchart
The jog motion program interprets the input from the joystick or
keyboard as a velocity reference of the moving platform. In each sampling
period, the new leg lengths are calculated using inverse kinematics. The
velocity of each leg is determined by dividing the difference between the new
and the current lengths by the sampling time. The velocity is sent to the
controller as soon as the calculation is completed. Figure 5.6 summarizes the
algorithm flowchart for the jog motion program.
66
Figure 5.6 Jog motion control flowchart
Figure 5.7 shows the flowchart of the interpolated motion program.
Interpolated motion is developed for executing the trajectory stored in a data
file. The trajectory for machining can be obtained from the NC code for a
certain job. In this program, more data points (via-points) between subsequent
coordinates in the trajectory are added. By using many via-points, the
discrepancies between the desired and the actual motion trajectory can be
reduced significantly. As can be seen in Figure 5.6, the algorithm is a
repeating point-to-point motion between the via-points. However, there is
limitation on the minimum sampling period to allow sufficient time for the
inverse kinematics calculation and communication with the controller while
performing the motion with desired velocity.
67
Figure 5.7 Interpolated motion control flowchart
5.4 Numerical Control Post-Processor for Stewart Platform
In this research, a commercial CAD/CAM software is used to design
parts and generate the tool paths for these parts. The CAM output is the NC
code to be sent to a standard CNC machine tool. The Stewart platform
controller cannot interpret this NC code format. Hence, additional processing
is required to convert the NC code to a data format which can be executed in
the Stewart platform controller. This conversion is developed using MatLab.
The NC codes used for controlling a machine tool are based on certain
conventions similar to computer programming languages. The NC codes form
a sequence of codes called an NC program. An example of an NC program is
shown in Figure 5.8. Table 5.1 summarizes a variety of NC codes and their
functions. Each code in an NC program corresponds to a command to be
executed by the CNC controller. The syntax of an NC code is a single
character followed by a number. For example, to set a standard CNC machine
68
to accept input values of coordinates of the tool in millimetres, a specific G-
code of ‘G21’ must be sent to initialize the machine.
Figure 5.8 Common NC program format
Table 5.1 NC codes and their functions
Character Function
A Additional rotary axis parallel and around the X axis
B Additional rotary axis parallel and around the Y axis
C Additional rotary axis parallel and around the Z axis
D Tool radius offset number, Depth of cut for multiple
repetitive cycles
E User macro character, Precise designation of thread lead
F Feed rate, Precise designation of thread lead
G Preparatory function
H Tool Length offset number
I Incremental X coordinate of circle centre or parameter of
fixed cycle
J Incremental Y coordinate of circle centre
K Incremental Z coordinate of circle centre or parameter of
fixed cycle
L Number of repetition
M Miscellaneous function
N Sequence or block number
O Program number
P Dwell Time, program number, and sequence number
designation in subprogram
Sequence number for multiple repetitive cycles
Q Depth of cut, shift of canned cycles
Sequence number for multiple repetitive cycles
69
R Point R for canned cycles, as a reference return value
Radius designation of a cycle arc
S Spindle-speed function
T Tool-function
U Additional linear axis parallel to X axis
V Additional linear axis parallel to Y axis
W Additional linear axis parallel to Z axis
X X coordinate
Y Y coordinate
Z Z coordinate
Since the codes beginning with ‘G’ and ‘M’ are most commonly used,
these codes are usually referred to as ‘G-code’ and ‘M-code’. In the current
work, the G-code and coordinate codes, such as ‘A’,’B’,’C’,’X’,’Y’, and ’Z’
are sufficient to control the Stewart platform as it does not have many
preparatory or miscellaneous functions. A sequence of NC codes can be
expressed as shown in the example shown next.
Word Word Word Word Word
N01 G00 X24 Y45 F40
Each word contains a character, followed by specific data (number)
which can further be separated as shown below.
Char Data Char Data Char Data Char Data Char Data
N 01 G 00 X 24 Y 45 F 40
By parsing and analysing these characters and data, the corresponding
machine operation can be decoded and a tool path can be generated. A
program block can consist of several G-codes. The list of G-codes that are
used in the post-processor is shown in Table 5.2.
70
Table 5.2 G-codes used for Stewart platform and their meaning
Code Group Function
G00 01 Rapid traverse positioning
G01 01 Linear interpolation
G02 01 Circular and helical interpolation CW (clockwise)
G03 01 Circular and helical interpolation CCW (counter-
clockwise)
G17 02 XY plane selection
G18 02 ZX plane selection
G19 02 YZ plane selection
G21 06 Input in millimetres
G90 03 Absolute programming command
G91 03 Incremental programming command
Based on the aforementioned explanation, a MatLab program is
created to fetch a NC code file to parse each block and to translate it to
coordinates in the Stewart platform coordinate system. These coordinates are
stored in the data file and are used to control the Stewart platform according
to the machining trajectory.
5.5 Stewart Platform Motion Emulation and Dynamics
The MatLab Virtual Reality Toolbox is used to model the complex
constraints in the motion of a Stewart platform so that the entire geometric
model (i.e., ball-socket joints, universal joints, platforms, compartments, and
work-pieces) is considered. As a result, the process can be analysed in real-
time. The collision avoidance algorithm uses the 3D model developed to
compute the minimum distances between objects. The MatLab
SimMechanics toolbox is used to model the dynamics of the Stewart platform.
Figure 5.9 and Figure 5.10 show the representation of the 3D model and
71
dynamic control of the Stewart platform in the SimMechanics toolbox and
Simulink, respectively. In this work, the integration of the motion animation
with the Stewart platform control is implemented. The dynamics simulation
is based on the data from the components vendor.
Figure 5.9 A 3D Stewart platform model
Figure 5.10 Stewart platform model in MatLab Simulink
72
CHAPTER 6
IMPLEMENTATION OF THE COOPERATIVE
MANIPULATORS AS MACHINE TOOL
6.1 Coordinate Mapping of the Cooperative Manipulators
The task description of a spatial PKM generally consists of three
linear coordinates and three angular coordinates describing the translation and
rotation displacements of the end-effector relative to the object. The
cooperative manipulators has a total of 12-DOF, consisting 6-DOF from the
tool-SP and 6-DOF from the table-SP. For machining applications, the task
description is defined in between three to five coordinates depending on the
type of the machining job. For the 2.5- and 3-axis machining, there are three
Cartesian coordinates. In complicated multi-axis machining, up to 2-DOF
angular motion can be defined. These 1- or 2-DOF angular motions add two
extra coordinates.
The developed post-processor is able to convert 5-axes machining to
five corresponding coordinates in the Stewart platform coordinate frame. The
machining can be performed using the tool-SP alone in the single
configuration or with both the tool-SP and the table-SP in the extended
configuration.
73
6.1.1 Single Stewart Platform Configuration
In the single configuration, the end-effector is the tool attached to the
moving platform of the tool-SP and the object is the work-piece fixed onto
the machining table. The detailed description of the single configuration with
attached coordinate systems is presented in Figure 6.1. It is assumed that the
base platform of the tool-SP and the work-piece are positioned in such a way
that it is parallel to the machining table and the origin of frame {F} and frame
{W} are aligned. The object or the work-piece is fixed onto the machining
table below the tool-SP with a distance of h. There is a distance of dt from the
moving platform to the tool tip. The tool is attached to the tool-SP in such a
way that the tool axis is coincident with the Z-axis of {P}.
The generalized coordinates used in the tool-SP are defined as the
position and the orientation of the moving platform frame {P} with respect to
the base frame {F}.
F
XP = [qx , qy , qz , θ1 , θ2 , θ3 ]T (6.1)
74
ZF
{F} YF Tool-SP
XF ZP
Height between
the tool-SP base {P} YP
and the Tool height dt
machining table XP
h
ZW
{W}
YW Work-piece
coordinate-frame
XW
Object or
work-piece
Figure 6.1 Coordinate system in the single configuration
The orientation coordinates must follow a certain convention for
describing the rotation of the tool. In this research, the orientation of frame
{P} with respect to frame {F} is specified by the following sequence of
rotations. First, by rotating frame {P} about axis ZF with an angle θ1, and
rotating about the new axis XP with an angle θ2 , and then rotating about the
new axis ZP with an angle θ3. Thus, the rotation matrix can be derived as
Equation 6.2.
F
RP = Rθ1 Rθ2 Rθ3
Cθ 1Cθ 2Cθ 3 − Sθ 1Sθ 3 −Cθ 1Cθ 2 Sθ 3 − Sθ 1Cθ 3 Cθ 1Sθ 2
= Sθ 1Cθ 2Cθ 3 + Cθ 1Sθ 3 − Sθ 1Cθ 2 Sθ 3 + Cθ 1Cθ 3 Sθ 1Sθ 2 (6.2)
− Sθ 2Cθ 3 Sθ 2 Sθ 3 Cθ 2
where Cθ = cos(θ) and Sθ = sin(θ), and
75
Cθ 1 − Sθ 1 0
Rθ1 = Sθ 1 Cθ 1 0
0 0 1
Cθ 2 0 Sθ 2
Rθ2 = 0 1 0
− Sθ 2 0 Cθ 2
Cθ 3 − Sθ 3 0
Rθ3 = Sθ 3 Cθ 3 0 (6.3)
0 0 1
In general, the task description of a manipulator is input coordinates
consisting of three linear coordinates and three angular coordinates describing
the translation and rotation displacements of the end-effector relative to the
object. For the PKM based machine tool, the input coordinates can be defined
as the tool contact point with the work-piece. The vector representing the
input coordinate in the coordinate frame {W} is defined as Equation 6.4.
{W}
Xi = [x , y , z , Φ, Γ , Ω]T (6.4)
The coordinates x, y, z are the translation and the coordinates Φ, Γ , Ω
are the rotation. The coordinate description in Equation 6.1 can be mapped to
the 6-DOF coordinate system for the tool-SP. However, according the
complexity of the machining task, fewer than 6-DOF may be required. For 3-
axis machining, only the translation coordinates are provided by the
CAD/CAM software as the tool orientation is not needed. For 4- or 5-axis
machining, one or two orientation coordinate(s) is used.
76
For machining, the coordinates Φ, Γ are used for defining the 4th and
the 5th rotation axes of the tool. The rotation coordinates Φ, Γ are defined as
follows. First, by rotating about the new axis XW with an angle Φ, and then
rotating about the new axis ZW(1) with an angle Γ. The last orientation
coordinate Ω corresponds to the rotation of the tool along the tool axis which
is redundant with the spindle rotation (cutter rotation for removing material).
Hence, this last orientation is not needed for machining and it represents an
extra DOF that could be used to optimize a certain index. This extra DOF
comes from the fact that the machining task only requires maximum five-axis
motion and the tool Stewart platform has six DOF.
The rotation angle coordinates can be mapped according to the
description above. The first two input angle coordinates are mapped to the
first two rotation coordinated of the tool SP, θ1 = Φ, θ2 = Γ . The free DOF
which is the coordinate Ω is mapped to the third rotation coordinates of the
tool-SP, θ3 = Ω. It can be set to a constant zero, Ω = 0 for a trivial choice. The
rotation matrix becomes Equation 6.5.
CΦ CΓ − SΦ CΦ S Γ
F
RP = SΦ CΓ CΦ SΦ SΓ (6.5)
− SΓ 0 CΓ
Alternatively, it can be set such that Ω = – Φ while maintaining the
optimum pose of the moving platform (Bonev et al. 2002). The rotation
matrix for this choice is given as Equation 6.6.
77
C 2 Φ CΓ + S 2 Φ SΦ CΦ ( CΓ − 1) CΦ SΓ
SΦ CΦ ( CΓ − 1) S Φ CΓ + C Φ S Φ SΓ
F 2 2
RP =
− S Γ CΦ − SΓ S Φ CΓ
(6.6)
The translation coordinates [x,y,z] describe the position of the tool
contact point with respect to frame {W}. These coordinates can be mapped to
the position of the tool contact point FTc (x’,y’,z’) by transforming the input
coordinates to the tool-SP base frame {F}. The transformation of frame {W}
with respect to frame {F} is given as Equation 6.7 based on the assumption
stated above.
1 0 0 0
0 1 0 0
F
TW = (6.7)
0 0 1 −h
0 0 0 1
Thus, the x and y coordinates can be directly equated with the
coordinates of tool-SP moving platform, x’ and y’. The z coordinate is
translated to z’ with offsetting the values according the location of the origin
of frame {W} with respect to frame {F}, which is the height h.
x’ = x ; y’ = y ; z’ = z - h (6.8)
Furthermore, since there is a distance between the tool contact point
and the centroid of the moving platform, the position of the moving platform
must be adjusted to achieve the correct location of the tool. This issue can be
78
described in Figure 6.2. The vector dt . vt is parallel to the tool axis as shown
in Figure 6.2. Therefore, the moving platform (the origin of frame {P})
position with respect to frame {F} in order to achieve the tool contact point
at FTc can be calculated as Equation 6.9. where vt can be obtained as the unit
vector of the Z axis of the moving platform with an opposite direction. The
coordinate mapping can be summarized in Figure 6.3.
F
OP = FTc - dt . vt (6.9)
ZF
Tool-SP
{F}
OF YF
XF
ZP
{P}
F
XP OP (qx , qy , qz)
d t . vt YP
F
Tc = (x’ , y’ , z’) Tool
Figure 6.2 The moving platform position to reach the tool contact point
79
{W} F
Xi XP
x’ = x
x , y, z y’ = y
z‘=z-h
= - d t . vt
Φ, Γ , Ω
θ1 = Φ
θ2 = Γ
θ3 = Ω = 0 ;
or θ3 = Ω = - Φ
Figure 6.3 Coordinate mapping in single configuration
6.1.2 Extended Configuration
In the extended configuration, both the tool-SP and the table-SP are
utilized. The coordinate system is presented in Figure 6.4. The table-SP is
placed below the tool-SP and installed in the frame. The height difference
between the Stewart platform bases is h2. The base frames {F} and {F2 } are
made parallel to each other and their base frame origins are aligned. Each SP
has six DOF, hence the cooperative manipulators in this extended
configuration has a total of 12 DOF. The generalized coordinates used in the
table-SP are defined as the position and the orientation of the moving platform
frame {P2} with respect to the base frame {F2}.
F2
XP2 = [qx2 , qy2 , qz2 , θ12 θ22 , θ32 ]T (6.10)
80
The transformation matrix of frame {F2} with respect to frame {F} is
given as Equation 6.11.
1 0 0 0
0 1 0 0
F
TF2 = (6.11)
0 0 1 −h 2
0 0 0 1
The coordinate mapping is formulated to position and orientate both
the tool and the work-piece in order for the tool to reach the desired tool
contact point on the work-piece by using both the tool-SP and the table-SP.
This means in order to fulfil the coordinate map, the tool-SP must move its
moving platform in such a way that the tool tip can reach the desired tool
contact point. This tool contact point is determined from the input coordinate
{W}
Xi = [x , y , z , Φ, Γ , Ω]T .
The work-piece in the extended configuration is fixed onto the moving
platform of the table-SP. It is assumed that the work-piece is setup such that
the description of the input coordinates in frame {W} is the same as the input
coordinates in frame {P2}. Thus, one can also write the input coordinate as
Equation 6.12.
{P2}
Xi = [x , y , z , Φ, Γ , Ω]T (6.12)
In addition, the tool-SP carries the tool so that the tool tip can reach
the coordinate described by FTc , which is the input coordinate relative to
frame {F} (the base frame of the tool-SP). Figure 6.5 explains the
configuration of the attached coordinate systems. The tool contact point on
81
the work-piece and the tool contact point that must be reached by the tool tip
are the same point such that, Xi = Tc must be fulfilled for the condition of
coordination of both SPs.
Thus, the translation coordinates can be mapped as follows. The tool-
SP moving platform translation follows the same derivation as in the single
configuration (Equation 6.9). However, the tool contact point FTc (x’,y’,z’)
{P2}
must be calculated from the input coordinates Xi . In order to determine
F
Tc (x’,y’,z’), the translation input coordinates x,y,z can be transformed with
respect to frame {F}. According to Figure 6.5, the tool contact point with
respect the table-SP base frame {F2} is the addition of two position vectors
in Equation 6.13, where F2RP2 is the rotation matrix of frame {P2} relative to
frame {F2}.
qx 2 qx 2 x
{F2}
Xi = q y 2 + F2RP2 {P2}Xi = q +
y2
F2
RP 2 y (6.13)
qz 2 qz 2 z
This rotation matrix can be derived in the following manner. Since
the relationship of the base frame {F} and {F2} is known in Equation 6.11,
x ' 0 qx 2 x 0
F
Tc ≡ y ' = Xi + 0 =
{F2} q + F 2 R y + 0
y2 P2 (6.14)
z ' −h2 qz 2 z −h2
The table-SP moving platform has three extra DOF. As can be inferred
from Equation 6.14, although the input coordinates are defined relative to
82
frame {P2}, the translation of the moving platform is not restricted as long as
the tool-SP moves the tool such that its tip touches the tool contact point FTc.
ZF
{F} YF
Tool-SP
XF ZP
{P} YP
Tool height dt
Height between
XP
the tool-SP base
and the table-SP
base ZP2
h2
YP2
Object or XP2 Table-SP
work-piece {P2}
ZF2
{F2} YF2
XF2
Figure 6.4 Coordinate system in extended configuration
83
ZF
{F}
OF
YF
XF
ZP
{P}
F
XP OP (qx , qy , qz)
d t . vt YP
F
Tc = (x’ , y’ , z’) Tool
{P2}
Xi = (x , y , z)
ZP2 Work-piece
The intended location of
the tool contact point on
the work-piece from {P2} Table-SP
input coordinates YP2
F2
XP2 OP2 (qx2 , qy2 , qz2)
ZF2
{F2}
YF2
OF2
XF2
Figure 6.5 Coordination of the tool-SP and the table-SP to reach the input
coordinate (the tool contact point on the work-piece)
Rotation coordinates mapping is more complicated if the sequence of
the rotation is arbitrary. However, the tool-SP has two rotation DOFs that are
used for machining and the third DOF is not used because it rotates about the
tool axis. This is also true for the table-SP. To simplify rotation coordinates
mapping, the rotation axes of the table-SP are made the same with the
84
rotation-axes of the tool-SP, so that the angles can be added arithmetically.
The first two rotation coordinates of the tool-SP are used for machining, as
the last one is a free DOF that can be set arbitrarily. It can also be set for the
table-SP that the first two rotations are the DOFs that are used in machining.
The two rotation axes (Φ, Γ ) are mapped to the four rotation DOFs of both
tool-SP and table-SP ( θ1,θ2,θ12,θ22 ), one in each couple is being redundant
to one another (i.e. θ1 is redundant to θ12 and θ2 is redundant to θ22).
Figure 6.6 describes the rotation axes used in the tool-SP in the single
configuration. The table-SP Euler angles must have the same convention as
the Euler angles of the tool-SP so that the angle can be combined. The table-
SP first rotation is along the ZP2 axis, which is parallel to the ZP axis of the
tool-SP. So, the first input rotation coordinate can be mapped using Equation
6.15..
Φ = θ1 – θ12 (6.15)
The minus sign in Equation 6.15 comes from the fact that the table-
SP locates the work-piece relative to the tool, so it becomes the opposite
direction. Moreover, the rotation matrix for the first rotation on the table-SP
can be calculated using Equation 6.16.
Cθ 1 − Sθ 1 0
Rθ12 = Sθ 1 Cθ 1 0 (6.16)
0 0 1
85
The second rotation axis of the table-SP is defined as parallel to the
second rotation axis of the tool-SP, so that the angle can also be summed
using Equation 6.17.
Γ = θ2 – θ22 (6.17)
The second rotation axis is along the new axis XP after the first rotation
on the tool-SP. The rotation sequence on the table-SP can be summarized in
Figure 6.7. The rotation matrix for the second rotation on the table-SP can be
constructed by using axis-angle representation of the rotation about the vector
XP’. The vector XP’ is the first column of matrix Rθ12 .
S 2θ 1 (1 − Cθ 22 ) + Cθ 22 − Sθ 1Cθ 1 (1 − Cθ 22 ) Cθ 1Sθ 22
Rθ22 = − Sθ 1Cθ 1 (1 − Cθ 22 ) C 2θ 1 (1 − Cθ 22 ) + Cθ 22 Sθ 1Sθ 22 (6.18)
−Cθ 1Sθ 22 − Sθ 1Sθ 22 Cθ 22
Although it is possible to define the third rotation, the use of the DOF
may not be necessary for the current machining application. Therefore, the
third rotation is fixed at 0, or it can be defined that the rotation matrix of the
third rotation on the table-SP is an identity matrix.
Rθ32 = I (6.19)
The rotation matrix of frame {P2} relative to frame {F2} is the
multiplication of the above rotation matrices according to the rotation
sequence.
F2
RP2 = Rθ12 Rθ22 Rθ32 (6.20)
86
ZP’’ ZP
YP’’
YP’
θ2
θ2
θ1 YP
θ1
XP
XP’
Figure 6.6 Rotation sequence in the tool-SP
ZP’’ ZP2
YP’’
YP2’
θ22
θ22
θ12 YP2
θ12 θ22
XP2 XP’
(of the tool SP)
XP2’
XP2’’
Figure 6.7 Rotation sequence in the table-SP
Coordinate mapping can be summarized in Figure 6.8.
87
{W}
Xi = {P2}Xi F
XP and F2XP2
x , y, z =
= - d t . vt
Φ = θ1 – θ12
Φ, Γ , Ω Γ = θ2 – θ22
θ3 = Ω = 0 ;
or θ3 = Ω = - Φ
θ32 = 0
Figure 6.8 Coordinate mapping in extended configuration
6.2 Extended Configuration Motion Planning
This section addresses the problem of optimal trajectory planning for
the cooperative manipulators. The objectives are to maximize the stiffness of
the structure and minimize the force requirement of the actuators while
satisfying the constraints of workspace and singularity. A computational
algorithm for searching the optimum location of machining trajectory is
developed. The performance index is based on the condition number of the
Jacobian matrix of the Stewart platforms.
The condition numbers are accepted as a measure of the manipulator
performance as the Jacobian matrix represents the relationships between the
velocities and the force acting on the actuators, and the velocities and the twist
88
of the end-effector (Stoughton and Arai 1993). If the Jacobian matrix
becomes singular (i.e., when the condition number reaches infinity), the
manipulator will gain additional freedom and lose its rigidity and ability to
move and apply forces in arbitrary directions. The condition number
expresses the amplification of the relative error in the joint coordinates to the
error of the moving platform pose. Thus, the condition number has to be as
small as possible. The following are the derivation of the Jacobian matrix of
the Stewart platform and the corresponding condition numbers.
6.2.1 Jacobian Matrix and Condition Number
For a PKM, the Jacobian matrix is defined as
lɺ = J(X) Xɺ (6.21)
where lɺ is the rate of change of the length of the legs and
T
Xɺ = [ qɺ T θɺT ]T = qɺ T ω T is the Cartesian velocity of the end effector.
Using inverse rate kinematics (Harib and Srinivasan 2003), both Jacobian
matrices can be obtained. The location of the ball-socket joints are
differentiated with respect to time, and the projections of these velocity
vectors on the axis of the legs yield the extension rates of legs of the tool-SP
and the table-SP as in the following.
lɺi = F aɺi ⋅ ni = F qɺP ⋅ ni + ω ⋅ ( F RP ⋅ P ai ) × ni , ω = θɺ1,θɺ2 ,θɺ3 ; i = 1,2,..6
T
(6.22)
lɺi 2 = F 2aɺi 2 ⋅ ni 2 = F 2qɺP 2 ⋅ n2i + ω2 ⋅ ( F 2 RP 2 ⋅ P 2ai 2 ) × ni 2 ; ω2 = θɺ12 ,θɺ22 ,θɺ32 ;
T
i = 1,2,..6 (6.23)
89
The Jacobian matrices can be obtained through comparing the
following equations derived from Equation (6.21).
ɺl = J Xɺ = J qɺ P
F
(6.24)
ω
F 2 qɺ P 2
lɺ2 = J 2 Xɺ 2 = J 2 (6.25)
ω2
Thus, the Jacobian matrix for the tool-SP and the table-SP are
nT ( RP ⋅ P a1 × n1 )
F T
1
T
( RP ⋅ a2 × n2 )
F P T
n
J = 2
⋮ ⋮ (6.26)
nT ( RP ⋅ a6 × n6 )
F P T
6
nT ( RP2 ⋅ P a12 × n12 )
F2 T
12
T
( RP2 ⋅ P a22 × n22 )
F2 T
n
J 2 = 22
⋮ ⋮ (6.27)
nT ( RP2 ⋅ a62 × n62 )
F2 P2 T
62
The condition number, κ , of the Jacobian matrix J is defined as
(Golub and Van Loan 1983),
κ = J J− (6.28)
κ 2 ( X 2 ) = J 2 J −2 (6.29)
90
where J denotes the 2-norm of the Jacobian matrix J. Since the Jacobian
matrix is a function of the moving platform pose, the condition number is also
a function of the pose.
The condition number will be inconsistent if it is derived directly from
the non-homogeneous Jacobian matrix J. To avoid this problem, the
rotational elements, i.e., the last three columns of the Jacobian matrices are
divided by the radius of the mobile platform of each manipulator (Ma and
Angeles 1991b) before the computation of the condition number.
In the next section, the condition number is used in the objective
function in the optimization procedure to solve the redundant parameters.
However, as there are two Stewart platforms in the cooperative manipulators,
an objective function that multiplies the condition numbers of both platforms
is defined.
f ( X , X 2 ) = κ ( X ) ⋅ κ2 ( X 2 ) (6.30)
The condition number ranges from 1.0 to infinity, i.e., 1 < κ < ∞ .
Thus, the value of the scalar multi-variable function f has the same range and
it reflects the performance of both Stewart platforms with respect to their
particular pose. Through minimizing this function, the optimal pose
configuration can be obtained for a given input.
For stiffness analysis, the stiffness measure is obtained by computing
the stiffness matrix which is given by Equation 6.31.
= (6.31)
91
where χ = diag [k1,k2,…,k6] and k1,k2,…,k6 are the axial stiffness of the legs
(El-Khasawnath and Ferreira 1999). The stiffness of the SP is taken as the
minimum eigenvalue of matrix ϑ. The optimization will yield trajectory
which direction is much affected by the maximum mean of the stiffness value.
In addition, in the optimization singularity must be avoided. It can be
formulated for each SP in the following. If the actuated leg lengths are
denoted by l and the location of the end-effector is denoted by X, then the
singularity constraint imposed on the SP is
= (6.32)
Conceptually, if the J’s determinant is equal to zero, the
corresponding condition number will reach an infinite value and thus,
infinitesimal motion of the end-effector can exist although all leg lengths are
fixed.
6.2.2 Optimization Procedure
Optimization (minimization) of the condition number is carried out to
determine the redundant parameters so that the resulting pose will give the
best dexterity for a given task. The reachable workspaces of both platforms
need to be validated prior to the minimization to ensure that the solution is
attainable in practice. The kinematic constraints that limit the Stewart
platform workspaces have been described in Section 5.1. In addition, the
optimization to determine the pose of both moving platforms is according to
the coordinate mapping in the extended configuration in Section 6.1.2. The
92
optimization will solve the coordinates of the moving platforms, FXP = [qx ,
qy , qz , θ1 , θ2 , θ3 ]T for the tool-SP and F2XP2 = [qx2 , qy2 , qz2 , θ12 θ22 , θ32 ]T
for the table-SP, given an input coordinate {P2}Xi = [x , y , z , Φ, Γ , Ω]T. In
the coordinate mapping, the value of θ32 is equal to zero, and the value of θ3
can be chosen to be zero or - Φ. Therefore, these two coordinates are fixed
and do not change during the optimization.
Simulation is carried out in MatLab using the fmincon function in the
MatLab Optimization Toolbox. The fmincon function finds a constrained
minimum of a scalar function of several variables with an initial guess and
any number of inequalities or equations. This is generally known as
constrained nonlinear optimization. This function is used to solve the
coordinates of the moving platforms with the minimum objective value as
defined in Equation 6.30. An additional constraint is added to the algorithm
so that the resulting trajectory is smooth. The algorithm will search for the
optimum pose with the lowest condition number, but within the adjacent
range from the previous pose. It is done by adding the following constraints.
( F
X Pi −1 − ∆X ) ≤ F
X Pi ≤ ( F
X Pi −1 + ∆X )
( X P 2i −1 − ∆X ) ( X P 2i −1 + ∆X )
(6.33)
F2
≤ F2
X P 2i ≤ F2
where i denotes the step number.
By limiting the range of the next pose, the trajectory will not change
drastically from one point to another point. The fmincon function only
estimates the local minimum within a certain range. If the first guess for all
the trajectory points in the optimization is set as a constant, there is a chance
that the solution will not converge to the target range specified. Therefore, the
93
previous pose is used as the first guess for the next computation. The
computation will begin from a constant guess only for the first point of the
trajectory. Figure 6.9 summarizes the algorithm.
Figure 6.9 Algorithm for motion planning in extended configuration
6.2.3 Straight-line Milling
{P2}
The proposed algorithm is tested with a sequence of Xi
corresponding to an arbitrary linear input trajectory. The values used in this
simulation are shown in Table 6.1. In addition, since the fmincon function is
based on a numerical approach, a first guess is required. In this simulation,
the first guess used was qz = qz2 = Z0/2, and the other parameters are zero ( qx
= qx2 = x0 =0 ; qy = qy2 = y0 = 0 ).
The first simulation is done without the smoothing constraint. The
result is shown in Figure 6.10 for qz. Although the proposed algorithm had
94
successfully solved the trajectory, the result is not satisfactory. From the
figure, it can be seen that there are discontinuities in these solutions. The
results of other coordinates also show similar characteristics. With the
additional smoothing constraints, the optimization with the same input was
repeated. The results are shown in Figure 6.11. No discontinuity was observed,
such that the actual motion based on this trajectory will be smooth. Hence,
with the added constraints, this trajectory planning becomes practical and
further filtering can be employed when a better trajectory is required. In
addition, a comparison of the trajectory with and without smoothing
constraints is shown in Table 6.2.
Table 6.1 Values used in the example
First guess value (qz, qZ2, θ1, θ2, 250 mm , 250 mm , 0°, 0°, 0°, 0°
θ12, θ22)
Smoothing constraint
∆X (1 mm, 1mm, 1mm, 0.25° , 0.25° ,
0.25° )
Input {F2}Xi (X, Y, Z, Φ, Γ, Ω)
Start pose 10 mm , 10 mm , 1 mm, 5°, 5°, 5°
End pose 10 mm , 10 mm , 50 mm, 5°, 5°, 5°
Z0 500 mm
x0 , y0 0 mm , 0 mm
Table 6.2 Trajectory Planning Result Summary
Without Smoothing With Smoothing
Constraints Constraints
∆qx
Mean (mm) 1.456 0.213
Standard Deviation 0.73 0.61
Mean of first derivative 0.567 0.342
(mm/s)
Standard Deviation of first 0.63 0.43
derivative
95
∆qy
Mean (mm) 2.023 0.145
Standard Deviation 0.58 0.45
Mean of first derivative 0.675 0.312
(mm/s)
Standard Deviation of first 1.13 0.87
derivative
∆qz
Mean (mm) 0.678 0.087
Standard Deviation 1.3 0.72
Mean of first derivative 2.134 1.536
(mm/s)
Standard Deviation of first 0.345 0.287
derivative
qz
Figure 6.10 Resulting trajectory plan qz from the optimization procedure
(t = step number, T = total number of steps = 50)
96
qx
(a)
qy
(b)
97
qz
(c)
Figure 6.11 A smooth trajectory from the algorithm with extra constraint;
showing: (a) qx, (b) qy, (c) qz (in mm, t = step number, T = total number of
steps = 50)
6.3 Stewart Platform Machining Framework with CAD/CAM
Software
Figure 6.12 shows a flowchart of how a machining job can be planned
and conducted for the cooperative manipulators. The arrangement of the
cooperative manipulators has been presented. The tool-SP and the table-SP
are anchored in a common frame. This is an arrangement so that both
platforms can work cooperatively to enlarge effective workspace. However,
each Stewart platform can be utilized independently or cooperatively. There
are three options that can be done with the cooperative manipulators:
98
1. Only the tool-SP moves with the cutting tool, the table-SP is
stationary. This is the same as the single configuration.
2. Only the table-SP moves with the work-piece, the tool-SP is
stationary. The table-SP must perform the motion in reverse to the
tool motion to achieve the desired result.
3. Both Stewart platforms move cooperatively. The motion of both
platforms must be coordinated. They can move one at a time or
both at the same time. This is the same as the extended
configuration.
In this research, the second option is not used. This is due to the design of the
table-SP that has a larger workspace but lower accuracy. To maintain high
accuracy and continuance in the machining operation, both Stewart platforms
move at different timings. This is due to the observation that the results are
not satisfactory because the error is too high when both SP moves at the same
time during cutting. Tool-SP is used to perform the material removal process,
while the table-SP is to move the work-piece while not cutting.
A fixture was specially designed to hold a work-piece of a maximum
90 mm by 90 mm by 70 mm on top the tool Stewart Platform or the machining
table (in the single configuration). The fixture consists of an L-shape locator,
which provides the reference point, and two clamps which provide the
clamping force. Clamps are fixed to the platform through two screws. A slot
is designed in each clamp, hence the position of the screw is adjustable and
the fixture will be able to accommodate various work-piece sizes within the
limit. The fixture can be seen in Figure 6.13.
99
Part Design
Design can also be done in Computer Aided Design
MasterCAM solids or by SolidWorks
directly defining machining
tasks
Computer Aided Manufacturing
MasterCAM
Post Processing:
Conversion to pose Computer Numerically Controlled
description of the end- Machine code
effector motion (6DOF) NC Code (text file .NC)
Store the motion data Run MATLAB
into .txt or .xls file to optimize the
tool path and
(Excel worksheet) control the
Stewart Platform,
tool, stock, and Stewart
other Platform
preparations.
Figure 6.12 Steps to machining with the cooperative manipulators
Figure 6.13 Fixture for holding work-piece on top of the tool-SP
100
A part can be designed using a CAD software. In this research, a CAD
software SolidWorks is used to design and model the mechanical objects. In
addition, a CAM software, MasterCAM is used to generate the necessary
machining tool path (NC code) to remove the material from a stock, so that it
becomes shaped as designed in CAD. There are two ways of creating a design
and generating the corresponding machining NC codes as shown in Figure
6.14.
a) Part geometry is defined in CAD (SolidWorks) and then it is saved
as Parasolid or MasterCAM Solid Feature Binary (.SFB) format
to be imported to CAM (MasterCAM). The tool path will be
generated in MasterCAM.
b) Part geometry is defined as solids in CAM (MasterCAM solids).
The tool path will be generated in MasterCAM.
The first approach is easier when the user is more familiar with
SolidWorks or other CAD software that can export Parasolid format. Because
an intermediate data file is used, some design features may not be fully
compatible or recognized by MasterCAM. The second approach is
recommended if the user is familiar with the modelling of the part directly in
CAM. This approach will not have any incompatibility issues as all the
processes are done in a single software.
In the next few sections, the first approach will be used, but with more
advanced integration between SolidWorks and MasterCAM for user- friendly
and easier procedure. This method uses an extension of MasterCAM and
SolidWorks, so that the part design and tool path generation can be done in
101
the SolidWorks user interface. This method can only process up to 3-axis
machining jobs. For 5-axis machining, the second approach must be used. In
addition, Figure 6.15 shows machining parameters that can be set for a
machining job in a CAM software.
Design CAD CAM NC Code
Idea Design
Machining Input
Parameters
(a)
Design CAM NC Code
Idea CAM Solids
Machining Input
Parameters
(b)
Figure 6.14 Information flow of part design and NC code generation
Machine tool: Cutting Parameters: Work-piece: Tool:
-Spindle Specification -Speeds -Material properties -Material properties
-dynamic behaviour -Feed, Feedrate -Shape, size, -Coated/Uncoated
-power -Depth of Cut geometry -Geometry
-machine rigidity
Figure 6.15 Machining Input Parameters
6.3.1 Tool Path Post-processing
In order to generate the tool path from the NC codes generated from
MasterCAM, it is necessary to go through post-processing, which acts as a
102
translator that reads the tool path instructions issued from MasterCAM and
writes an appropriate trajectory for the Stewart platform. The post-processor
is written in MatLab.
The MatLab program does not need commentaries or additional
information which is usually included in the NC file. In order to convert the
NC file to a compatible format, the first few lines up which contain additional
information till the line before ‘G21’ should be deleted as shown in Figure
6.16 in blue colour. Besides that, the block numbers should be removed by
choosing the menu ‘NC Functions’, and select ‘Remove Block Numbers’
similarly shown in Figure 6.17.
:0001 ( PROGRAM – TEST13 )
( DATE – 13-03-12 TIME – 17:45 )
G21
G0 G17 G40 G80 G94 G98
G0 G28 G91 Z0.
( 3. FLAT END MILL HSS TOOL -1 DIA. OFF. – 1. )
( WCS: TOP )
T1 M6
G0 G54 G90 X18.506 Y25.002 C89.851 s2546 M3
…
Figure 6.16 Additional information in a G-Code file that cannot be
processed by the MatLab post processor (blue codes)
103
Figure 6.17 Block Numbers removal in the NC code
The MatLab post-processor converts the NC codes to the input
coordinates for the cooperative manipulators according to the coordinate
mapping defined in Section 6.1. There are two steps in conducting the MatLab
post-processor. First, the program will scan the NC file and extract the G-
codes and motion types.
Secondly, the cutter locations that will be input to the MatLab
program for moving the Stewart platform must be able to be interpreted by
the Stewart platform. This process includes extraction and interpolation of
Cutter Location (CL) Data. This process continues from the previous step,
where the via points are inserted in between two subsequent CLs with a
predetermined interpolation step.
Furthermore, before conducting any machining, a test on the tool path
must be performed to check whether all the points are reachable to ensure
proper operation. The workspace simulation can be superimposed here to
result in a plot such as shown in Figure 6.18. Accessible trajectory CL points
104
are plotted continuously in blue colour. If some parts of the simulated
trajectory are plotted in red colour, then these parts are not accessible by the
Stewart platform. In the case that the whole trajectory has filled the entire
Stewart platform workspace, the machine origin must be fine-tuned to
machine this trajectory in such a way that the trajectory starting point is near
the boundary, so that the trajectory can be covered by the workspace as shown
in Figure 6.18c.
50
-50
-100
-150
-200
-250
-300
-350
100
0 100
50
0
-100 -50
-100
(a)
-270
-295
-280
-300
-290
-305
-300 -310
-315
-310
-320
-320
-325
-330 -330
-335
-340
100 100 50
50 50 50
0 0 0 0
-50 -50 -50 -50
-100 -100
(b) (c)
Figure 6.18 Testing feasibility of the resulting trajectory, (a) Inaccessible
trajectory (b) Partially accessible trajectory (c) Accessible trajectory
105
6.3.2 Determining Machine Origin
The machine origin is defined as the position (0,0,0) of the tool when
all the Stewart platform joint coordinates are zeroed (home position). This is
used as the reference for determining the start point of the trajectory. The start
point may not be at the machine origin, but its location is determined with
respect to the machine origin. The procedure to determine the machine origin
is done manually. In the single configuration, the tool-SP is moved to its home
position. Next, the distance along the base ZF axis between the tool tip and
the machining table (Z_distance) is measured (Figure 6.19). The machine
origin is then simply given by the following equations.
X_Machine_origin = X_Current_Cutter_Location;
Y_Machine_origin = Y_Current_Cutter_Location;
Z_Machine_origin = Z_Current_Cutter_Location – Z_distance; (6.34)
In the extended configuration, similar procedures can be used. The
tool-SP and the table-SP are configured in their home position before the
procedure is conducted. The distance between the tool- and the table-SP
moving platform is measured.
106
Current
Cutter
Z_distance Location
Machine
origin
Figure 6.19 Determining the machine origin
6.4 Machining Case Studies
The parts to be machines are designed in SolidWorks which is a
Parasolid-based solid modeller, and utilizes a parametric feature-based
approach to create models and assemblies. Parameters refer to constraints
which values determine the shape or geometry of the model or assembly.
Parameters can be either numeric parameters, such as line lengths or circle
diameters, or geometric parameters, such as tangent, parallel, concentric,
horizontal or vertical, etc.
6.4.1 Machining an ‘NUS’ Pocket
After the stock has been defined, a sketch is created on the top surface
(Figure 6.20). The Extruded Cut feature is used to generate the pocket. The
tool path is generated using the MasterCAM Feature Based tool path function
107
that will recognize any feature after the stock dimension has been defined.
The cutter diameter is also selectable in the interface. Each tool path group
can be exported to a separate NC file if needed, especially when machining
curve or complex surfaces with a large number of data points, because one
whole chunk of the tool paths will involve huge amount of memory for
processing in MatLab and may halt the Stewart platform in the middle of an
operation. The resulting trajectory and the actual machining are shown in
Figure 6.21.
Figure 6.20 Sketch for the ‘NUS’ pocket on top of the stock model
108
(a)
(b)
Figure 6.21 (a) The tool-path generated (b) Actual machining
6.4.2 Machining a Dome
Using SolidWorks cut extrude followed by boss extrude will form a
base as shown in Figure 6.22a. A dome-shaped island is added on top of the
rectangular base. The machining is done in two stages: roughing and finishing.
Accordingly, the MasterCAM generates one tool path group for each tool path
109
group. The roughing creates a pocket with a rectangular island. The finishing
produces a tool path trajectory with high density of coordinates to machine
the dome shape (Figure 6.23). The finishing tool path is generated based on
the dome surface, which is selectable in the interface (Figure 6.22b). In
addition, the tool path must be contained in the pocket to make sure no tool
path is generated outside the pocket. The tool containment is defined as a
closed curve in Figure 6.24.
110
(a) Pocket with an island
Surface to be machined
(b) Dome on top of the island
Figure 6.22 Machining a dome (wire frame display)
111
(a) Tool path
(b) Result
Figure 6.23 Machining a dome
Figure 6.24 Tool containment constraint
112
6.4.3 Machining a Test Part
A part for testing 2.5D machining on the cooperative manipulators is
designed as shown in Figure 6.25. Machining consists of several stages, in
each stage a feature is machined. The depth of the cut is selected within the
safe limit of the tool. Machining was done after simulation in MasterCAM.
The machined work-piece is shown in Figure 6.26. It can be seen that the
cooperative manipulators can machine such a part with a good finish. The
cooperative manipulators was further evaluated by machining another test
part with thin wall features (Figure 6.27). Thin walls with thickness of 0.1
mm were machined by the cooperative manipulators. It was observed that thin
walls with lower thickness could also be machined. However, due to the
brittleness of the material, the thin walls collapsed as the tool passed through
them during machining. In addition, to test 3D machining, a 3D part is
designed as shown in Figure 6.28. With two stages of machining, facing and
finishing cut, the resulting cut can be seen in Figure 6.29.
113
(a) Part model and machining tool path
(b) Machining simulation
Figure 6.25 The first test part comprising of 2-axes machining features
114
Figure 6.26 Machined test part
(a) Part model and machining tool path
115
(b) Machining results
Figure 6.27 The second test part with thin wall features
(a) Part model and machining tool path
(b) Machining simulation
Figure 6.28 The part with 3-axis machining
116
Figure 6.29 Three-axis machining result
6.4.4 Machining with Rotation Axes
Due to the limitation of the integration of SolidWorks and
MasterCAM, machining with rotation axes, i.e., 4- or 5- axes machining, must
be done entirely in MasterCAM. The rotation axes defined in MasterCAM
must be compatible with the MatLab post-processor. Therefore, a specific
tool path processor in MasterCAM is selected as the active post-processor for
five-axis machining. This tool path processor has the same rotation axes and
sequence as defined in the coordinate mapping for the Stewart platform.
A tapered circular island has been designed as shown in Figure 6.30.
The wall of the island has an angle of five degrees. By machining this part,
the Stewart platform can demonstrate the accessibility of the cutting tool to
reach the tool nutation angle of five degrees while the tool precession angle
changes continuously in interval [0, 360°]. The resulting work-piece is shown
in Figure 6.31. In addition, an S-shape swarf wall was also designed as shown
117
in Figure 6.32. This shape could be a component of a more complex part such
as a turbine blade.
(a) SolidWorks model (b) MasterCAM tool path
Figure 6.30 Circular Tapered Island
Figure 6.31 Machining result of tapered circular island on wax
118
(a) 3D model of the part with the S-shape swarf (wire frame display)
(b) Machining result on wax
Figure 6.32 Machining a S-shape swarf
119
CHAPTER 7
STEWART PLATFORM MACHINING
OPTIMIZATION AND EVALUATION
7.1 Machining Workspace Analysis
The machining workspace is characterized by the locations on the
work-piece that can be reached by the tool on the tool-SP. The work-piece
itself is moved by the table-SP. Therefore, this is considered the farthest
distant points reachable for every direction of the tool in the combined
workspace evaluation. In addition, the distance between the bases of the two
Stewart platforms is 1200 mm and the distance from the tool-SP moving
platform origin to the tool tip is 150 mm. These data are based on the actual
condition of the cooperative manipulators. To simplify the analysis, inscribed
circles are used among the layers of the workspace plot as shown in Figure
7.1. This approximation guarantees that all the points in the inscribed circles
are inside the actual workspace. Thus, it is named the inscribed workspace.
Using the inscribed workspace, it is simplified to analyse the workspace of
the cooperative manipulators.
The workspace of the cooperative manipulators is defined as the
collection of tool location relative to the object resulting from the movement
of the tool-SP and the table-SP. To analyse the cooperative manipulators
workspace, every combination of a location in the tool-SP workspace and a
120
location in the table-SP workspace is tested. Consider Figure 7.2, for an
instance of a 3D translation workspace combination (zero rotation). The
volume of the cooperative manipulators (the combined workspace) is
significantly larger compared to the volume of workspace of the tool-SP. This
is due to the difference between the much bigger translational workspace of
the table-SP. The table-SP can position the work-piece such that the tool-SP
can reach farther location on the work-piece.
It is interesting to analyse the workspace for 4- or 5- axis machining.
In order to define workspace for five-axis machining, the following
description is used. While the tool tip reaches the point Tc in the workspace,
the tool tilt angle can be changed in the conic face of β (Figure 7.3) . Based
on this, the smallest reachable nutation angle (βmin) is defined as the smallest
value of the nutation angle of the moving platform while the tool tip is at point
Tc and the precession angle α changes continuously in interval [0, 360°]. The
machining workspace is then defined as the collection of points that the tool
tip can reach given a value of the smallest reachable nutation angle, βmin
(Wang et al. 2001).
The machining workspaces of the cooperative manipulators for
various tilt angles of the tool axis have been simulated. The result is presented
in Table 7.1.
121
60
40
Y (mm) 20
-20
-40
-60
-60 -40 -20 0 20 40 60
X (mm)
Figure 7.1 Inscribed circle (blue) inside the workspace boundary (2D
workspace)
800
750
700
650
600
550
500
200 200
0 0
-200 -200
(a) Workspace of the table-SP
-440
-450
-460
-470
-480
-490
50
50
0
0
-50 -50
(b) Workspace of the tool-SP
122
250
200
150
100
50
0
400 400
200 200
0 0
-200 -200
-400 -400
(c) Combined Workspace of both SPs
Figure 7.2 Combination of 3D translation workspace
Change curve of
minimal reachable
nutation angle βmin
Change curve of
maximal reachable
nutation angle βmax
Tool axis
Tc(x,y,z)
Figure 7.3 Tool axis pose in the workspace analysis
123
Table 7.1 Workspace Volume with various tilt angles of the tool axis
Single Tool-SP Extended
configuration
3D Workspace 331,220 mm3 33,060,000 mm3
(βmin = 0)
βmin = 1 degree 218,650 mm3 31,823,000mm3
βmin = 2 degree 167,140 mm3 29,926,000mm3
βmin = 3 degree 121,900mm3 28,750,000mm3
βmin = 4 degree 72,270mm3 26,413,000mm3
βmin = 5 degree 41,488mm3 25,056,000mm3
βmin = 6 degree 17,618mm3 23,318,000mm3
βmin = 7 degree 3,801mm3 21,696,000mm3
7.2 Application of workspace data for optimal setup in machining
For optimal machining, workspace data must be considered to utilize
the full capabilities of the cooperative manipulators. The application of the
information when developing a process plan and tool paths is examined. Since
the workspace of the tool-SP is not a rectangle and it changes size and shape
based on the tool angle, a visual method is called for. The method is to use
the 3D plot of the tool-SP workspace for a given tool angle in the CAD/CAM
system to choose the location and orientation of the work-piece. The work-
piece is stationary in the single configuration and it is fixed on the machining
table. In the extended configuration, the work-piece is carried by the moving
platform of the table-SP. Therefore, the work-piece location and orientation
can change during machining which allows the tool-SP to access a larger
volume.
124
Several parameters influence the work volume as it is applied to
process planning and tool path generation. First, the volume is based on the
position of the moving platform, and the orientation of the tool axis vector. If
the tool is not tilted with respect to the table, three-axis machining, the
workspace takes on a different shape, in this case larger. Large tool axis vector
tilt angles would be required to remove the needed material, if the volume is
much smaller. This is the case because positioning uses a finite resource, the
available lengths of the legs.
Because only five DOF motions are used for machining, the rest of
the DOFs can be used for optimizing certain indices. However, to minimize
error in the motion and maintain accuracy in machining, it is advised that only
the tool-SP is used to cut the material and the table-SP is stationary in
processing a trajectory. The focus of the table-SP is to provide larger
workspace for the tool-SP if required. Therefore, the table-SP is used to
position and orientate the object or work-piece with respect to the tool-SP and
the tool-SP is used to perform the intended operation or the cutting of the
material.
7.3 Calibration and Accuracy Improvement
Manufacturing tolerances of building a manipulator often cause an
inherent problem where the kinematic parameters are not exactly equal to the
values in the kinematic model. Kinematic calibration is a process of
identifying the actual values of the kinematic parameters in the kinematic
model. Thus, by updating these parameters, the inverse kinematic calculation
125
of the required joint angles will result in an accurate end-effector pose. The
calibration process generally consists of four basic steps, namely (1)
development of a kinematic model that contains a set of parameters to
determine the relationship between the actuated joint angles and the end-
effector pose, (2) measurement and recording of the manipulator poses, (3)
error minimization through searching for the optimum kinematic model
parameters of the manipulator from the pose measurements and manipulator
actuated joint angles, and (4) correction for the geometric parameter errors in
the manipulator kinematic model.
The kinematics formulas are crucial in the control of a PKM to move
its end-effector to a desired location. However, this requires the PKM to be
built according to the nominal kinematic parameters, such as joint locations
and leg offsets. The main purpose of the calibration is to find the actual
kinematic parameters that have deviated from their nominal values due to the
imperfect assembly and manufacturing tolerances.
The first step of calibration is error modelling. Error modelling is
important for profiling geometric entities that cause motion errors of the end-
effector. Error modelling is the process of determining the parameters that
affect the motion of the end-effector. Errors considered in the calibration are
geometric errors and they are treated as static values or constants. The
following explanation uses the designation for the tool-SP, although the same
derivation can be made for the table-SP.
It is generally accepted that 42 kinematic parameters can fully
describe a configuration of a nominal Stewart platform (Wang and Masory
126
1995). These kinematic parameters are of main importance and they are
namely, the leg length offset LOi (one parameter), locations of the universal
joints, bi (three parameters each), and locations of the spherical joints, ai
(three parameters each). Hence, there are seven kinematic parameters per leg.
Kinematic calibration can solve kinematic parameters with respect to
an arbitrarily fixed coordinate system located at the base frame {F} and the
moving platform {P}. The absolute locations of the origins of frame {F} and
frame {P} have no effect on the motion error, because the kinematic
parameter errors are defined within the Stewart platform closed kinematics
loops relative to these origins. Therefore, six parameters that are defined with
respect to frame {F} and six parameters with respect to frame {P} can be
eliminated or excluded from the calibration. In the calibration, the parameters
b1, b2, Pa1, Pa2 are set to the nominal/reference values and the other 30
parameters define the kinematic configuration of the Stewart platform.
Measurement data are taken from the digital indicators during
calibration. These digital indicators provide additional redundant sensing
which is used to calibrate the Stewart platform. The digital indicators are
installed on the machining table (single configuration) with reference to the
world coordinate system. There is no specific constraint on the position of the
digital indicators; however, they must be installed with the axes of the gauges
perpendicular to the table. The purpose is to measure the distance along the
axes of the gauges between the machining table and the moving platform.
Figure 7.4 shows an illustration of the Stewart platform with the setup. The
coordinate frames {T} and {W}, refer to the tool coordinate system and the
127
world coordinate system, respectively. The digital indicator measurement is
achieved relative to the world coordinate system. The world coordinate frame,
{W} is attached on the machining table on which the digital indicators are
installed, as the reference for the measurement.
(a) 3D model
128
(b) Coordinate systems
(c) Dial gauges for measurement
Figure 7.4 Calibration setup
129
A numerical nonlinear gradient descent optimization method is used
to solve for the real kinematic parameters from the measurement data. The
digital indicators are measurement devices to verify the location of the
moving platform and determine the error between the desired and the actual
locations. The readings of the digital indicators give different values than the
expected values (nominal values) if there is a discrepancy or an error in the
moving platform location. An error at one pose can be formulated as Yi – Fi(Qi,
η), i=1..N, where η denotes the real (unknown and to be found) kinematic
parameters, Yi is the measurement vector taken from the readings of the
indicators, and Qi is the leg length computed from the inverse kinematics
corresponding to the ith pose when the measurement is taken. Fi is a function
that gives the nominal values of readings of the digital indicators when the
moving platform is at the desired location. When there is no error Yi is equal
to Fi. In other words, if there is no difference in the nominal kinematic
parameters and the real kinematic parameters, there would be no kinematic
errors. The calibration is solved by optimization that is formulated as
Equation (7.1).
Y1 − F1 ( Qi ,η )
min ... (7.1)
η
YN − FN ( Qi ,η )
N is the number of measurement poses taken. Equation 7.1 will
minimize the error by finding the real kinematic parameters. This equation is
also known as the Least Squares Formulation (LSF). The LSF procedure
130
determines the real kinematic parameters, which are initially set according to
the nominal parameters of the Stewart platform. When the errors in Equation
7.1 are minimized, the set of kinematic parameters that best represent the
actual kinematic parameters of the Stewart platform can be found.
The forward kinematics calculation is performed in each function, Fi
to obtain the expected values of readings of the digital indicators. The
function Fi in Equation (7.1) gives the pose of the moving platform from the
leg lengths, Qi and a set of kinematic parameters, η. The function Fi computes
the expected digital indicator readings based on the machining table location,
on which the frame {W} is attached, with respect to the base coordinate frame
{F}. Using the end-effector location that has been calculated, the expected
digital indicator readings are determined by finding the distances between the
machining table and the moving platform.
Simulation is carried out to verify the calibration method before
applying the method to the actual Stewart platform calibration with real data.
In the simulation, a real model is assumed and used to test the calibration
method. Table 7.2 summarizes the deviations of the real model from the
nominal model. The nominal model is used for initialization of the kinematic
parameters at the start of the error minimization in the LSF. Table 7.3 presents
the various simulation results with a random error applied in the measurement
data. This error emulates the noise and disturbance in the measurement. The
error is modelled as the Gaussian white noise with the corresponding
variances (noise level) listed in Table 7.3.
131
The calibration process has been performed in several stages to
improve and obtain the best geometric parameters that can minimize the
errors. Some measurement errors due to instability were observed during the
calibration process where the data obtained for the same pose differed by
around 10-50 micrometre, depending on the previous state of the Stewart
platform. This could be caused by the backlash in the passive joints. To
compensate this error, two sets of data were taken for each configuration
during measurement and the average was recorded. The result shows the
calibration using optimally selected poses gives the least error. Table 7.4
summarizes the calibration result with a set of 110 poses.
Note that each SP is calibrated with respect to the local coordinate
frame which origin is at the SP’s base. While it is true that the calibration of
each SP may yield worse result than the simultaneous calibration of both SPs,
the latter is actually not feasible for practical reasons. For the calibration to
work, the coordinate frame must be fixed accurately. In the setup of 2-SP
inside a custom-made frame, the relative location of both platforms with
respect to each other is not known beforehand. Although the values can be
approximated, a small error will lead to wrong geometric parameters. In
addition, for simultaneous calibration over 70-80 parameters will need to be
optimized. The number of parameters is quite large and the optimization has
low chance to find good representative geometric parameters, especially if
approximation is used.
132
Table 7.2 Real model for calibration simulation
Assumed leg lengths offset error (mm)
δLO1 δLO2 δLO3 δLO4 δLO5 δLO6
1 2 3 1 2 3
Assumed position error of the base joints (universal joints) (mm)
δb1 δb2 δb3 δb4 δb5 δb6
X 0 3 0.1 -0.1 -0.1 -0.1
Y 0 0 1.8 1.8 1.8 1.8
Z 0 0.3 0.7 0.7 0.7 0.7
Assumed position error of the platform joints (spherical joints) (mm)
δa1 δa2 δa3 δa4 δa5 δa6
X 0 -0.9 3 3 3 3
Y 0 0.67 -2 -2 -2 -2
Z 0 0.53 0.7 0.7 0.7 0.7
Table 7.3 Error comparison of calibration simulation
NL EE DE PEb Pea Per
0.0001 99.99% 99.81% 0.12, 3.11, 0.0006 ,0.001 99.41%, 99.97%,
1.27 ,0.002 99.85%
0.001 99.95% 99.57% 0.12, 2.99, 0.002, 0.004, 98.28%, 99.87%,
1.26 0.003 99.79%
0.01 99.48% 97.78% 0.12, 2.78, 0.01, 0.006, 0.023 90.90%, 99.78%,
1.24 98.13%
0.1 93.80% 94.61% 0.12, 3.24, 0.014, 0.05, 0.04 88.15%, 98.59%,
1.23 96.60%
Column Info: NL: Noise Level (variance) , EE: reduction in estimation error
(sum of squares of F1..FN), DE: reduction in mean difference between
nominal and real digital indicators reading , PE(x): pose error mean in α
(degree), β (degree), and Fz0 (mm), respectively (PEb:before, PEa:after, PEr:
error reduction).
133
Table 7.4 Kinematic Parameters after Calibration
Leg lengths offset (mm)
LO1 LO2 LO3 LO4 LO5 LO6
1 2 3 1 2 3
Position of the base joints (universal joints) (mm)
b1 b2 b3 b4 b5 b6
X -152.5 -129.069 76.15 131.9689 76.15 -0.1
Y 0.0 76.25 130.269 -74.45 133.869 154.3
Z 0.0 0.3 0.7 0.7 0.7 0.7
Position error of the platform joints (spherical joints) (mm)
P P P P P P
a1 a2 a3 a4 a5 a6
X -60.622 -0.9 3 73 63.22 -32
Y 35 0.67 -72 -2 33 58.622
Z 0.0 0.53 0.7 0.7 0.7 0.7
7.3.1 Perpendicularity of Dial Gauges
In the case that the dial gauges are not perfectly mounted onto the
external plate, additional parameters representing the error angle on the
perpendicularity for each gauge, ϒi , i=1..3. This error angle can be described
in Figure 7.5. The dial gauge measurement axes are supposed to be parallel
to ZW. However, due to assembly error and manufacturing tolerance, it may
have deviation. Thus, the actual distance reading from each gauge should be
corrected by a factor of cos ϒi for each dial gauge. These parameters can be
included in the optimization to solve the problem as given in the following
equation.
Fi(Q0 + ∆Q,Y0 + ∆Y, η0 + ∆η), i=1..N (7.2)
where the superscripts 0 indicate that the variables or parameters are at their
nominal or supposed values (measured or selected) and ϒi is included in the
measurement error ΔYi = Yi – Yi.cos(ϒi). An extended Kalman Filter or also
known as the Total Least Squares Formulation (TLSF) method is proposed to
134
counter this problem. A priori notions of accuracy of the measurement that
can be modeled using statistical approach with a maximum-likelihood
principle are considered in the TLSF. In this context, the LSF does not take
into account measurement and input errors, ∆Q = 0 and ∆Y = 0. In other
words, it is assumed that the measurement is perfect and the real leg lengths
are known without any discrepancies. Consequently, it only finds the ∆η that
minimizes the discrepancies.
Vector parallel to ZW
Measurement axes
Zw ϒi
Yw External Plate
Xw
Figure 7.5 Error angle on dial gauges
On the other hand, the TLSF needs to consider non-zero ∆Q and ∆Y.
If this is the case, the TLSF must find the best combination of ∆Q, ∆Y, and
∆η. Because these errors are assumed to be Gaussian with zero mean, the best
combination is the minimizer of (Patel and Ehmann 2000; Wampler et al.
1995):
N
χ2 = ∑ ∆Q W ∆Q + ∆Y
i =1
T
Q
T
WY ∆Y + ∆η T Wη ∆η (7.3)
135
where WQ, WY, Wη can be obtained from the inverse of the covariance
matrices for ∆Q, ∆Y , and ∆η , respectively. Thus, TLSF minimizes (7.3)
subject to (7.1). With the TLSF, it is ready to calibrate the real case with
imperfect information from measurements. However, prior to real calibration,
TLSF simulations are performed with the same settings with the previous
simulation using the LSF. Equation (14) shows that the TLSF tries to guess
what is happening in the real case where errors are not just caused as a result
from only the kinematic parameters deviation, but also error angle, noises and
other un-modeled errors in terms of joints coordinates. Table 7.5 summarizes
the calibration results with error angles imposed on the dial gauges.
Table 7.5 Error comparison of calibration with measurement errors
NL EE DE PEb PEa PEr
0 (LSF) 99.92% 97. 703% 1.09, 0.61, 0.021, 0.028, 98.09%,
1.26 0.018 95.31%, 98.6%
0.0001 92.67% 66.76% 2.96, 2.99, 1.98, 2.93, 32.88%, 2.12%,
2.92 2.18 25.48%
0.001 99.52% 71.53% 2.54, 2.81, 2.06, 2.62, 18.92%, 6.99%,
3.15 1.97 37.54%
0.01 99.05% 67.02% 2.28, 2.73, 2.11, 2.78, 7.29%, -1.73%,
3.38 1.84 45.61%
0.1 50.25% 26.64% 1.903, 2.79, 1.95, 2.76, -2.51%, 1.03%,
1.82 1.65 8.82%
Column Info: NL: Noise Level (variance) , EE: reduction in estimation error
(sum of squares of F1..FN), DE: reduction in mean difference between
nominal and real digital indicators reading , PE(x): pose error mean in α
(degree), β (degree), and Fz0 (mm), respectively (PEb:before, PEa:after, PEr:
error reduction).
136
7.3.2 Pose selection for Calibration
When the calibration is being performed, measurement data are
collected from the various poses or configurations of the PKM. These data,
however, can be affected by noise and physical instability. The set of
measurement poses can be optimized to yield robust calibration. The quality
of a certain set of configurations/poses with respect to the calibration can be
approximated based on an observability index that can be obtained from the
identification Jacobian matrix, JP. This matrix has components that are
calculated using the linearized version of Equation (7.1)
∆Y = J P . ( X P ) ∆η (7.4)
In Equation (7.4), ∆Y is the difference in the measurement variable
(digital indicator readings) when an error ∆η is induced in the kinematic
parameters, and XP is the pose set (collection of the moving platform location
where measurements are taken) used in the calibration. The identification
Jacobian matrix, JP, can be computed numerically by assuming small
difference in ∆η (Huang et al. 1998). There are four options that have been
proposed in the literature for the definition of the observability index from the
matrix JP (Nahvi and Hollerbach 1996; Daney 2002). One of the best indices
that relate well to error calibration sensitivity is the Noise Amplification
Index, O given in equation (5), where σi are the singular values of JP ordered
from largest to smallest so that σ1 >= σ n.
O = σn2 / σ1. (7.5)
137
In this research, a swarm-based PSO (Particle Swam Optimization)
(Kennedy and Eberhart 1995) search method shown in Figure 7.6 is proposed
to search for the best set of poses for calibration measurement. The PSO
method models the optimum candidates as moving particles over the search
domain, which is the PKM workspace. The aim is to find a pose set within
the workspace of the PKM that maximizes the objective function O(JP(XP)),
where O() is a function to compute the observability index of JP(XP), which
is the identification Jacobian matrix characterizing the error relationship
between the kinematic errors and the measurement variables. The framework
of this algorithm is a modification of the one proposed in (Daney et al. 2005),
but with PSO as its search method. The steps are as follows:
1. Initial set, X1 … XN for N poses needed for selection, is formed by
random poses in the workspace.
2. AddFind step: Using PSO, find an additional XN+1, that maximizes
O(JP(XP)) where XP includes X1 … XN, and XN+1.
a. To find XN+1, randomize the particles over an allowable range of
the search domain. Each particle represents a solution candidate
for XN+1.
b. Compute the best position of each particle and find the global best
particle position (the best candidate that gives the highest
objective value among all the particles). The objective value is
computed from O(JP(XP)) where X includes X1 … XN, and the pose
represented by the position of each particle being evaluated.
138
c. Update the particle position (move the solution candidates to new
positions) based on standard PSO governing equation using the
information from the previous step.
d. Stop if the global best objective value is not improving for several
iterations or the number of iterations exceeds a predefined limit.
The XN+1 found is the pose represented by the last global best
particle position.
e. RemoveFind step: Find an X- within the set X1 … X N+1 that if this
X- is removed from the set, the remaining poses give a higher
objective value of O(JP(X)) where X includes all poses in X1 …
XN+1 except X-.
4. Repeat steps (2) and (3) until X- = XN+1, which implies that the
newly added pose is not improving the current pose set, and stop.
The initial pose set is initialized randomly over the workspace. Next,
the PSO is conducted to search for one additional pose such that if this pose
is added to the group, the group will have the best performance index; this is
the AddFind step. Immediately after a new pose has been added to the group,
the RemoveFind step is executed to find one pose of the group that if this pose
is removed, the group will have the best objective value. These two main
processes, AddFind and RemoveFind, are repeated until no better objective
value for the pose set can be found. The PSO search domain is defined by the
boundary of the Stewart Platform PKM workspace. In addition, the optimum
configurations are often found near the boundary of the workspace. This
139
implies that the initial guesses of the PSO search can be taken near the
boundary to expedite the search process.
Figure 7.7 shows the global best performance index (observability)
optimized in the AddFind process based on the global best pose. The PSO
algorithm is stopped after 40 iterations and it is concluded the pose with the
global best performance as the candidate to be added to the pose group in
AddFind. RemoveFind is implemented sequentially to check each and every
pose in the group by calculating the index. If one pose is added, and then one
pose is removed in the AddFind-RemoveFind cycle. If the pose to be added
and the pose to be removed are the same, then the search for pose stops.
Initialize a AddFind: RemoveFind: Stop if a
random set of Add a pose Remove a better set
N poses to the set pose from the cannot be
selections set found
PSO search method
Search Iteration: Stop if
Initialize Evaluate current termination
random particles condition is true
candidates for a Move particles to and return the
new pose over new positions best candidate as
the workspace the new pose
Maximal
Figure 7.6 PSO algorithm for optimum poses selection
140
-34
x 10
10
7
Global
6
Best
5
2
0 5 10 15 20 25 30 35 40
Figure 7.7 Global best evaluation in PSO search
The number of poses required to obtain a good calibration result has
been investigated. Figure 7.8 presents the plot of the error in the kinematic
parameters after calibration against the number of pose measurements. In this
test, random poses are used. It can be concluded that a larger number of poses
will lead to better calibration results. In addition, there is a limit where the
errors cannot be reduced further with additional number of poses.
141
-5
x 10
1
mean
0.9 Standard Deviation
0.8
0.7
0.6
Error in mm
0.5
0.4
0.3
0.2
0.1
0
20 30 40 50 60 70 80 90 100 110 120
Number of poses
Figure 7.8 Calibration improvement with a larger number of poses
7.3.3 Online Calibration for Kinematic Parameters Error
Compensation
The proposed calibration has potential for automated calibration as
long as the selected poses are measurable and reachable. The calibration
procedure presented in the previous section is extended to form a self-
calibration procedure to improve the accuracy of the PKM when it is in
operation. The measurement method for the self-calibration procedure is
using the draw-wire sensors installed between the base and the moving
platform of the Stewart platform. The objective is to form a passive 6-legged
mechanism using the wire sensors, so that the wire sensor readings can fully
capture the moving platform pose. As the wire sensors are used for the
calibration procedure, there is no interference between the sensors and the
142
Stewart platform. This arrangement allows machining and measurement to be
performed at the same time (Figure 7.9).
Draw-wire
sensor pool
Figure 7.9 Draw-Wire Sensor Assembled to tool-SP
Under a constant voltage input, the draw-wire sensor will have
different output according to the length of the wire that has been drawn out
of the sensor. The output is a voltage signal instead of the reading of the wire
length. Hence, a calibration of the draw-wire sensor is also needed to find the
relationship between voltage output and the wire length as well as to check
the linearity of the output.
143
In order to stabilize the voltage input to have a steady result, an
external power supply is used. A height gauge with an accuracy resolution of
10 μm is used to calibrate the wire sensor. The results shown in Figure 7.10
show good linearity. With these data, the bias offset in voltage reading is
obtained. With interpolation or extrapolation, the length of the drawn wire
can be calculated.
Figure 7.10 Relationship between voltage output and Draw-Wire sensor
length
7.4 Machining Evaluation
A circular pocket was created using the cooperative manipulators
before and after the calibration. Early investigation shows that the accuracy
of the Stewart platform differs significantly such that it can be seen in the
shape of the work-piece in Figure 7.11. Before calibration, the shape is more
like an ellipse than a circle. The accuracy of Stewart platform improves after
144
calibration took place. The updated kinematic parameters resulted in a better
shape for the created part.
(a) before calibration
(b) after calibration
Figure 7.11 A circular pocket part during calibration
Further, facing operation was performed to observe the smoothness of
the surface such as shown in Figure 7.12. With careful selection of the
145
distance between two subsequent via points, the interpolation can give
sufficient number of points in order for the tool-SP to machine a good surface
finish.
Figure 7.12 Facing operation of a work-piece
In addition, although machining can be performed using the
cooperative manipulators without any problems, the chips accumulated from
the work-piece could not be removed automatically. In this cooperative
146
manipulators, the author tried his best to clear off the chips during machining
(Figure 7.13). The material of the work-piece is relatively soft compared to
metal. Hence, no tool jam was observed although the chips were piling around
the work-piece during machining. However, for metal cutting the chips could
jam the tool. An automated chips removal system may need to be installed on
the cooperative manipulators for metal cutting. In this research, the study is
limited to machining wax because it is soft and can be machined quite easily.
The machined part can reflect the tool movement of the cooperative
manipulators and can thus ensure safe operation.
Figure 7.13 Chip accumulation during machining
In order to test the rotation capability of the tool-SP (in single
configuration), a part with a tapered wall was machined. The inclination angle
of the wall is 10 degrees. The maximum angle in one axis of rotation of the
tool-SP is about 13-15 degrees. The result is shown in Figure 7.14. In addition,
147
Figure 7.15 shows the tool-SP when it was machining the tapered wall. As
can be seen, at the configuration during machining the tapered wall, the tool-
SP has almost reached the limit of some of its ball-socket joints. In fact, the
ball-socket joint constraint is the main limitation for the rotation of the tool.
Thus, the table-SP in the extended configuration could be used to position the
work-piece to allow the tool to approach the work-piece with a larger rotation
angle.
Figure 7.14 A tapered wall machined from the most tilted angle of the tool
of the tool-SP
148
Figure 7.15 The tool-SP in its most tilted configuration
A 2 ½D work-piece, which is the dome shaped part, was designed to
verify the error of the calibration process. This verification method includes
the identification of error factors from the Stewart platform kinematics and
dynamics errors during machining. A CMM with a pre-calibrated touch
trigger probe was used to measure points on the machined surface. The CMM
was programmed to follow curves on the machined profile, which is
generated from the local part coordinates obtained from the vertices of the
faces forming the surface of the work-piece as shown in Figure 7.16. Figure
7.17 shows the 3D plot of the measured points on the surface. The mean of
149
the absolute deviation from the defined work-piece before and after
calibration is less than 0.5 mm and 0.05 mm respectively.
7.5
6.5
5.5
5
10
10
8
6
4
0 2
0
Figure 7.16 Surface model and extracted vertices of the dome shaped test
part
8.5
7.5
6.5
5.5
10
5
0 5
2
4
6
8 0
10
(a) Before calibration
150
8.5
7.5
6.5
5.5
5
10
10
8
5 6
4
2
0 0
(b) After calibration
Figure 7.17 Comparison of surface measurement
7.5 Stewart Platforms Evaluation
Since the emergence of the first Stewart platform based machine tool
decades ago, researchers are investigating whether the full potential of the
PKM will actually change the machining processes. The promising
characteristics that have been observed in Stewart platforms are higher
payload-to-weight ratio, noncumulative joint error, higher structural rigidity,
and modularity. Nevertheless, the design of these PKMs is the key to bring
out these benefits to the machining industry. This researcher has experienced
difficulties in finding good components, especially the passive joints of the
Stewart platform. In addition, the frame in which the Stewart platforms are
built must be rigid to maintain accuracy after calibration. Furthermore, the
connection between components of the Stewart platforms creates additional
error on top of these errors.
Due to these limitations, the micrometre machining accuracy is
difficult to be achieved with the current setup. The tool-SP was built with the
best joints in the market, and presently, it has the highest accuracy among the
151
others. However, it is still below the performance of conventional CNC
machines. Finally, the accuracy is not satisfactory because a few main causes:
the vibration of the frame structure, tool holder misalignment, and mostly
from the spherical joint clearance of the table-SP which produces
uncontrollable motion at its end-effector.
In addition, there are a few aspects which are worth mentioning in this
evaluation. While it is generally accepted that PKMs have a low workspace-
to-footprint ratio, simply adding some automation components could increase
the workspace efficiently. For example, an additional rotary table with one or
two DOF would increase the tool approach angle of the tool-SP. The rotary
table could add more flexibility compared to the table-SP. Since the tool-SP
workspace is strongly affected by the tool orientation, the rotary table could
replace the orientation capability of the platform, making the job easier and
maintain the translational workspace of the tool-SP to maximum at all times.
This is similar to the decoupling concept in the serial manipulator, where the
hand performs all the translation and the wrist performs all the rotation
required by the task. Fortunately, the table-SP can act as a multi-DOF table.
Thus, one can see the second SP not as another manipulator, but as an
automation component to help the main tool-SP perform its job.
Calibration of a Stewart platform to eliminate its static error is not
easy. The calibration of PKMs has predominantly been evaluated in terms of
analyses using kinematic-based models. Basically, the errors come from
imperfect knowledge of manipulators’ parameters. Some problematic error
sources that are hard to address include localized heat generation in the
152
screw/nut system, axis errors, and nonlinear kinematic mapping errors. It is
interesting that the calibration issue is also related to stiffness analysis, work-
piece placement, and controller development. All of these other issues can
affect the accuracy in different ways. Due to its complexity and time-
consuming calibration, the introduction of PKM in the manufacturing context
needs the development of a fast and automated calibration technique.
The control system of a manipulator is also a crucial issue. The control
of a PKM is important when dealing with the effect of leg inertia and
dynamics behaviour. What is required for the control system of a PKM is the
possibility of introducing a real-time, singularity-free path as well as
workspace verification. This should also be done a priori in the CAM
simulation and NC verification procedures. In addition, it is necessary to
check that the trajectory does not exceed the workspace limits. The
performance of a PKM varies over its workspace, thus when a tool-path is
executed, optimization is needed to assure a smooth machining operation to
provide a maximum, stiffness along the machining path between the work-
piece and the tool.
153
CHAPTER 8
CONCLUSIONS AND RECOMMENDATIONS
8.1 Conclusions
In this thesis, an introduction to the closed-chain mechanism of PKMs
has been presented, including its brief history, literature review, applications,
and comparison with serial kinematic manipulators. Compared to serial
kinematic manipulators, Stewart platforms have the advantages of higher
structural stiffness, larger payload capacity and lower friction. The primary
shortcoming of Stewart platforms is the limited workspace.
In this research, the development of two Stewart platforms has been
presented and they can be applied in various applications, such as machining
and positioning. The focus is to control and plan the motion of Stewart
platforms to achieve better performance compared to serial counterparts.
Cooperative manipulators consisting of two Stewart platforms has
been fabricated and developed in this research. A frame is designed for the
two Stewart platforms to be installed, such that one Stewart platform is
installed in a normal configuration and a second Stewart platform is installed
in an inverted configuration, such that they can be used as a hybrid
mechanism to perform a task cooperatively. Various components for
calibration and machining purposes are installed and attached on this frame,
154
such as wire-sensors, camera, cabling compartments, machine tool
attachment point, and other peripherals that are needed for the control systems.
Software development in this research includes a graphical user
interface which uses a mouse and a joystick as devices to control the motion
of the Stewart platforms. In addition, a user can control the movement of the
Stewart platforms by providing input parameters in terms of the coordinates
in the various coordinate systems, namely, the pose of the moving platform
and the joint coordinates of the actuators. Inverse and forward kinematics
analyses are included in this control system to observe the motion of the
Stewart platforms continuously. The inverse kinematics is found to have a
direct solution while the forward kinematics can be solved by using the
Newton-Raphson numerical method. A graphical simulation program based
on M-file programming and MatLab SimMechanics has been developed and
used as a valuable design tool to investigate the effects of geometric
parameters and joint constraints on the motion of the Stewart platforms and
to provide useful information about the workspace, joint angle, and dynamics
of the leg lengths. The graphical simulation program is also used with the user
interface to simulate the experiments, check software malfunction and
validate the control process.
The approach based on least squares formulation (LSF) is developed
to correct the positioning of the moving platform of the Stewart platforms and
compensate for the deviations of the kinematic parameters from the nominal
values due to manufacturing and assembly errors of the Stewart platforms
components. The kinematic error model of the Stewart platforms is
155
formulated based on a standard parameters set that is accepted in most
research literature. The analytical formulas are developed and simulated prior
to applying the data collected from sensors. In particular, dial gauges and
wire-sensors are used to calibrate the Stewart platforms. Significant error
reduction was observed after the calibration process. Furthermore, a PSO
search finding measurement for optimum calibration is developed. A
calibration work-piece was used to verify the translation movement of the
Stewart platform and the improvement of the control software. The
calibration results are useful to improve the absolute and relative accuracy
and verify the resolution of the Stewart platform.
The machining process using the cooperative manipulators is
achieved by using NC codes. Data files containing a series of poses can be
used to move the platform in pre-computed trajectories. These trajectories are
generated from a conversion of these NC codes using MatLab programming,
which could generate sufficient trajectory description of a Stewart platform
configuration from an NC file. Three- and five-axis machining processes have
been achieved for the cooperative manipulators based on coordinate mapping.
The NC file is generated using commercial CAM software, namely
MasterCAM in this project, based on 3D models of the work-pieces to be
machined.
8.2 Research Contributions
This work has incorporated two fields of research, namely, the field
of machine tools, and the subject of parallel kinematic manipulation. Detailed
156
studies of the proposed cooperative manipulators have led to some generally
applicable concepts for hybrid manipulation for tasks that require high
stiffness and large workspace. The following points summarize the research
achievements in this thesis.
• Design and implementation of the tool-SP and the table-SP.
• Kinematic analysis and control of the Stewart platforms.
• Implementation of user interface for controlling both Stewart
platforms.
• Integration of the two Stewart platforms to form the cooperative
manipulators to perform machining tasks.
• Conversion of CNC codes from a CAD/CAM package to be used
in the cooperative manipulators as a machine tool.
• Analysis of workspace improvement from the single configuration
to the extended configuration.
• Development of calibration methods to improve motion accuracy.
• Machining case studies with the proposed cooperative
manipulators.
The main contribution of this thesis consists of the experimental
verification of the proof of concept of two cooperative PKMs as a machine
tool, the proposed method for finding optimum motion planning of two
parallel manipulators in performing a single machining task, and the proposed
method of accuracy improvement of the Stewart platforms.
157
8.3 Future Work
New methods and innovative design solutions are needed to solve
problems associated with vibration, stiffness, accuracy, calibration and
temperature compensation. The benefits of the PKMs can be achieved but
they require considerable efforts for their proper application and
generalization. Hence, future works can be focused on the following aspects.
1. With some modifications, the algorithms proposed can be used in
other applications, such as grinding and polishing. More accurate
results can be expected by installing 6-DOF positional or force sensors
on the moving platform and using the additional feedback for the
control algorithms. Currently, the control system used is based on
industrial PID, which controller cards are hard-coded with un-
modifiable control loop for each individual leg. If the controller
feedback includes information from the sensors on the end-effector,
dynamic effects caused by the interaction of the tool and the work-
piece can be compensated.
2. The design of the Stewart platform or the synthesis of a new type of
PKM can be improved. This should include a dynamic model that
could be used in accuracy analysis, dynamic stiffness analysis, and
advanced control. In order to obtain good resolution and accuracy, a
thorough modelling of the issues could be used for error compensation
or redesigning of a new component. Currently, the main problem is
due to the backlash of the joints, platform vibration, limited
calibration devices, and manual fixation of the Stewart platform base.
158
3. The mechanical design can be reviewed to increase the present
working envelope. The results have shown that the table-SP can
increase the workspace of the tool-SP significantly. One option is to
redesign the structural arrangement of the joints of the table-SP. An
approach would be to combine serial and parallel chain mechanisms
to build cooperative manipulators structure with a larger positional
workspace. Other arrangements of the kinematics chains can be
explored to extend the cooperative manipulators depending on the
application requirement.
4. In future, medical applications of Stewart platforms or PKM can be
explored. A few research studies on this area have proven that the
utilization of PKMs in the medical field is more preferred than serial
kinematic manipulators because of their capabilities of handling small
devices with higher accuracy and with smaller installation space. A
PKM can also be implemented as the end-effector of a serial robot.
The development of an effective user interface for conveying the
commands of a surgeon to the machine is a key prerequisite for a
medical robotic assistant.
159
BIBLIOGRAPHY
Andreff N., Renaud P., Martinet P., Pierrot F.. (2004). Vision-based
kinematic calibration of an H4 parallel mechanism: practical
accuracies. Industrial Robot; 31(3): p. 273-83.
Andreff N., Dallej T., Martinet P., (2007). Image-based visual servoing of a
Gough-Stewart parallel manipulator using leg observations.
International Journal of Robotics Research; 26(7): p. 677-87.
Bhattacharya S., Hatwal H., Ghosh A., (1998). Comparison of an exact and
an approximate method of singularity avoidance in platform type
parallel manipulators. Mechanism and Machine Theory; 33(7): p.
965-974.
Bonev I.A., Ryu J., (2001). A new approach to orientation workspace analysis
of 6-DOF parallel manipulators. Mechanism and Machine Theory;
36(1): p. 15-28.
Bonev I.A., Zlatanov D., Gosselin C.M., (2002). Advantages of the modified
Euler angles in the design and control of PKMs. In: Proceeding of the
Parallel Kinematic Machines International Conference. Chemnitz,
Germany; p. 171-188.
Callegari M., Suardi A. (2003) Hybrid Kinematic Machines for Cooperative
Assembly Tasks. In: Proceedings of International Workshop:
Multiagent Robotic Systems: Trends and Industrial Applications,
Robocup. Padova; 7 July 2003.
160
Carbone G., Ceccarelli M. (2005). A serial-parallel robotic architecture for
surgical tasks. Robotica; 23(3): p. 345-54.
Chanal H., Duc E., Ray P., Hascoët J.Y., (2007). A new approach for the
geometrical calibration of parallel kinematics machines tools based on
the machining of a dedicated part. International Journal of Machine
Tools and Manufacture; 47(7-8): p. 1151-63.
Chai K.S., Young K. (2001) Designing a Stewart platform-based Cooperative
System for Large Component Assembly. In: Proceedings of IEEE
Conference on Methods and Models in Automation and Robotics.
Miedzyzdroje; 28-31 August 2001.
Chen S.-H., Fu L.-C. (2006) The forward kinematics of the 6-6 Stewart
platform using extra sensors. In: Proceedings of IEEE Conference on
Systems, Man and Cybernetics. Taipei, 8-11 October 2006.
Chen Y., McInroy J.E., Yi Y. (2003). Optimal, fault-tolerant mappings to
achieve secondary goals without compromising primary performance.
IEEE Transactions on Robotics and Automation; 19(4): p. 680-91.
Cheok K.C., Overholt J.L., Beck R.R. (1993). Exact methods for determining
the kinematics of a Stewart platform using additional displacement
sensors. Journal of Robotic Systems; 10(5): p. 689-707.
Chiacchio P., Pierrot F., Sciavicco L., Siciliano B. (1993). Robust design of
independent joint controllers with experimentation on a high-speed
parallel robot. IEEE Transactions on Industrial Electronics; 40(4): p.
393-403.
161
Clavel R., Helmer P., Niaritsiry T., Rossopoulos S., Verettas I. (2005) High
precision parallel robots for micro-factory applications. In:
Proceedings of 2nd International Colloquium, Collaborative Research
Centre 562. Braunschweig; 10-11 May 2005. p. 285-296.
Codourey A., Burdet E. (1997) A body-oriented method for finding a linear
form of the dynamic Equation of fully parallel robots. In: Proceedings
of IEEE Conference on Robotics and Automation. Albuquesrque,
USA; 20-25 April 1997. p. 1612-1618.
Cortes, J. Simeon T. (2003) Probabilistic motion planning for parallel
mechanisms. In: Proceedings of IEEE Conference on Robotics and
Automation. Taipei, Taiwan; 14-19 September 2003. p. 4354-4359.
Dallej T., Hadj-Abdelkader H., Andreff N., Martinet P. (2006a) Kinematic
calibration of a Gough-Stewart platform using an omnidirectional
camera. In: Proceedings of IEEE/RSJ Conference on Intelligent
Robots and Systems. Beijing, China; 9-15 October 2006. p. 4666-
4671.
Dallej T., Andreff N., Mezouar Y., Martinet P. (2006b) 3D pose visual
servoing relieves parallel robot control from joint sensing. In:
Proceedings of IEEE/RSJ Conference on Intelligent Robots and
Systems. Beijing, China; 9-15 October 2006. p. 4291-4296.
Daney D. (2002) Optimal measurement configurations for Gough platform
calibration. In: Proceedings of IEEE Conference on Robotics and
Automation. Washington, DC, USA; 11-15 May 2002. p. 147-152.
162
Daney D., Madeline B., Papegay Y. (2005). Choosing measurement poses for
robot calibration with the local convergence method and Tabu search.
International Journal of Robotics Research; 24(6): p. 501-508.
Daney D., Andreff N., Chabert G., Papegay Y. (2006). Interval method for
calibration of parallel robots: Vision-based experiments. Mechanism
and Machine Theory; 41(8): p. 929-944.
Dasgupta B., Choudhury P. (1999). A general strategy based on the Newton-
Euler approach for the dynamic formulation of parallel manipulators.
Mechanism and Machine Theory; 34(6): p. 801-824.
Dasgupta B., Mruthyunjaya T.S. (2000). Stewart platform manipulator: A
review. Mechanism and Machine Theory; 35(1): p. 15-40.
Dietmaier, P. (1998) The Stewart-Gough Platform of General Geometry can
have 40 real postures. In: Proceedings of 6th International Symposium
on Advances in Robot Kinematics: Analysis and Control. Strobl,
Austria; 29 June - 4 July 1998. p. 7-16.
Du Plessis L.J., Snyman J.A. (2001). A numerical method for the
determination of dextrous workspaces of Gough-Stewart platforms.
International Journal for Numerical Methods in Engineering; 52(4): p.
345-69.
El-Khasawnath BS, Ferreira PM, (1999). Computation of stiffness and
stiffness bounds for parallel link manipulators. International Journal
of Machine Tools and Manufacturing; 39(2): p. 321–342.
Fassi I., Wiens G.J. (2000). Multiaxis machining: PKMs and traditional
machining centres. Journal of Manufacturing Processes; 2(1): p. 1-14.
163
Fichter E.F. (1986). A Stewart-Platform based manipulator: general theory
and practical construction. International Journal of Robotic Research;
5(2): p. 157-182.
Gallardo, J., Rico J.M., Frisoli A., Checcacci D., Bergamasco M. (2003).
Dynamics of parallel manipulators by means of screw theory.
Mechanism and Machine Theory; 38(11): p. 1113-1131.
Geng, Z., Haynes L.S., Lee J.D., Carroll R.L. (1992). On the dynamic model
and kinematic analysis of a class of Stewart platforms. Robotics and
Autonomous Systems; 9(4): p. 237-54.
Geng, Z., Haynes L.S.(1994) An Effective Kinematic Calibration Method for
Stewart platforms. In: Proceedings of International Symposium on
Robotics and Manufacturing. Hawai; 15-17 August 1994. p. 87-92.
Geldart M., Webb P., Larsson H., Backstrom M., Gindy N., Rask K., (2003).
A direct comparison of the machining performance of a variax 5 axis
parallel kinetic machining centre with conventional 3 and 5 axis
machine tools. International Journal of Machine Tools and
Manufacture; 43(11): p. 1107-1116.
Golub G. H., Van Loan, C. F. Matrix Computations, Baltimore: The Johns
Hopkins University Press, 1983.
Gosselin, C., Angeles J. (1990a). Singularity analysis of closed-loop
kinematic chains. IEEE Transactions on Robotics and Automation;
6(3): p. 281-290.
Gosselin, C. (1990b). Determination of the Workspace of 6-DOF Parallel
Manipulators. Journal of Mechanical Design; 112(3): p. 331-336.
164
Gosselin, C., Angeles J. (1990c). Kinematic inversion of parallel
manipulators in the presence of incompletely specified tasks. Journal
of Mechanical Design; 112(4): p. 494-500.
Gough, V.E. Whitehall S.G. (1962) Universal tire test machine. In:
Proceedings of the 9th FISITA World Automotive Congress. London;
1962, p. 117-137.
Han C., Kim J., Kim J., Park F.C., (2002). Kinematic sensitivity analysis of
the 3-UPU parallel mechanism. Mechanism and Machine Theory;
37(8): p. 787-798.
Harib K.H, Srinivasan K. (2003). Kinematic and dynamic analysis of Stewart
platform-based machine tool structures. Robotica; 21(5): p. 541-554.
Harib K.H., Sharif Ullah A.M.M., Hammami A. (2007). A hexapod-based
machine tool with hybrid structure: Kinematic analysis and trajectory
planning. International Journal of Machine Tools and Manufacture;
47(9): p. 1426-32.
Harris, D.M.J. (1995) Parallel-linkage robot coordinate transformation
horugh screw theory. In: Proceedings of 9th World Congress on the
Theory of Machines and Mechanism. Milan; 30 August-2 Sept 1995.
p. 1565-1568.
Huang C.-I., Chang C.-F.. Yu M.-Y., Fu L.-C. (2004) Sliding-mode tracking
control of the Stewart platform. In: Proceedings of 5th Asian Control
Conference. Melbourne, Australia; 20-23 July 2004. p. 562-569.
Huang C.-I., Fu L.-C. (2004) Adaptive backstepping tracking control of the
Stewart platform. In: Proceedings of 43rd IEEE Conference on
165
Decision and Control. Nassau, Bahamas; 14-17 December 2004. p.
5228-5223.
Huang T., Wang J., Gosselin C.M., Whitehouse D.J. (1999). Determination
of closed form solution to the 2-D orientation workspace of Gough-
Stewart parallel manipulators. IEEE Transactions on Robotics and
Automation; 15(6): p. 1121-5.
Huang T., Whitehouse D.J. (2000). A Simple yet Effective Approach for
Error Compensation of a Tripod-Based Parallel Kinematic Machine.
CIRP Annals - Manufacturing Technology; 49(1): p. 285-288.
Huang T., Chetwynd D.G., Whitehouse D.J., Wang J. (2005). A general and
novel approach for parameter identification of 6-DOF parallel
kinematic machines. Mechanism and Machine Theory; 40(2): p. 219-
239.
Hunt K.H., Kinematic Geometry of Mechanism, Oxford: Clarendon Press,
1978.
Kennedy J., Eberhart R. (1995) Particle swarm optimization. In: Proceedings
of IEEE Conference on Neural Networks. Perth, Australia; 27
November-1 December 1995. p. 1942-1948.
Kim S., S. Lee. (1992) Cartesian space dynamic model of serial-parallel
manipulator systems and their dynamic performance evaluation. In
Proceedings of 31st IEEE Conference on Decision and Control.
Tucson, AZ, USA; 16-18 December 1992. p. 327-328.
Kock S., Schumacher W. (2000) A mixed elastic and rigid-body dynamic
model of an actuation redundant parallel robot with high-reduction
166
gears. In: Proceedings of IEEE Conference on Robotics and
Automation. San Francisco, CA, USA; 24-28 April 2000. p. 1918-
1923
Lallemand J.P., Goudali A., Zeghloul S. (1997). The 6-DOF 2-Delta parallel
robot. Robotica; 15: p. 407-16.
Li C.-G., Ding H.-S., Wu P.-D. (2003) Application of MRAC to a 6-dof
parallel machine tool. In: Proceedings of IEEE Conference on
Machine Learning and Cybernetics. Xi’an, China; 2-5 November
2003. p. 2164-2167.
Liu G.F., Wu Y.L., Wu X.Z., Kuen Y.Y., Li Z.X. (2001) Analysis and control
of redundant parallel manipulators. In: Proceeding of IEEE
Conference on Robotics and Automation. Seoul, Korea; 21-26 May
2001. p. 3748-3754.
Liu, K., Lewis F., Lebret G., Taylor D. (1993). The singularities and
dynamics of a Stewart platform manipulator. Journal of Intelligent and
Robotic Systems; 8(3): p. 287-308.
Ma O., Angeles J. (1991a) Architecture singularities of platform manipulators.
In: Proceedings of IEEE Conference on Robotics and Automation.
Sacramento, CA, USA; 9-11 April 1991 p. 1542-1547.
Ma O., Angeles J. (1991b) Optimum Architecture Design of Platform
Manipulators. In: Proceedings of the 5th International Conference on
Advanced Robotics. 19-22 June 1991. p. 1130-1135.
Marquet F., Company O., Krut S., Pierrot F. (2002) Enhancing parallel robots
accuracy with redundant sensors. In: Proceedings of IEEE Conference
167
on Robotics and Automation. Washington, DC, USA; 11-15 May
2002. p. 4114-4119.
Masory, O., Wang J., (1995). Workspace evaluation of Stewart platforms.
Journal of Advanced Robotics; 9(4): p. 443-61.
Masory O., Wang J., Zhuang H. (1997) Kinematic modelling and calibration
of a Stewart platform. Journal of Advanced Robotics; 11(5): p. 519-
541.
Meng J., Li Z. (2005) A general approach for accuracy analysis of parallel
manipulators with joint clearance. In: Proceedings of IEEE/RSJ
Conference on Intelligent Robots and Systems. Canada; 2-6 August
2005 p. 2468-2473.
Merlet J.P. (1992) On the infinitesimal motion of a parallel manipulator in
singular configurations. In: Proceedings of IEEE Conference on
Robotics and Automation. Nice, France; 12-14 May 1992. p. 320-325.
Merlet J.P. (1994) Trajectory verification of parallel manipulators in the
workspace. In: Proceedings of IEEE Conference on Robotics and
Automation. San Diego, CA, USA; 8-13 May 1994. p. 2166-2171.
Merlet J.P. (1999) Parallel Robots: Open Problems. In: Proceedings of 9th
International Symposium of Robotics Research. Snowbird, Utah,
USA; 9-12 October 1999.
Merlet J.P., Perng M-W., Daney. D. (2000) Optimal Trajectory Planning of a
5-axis Machine-Tool based on a 6-axis parallel Manipulator. In:
Porceedings of 7th International Symposium on Advances in Robot
Kinematics. Piran, Slovenia; June 2000. p. 315-322.
168
Merlet J.P. (2002) Still a long way to go on the road for parallel mechanisms.
In: ASME 2002 International Design Engineering Technical
Conferences and Computers and Information in Engineering
Conference. Montreal, Quebec, Canada; 29 September-2 October
2002.
Merlet J.P., Parallel Robot. Solid Mechanics and Its Application., Dordrecht:
Springer, 2006.
Merlet J.P. (2007). A formal-numerical approach for robust in-workspace
singularity detection. IEEE Transactions on Robotics; 23(3): p. 393-
402.
Nahvi A., Hollerbach J.M. (1996) The noise amplification index for optimal
pose selection in robot calibration. In: Proceedings of IEEE
Conference on Robotics and Automation. Minneapolis, MN, USA;
22-28 April 1996. p. 647-654.
Neugebauer R., Wieland F., Ihlenfeldt S. (2000). Comparison of Parallel
Structure Concepts for Five-Axis Machining. Journal of
Manufacturing Processes; 2(1): p. 59-66.
Neumann K.E., (2006). Exechon Concept. In: The 5th Chemnitz Parallel
Kinematics Seminar. Allemagne; p.787-802.
Nguyen C.C., Pooran F.J. (1989). Dynamic analysis of a 6 DOF CKCM robot
end-effector for dual-arm telerobot systems. Robotics and
Autonomous Systems; 5(4): p. 377-94.
Nguyen C.C., Zhou Z.-L., Antrazi S.S., Campbell C.E., Jr. (1991) Efficient
computation of forward kinematics and Jacobian matrix of a Stewart
169
platform-based manipulator. In: Proceedings of IEEE SoutheastCon
1991. Williamsburg, VA, USA; 7-10 April 1991. p. 869-874.
Niaritsiry F.-T., Fazenda N., Clavel R. (2004) Study of the source of
inaccuracy of a 3 DOF flexure hinge-based parallel manipulator. In:
Proceedings of IEEE Conference on Robotics and Automation. New
Orleans; 26 April-1 May 2004. p. 4091-4096.
Parenti-Castelli V., Di Gregorio R. (1995). Determination of the actual
configuration of the general Stewart platform using only one
additional displacement sensor. Journal of Mechanical Design;
121(1):p. 21-25.
Parenti-Castelli V., Di Gregorio R. (2000). A New Algorithm Based on Two
Extra-Sensors for Real-Time Computation of the Actual
Configuration of the Generalized Stewart-Gough Manipulator.
Journal of Mechanical Design; 122(3): p. 294-298.
Parikh P.J., Lam S.S.Y. (2005). A hybrid strategy to solve the forward
kinematics problem in parallel manipulators. IEEE Transactions on
Robotics; 21(1): p. 18-25.
Patel A.J., Ehmann K.F. (1997). Volumetric error analysis of a Stewart
platform-based machine tool. CIRP Annals - Manufacturing
Technology; 46(1): p. 287-290.
Patel A. J., Ehmann K. F., (2000). Calibration of a hexapod machine tool
using a redundant leg. International Journal of Machine Tools and
Manufacture; 40(4): p. 489-512.
170
Pierrot F. (2002) Parallel mechanism and redundancy. In: Proceedings of 1st
International Colloquium, Collaborative Research Centre 562.
Braunschweig; 29-30 May 2002. p. 261-277.
Pritschow G., Eppler C., Garber T. (2002). Influence of the dynamic stiffness
on the accuracy of PKM. In: Proceedings of the 3rd Chemnitzer
Parallelkinematik Seminar. Chemnitz, Germany; May 2002. p. 313-
333.
Pugazhenthi S., Nagarajan T., Singaperumal M. (2002). Optimal trajectory
planning for a hexapod machine tool during contour machining.
Proceedings of the Institution of Mechanical Engineer, Part C: Journal
of Mechanical Engineering Science; 216(12):p. 1247-1257.
Reboulet C., Berthomieu T. (1991) Dynamic models of a six degree of
freedom parallel manipulators. In: Proceedings of the 5th International
Conference on Advanced Robotics. 19-22 June 1991. p. 1153-1157.
Refaat S., Herve J.M., Nahavandi S., Trinh H., (2007). Two-mode
overconstrained three-dofs rotational-translational linear-motor-
based parallel-kinematics mechanism for machine tool applications.
Robotica; 25(4): p. 461-466.
Renaud P., Andreff N., Lavest J.-M., Dhome M. (2006). Simplifying the
kinematic calibration of parallel mechanisms using vision-based
metrology. IEEE Transactions on Robotics; 22(1): p. 12-22.
Ropponen T., Arai T. (1995) Accuracy analysis of a modified Stewart
platform manipulator. In: Proceedings of IEEE International
171
Conference on Robotics and Automation. Nagoya, Aichi, Japan; 21-
27 May 1995. p. 521-525.
Stewart D. (1965). A platform with six degrees of freedom. In: Proceedings
of the Institution of Mechanical Engineers; 180(15): p. 371-386.
St-Onge B.M., Gosselin C.M. (2000). Singularity analysis and representation
of the general Gough-Stewart platform. International Journal of
Robotics Research; 19(3): p. 271-88.
Stoughton R. S., Arai T. (1993). A Modified Stewart Platform Manipulator
with Improved Dexterity. IEEE Transactions on Robotics and
Automation; 9(2):p. 166-73.
Schwaar M., Jaehnert T., Ihlenfeldt S., (2002). Mechatronic design,
experimental property analysis and machining strategies for a 5-strut-
PKM. In: The 3rd Chemnitz Parallel Kinematics Seminar:
Development Methods and Application Experience of Parallel
Kinematics. Verlag Wissenschaftliche Scripten, Zwickau; p. 671-681.
Takeda Y., Shen G., Funabashi H. (2004). A DBB-based kinematic
calibration method for in-parallel actuated mechanisms using a
Fourier series. Journal of Mechanical Design; 126(5): p. 856-65.
Tanaka W., Arai T., Inoue K., Takubo T., Park C.S. (2006) Calibration
method for parallel mechanism using micro grid pattern. In:
Proceeding of IEEE Conference on Robotics and Automation.
Orlando, FL, USA; 15-19 May 2006. p. 763-768.
Tanev T.K. (2000). Kinematics of a hybrid (parallel-serial) robot manipulator.
Mechanism and Machine Theory; 35(9): p. 1183-1196.
172
Tarokh M. (2007) Real time forward kinematics solutions for general Stewart
platforms. In: Proceeding of IEEE Conference on Robotics and
Automation. Rome, Italy; 10-14 April 2007. p. 901-906.
Terrier M., Giménez M., Hascoët J.Y., (2005). Verne - a five-axis parallel
kinematics milling machine. Proceedings of the Institution of
Mechanical Engineers, Part B: Journal of Engineering Manufacture;
219(3): p. 327-336.
Tischler C.R., Samuel A.E. (1998) Predicting the Slop of In-Series/Parallel
Manipulators caused by Joint Clearances. In: Proceedings of 6th
International Symposium on Advances in Robot Kinematics: Analysis
and Control. Strobl, Austria; 29 June - 4 July 1998. p. 227-236.
Tlusty J., Ziegert J.C., Ridgeway S. (1999). Fundamental comparison of the
use of serial and parallel kinematics for machines tools. CIRP Annals
- Manufacturing Technology; 48(1): p. 351-356.
Tlusty J., Ziegert J.C., Ridgeway S. (2000). A Comparison of Stiffness
Characteristics of Serial and Parallel Machine Tools. Journal of
Manufacturing Processes; 2(1): p. 67-76.
Tsai L.-W. (2000). Solving the Inverse Dynamics of a Stewart-Gough
Manipulator by the Principle of Virtual Work. Journal of Mechanical
Design; 122(1): p. 3-9.
Tsai L.-W., Joshi S. (2002). Kinematic Analysis of 3-DOF Position
Mechanisms for Use in Hybrid Kinematic Machines. Journal of
Mechanical Design; 124(2): p. 245-253.
173
Voglewede P.A., Ebert-Uphoff I., (2005). Overarching framework for
measuring closeness to singularities of parallel manipulators. IEEE
Transactions on Robotics; 21(6): p. 1037-1045.
Wampler C.W., Hollerbach J.M., Arai T. (1995). An implicit loop method for
kinematic calibration and its application to closed-chain mechanisms.
IEEE Transactions on Robotics and Automation; 11(5): p. 710-724.
Wang H., Xue C., Gruver W.A. (1995) Neural network control of a parallel
robot. In: Proceedings of IEEE Conference on Systems, Man and
Cybernetics. Vancouver, BC, Canada; 22-25 October 1995. p. 2934-
2938.
Wang J., Masory O. (1995) On the accuracy of Stewart Platform – Part I the
effect of manufacturing tolerances. In: Proceedings of IEEE
International Conference on Robotics and Automation. Nagoya, Aichi,
Japan; 2-6 May 1995. p. 114 – 120.
Wang J., Gosselin C.M. (1998). A New Approach for the Dynamic Analysis
of Parallel Manipulators. Multibody System Dynamics; 2(3): p. 317-
334.
Wang J., Gosselin C.M. (2004). Kinematic analysis and design of
kinematically redundant parallel mechanisms. Journal of Mechanical
Design; 126(1): p. 109-18.
Wang, S.-M., Ehmann K.F. (2002). Error Model and Accuracy Analysis of a
Six-DOF Stewart platform. Journal of Manufacturing Science and
Engineering; 124(2): p. 286-295.
174
Wang, Y. (2006) An incremental method for forward kinematics of parallel
manipulators. In: Proceedings of IEEE Conference on Robotics,
Automation and Mechatronics. Bangkok, Thailand; 1-3 June 2006. p.
1-5.
Wang, Z., Liu W., Lei Y. (2001). A study on workspace, boundary workspace
analysis and work-piece positioning for parallel machine tools.
Mechanism and Machine Theory; 36(5): p. 605-622.
Weck M., Staimer D., (2002). Parallel Kinematic Machine Tools – Current
State and Future Potentials. CIRP Annals - Manufacturing
Technology; 51(2): p. 671-683.
Wohlhart K. (1999). Degrees of shakiness. Mechanism and Machine Theory;
34(7): p. 1103-1126.
Yurt S.N., Őzkol I., Kaya M.O., Haciyev C., (2002). Optimization of the PD
coefficient in a flight simulator control via genetic algorithms.
Aircraft Engineering and Aerospace Technology; 74(2): p. 147-153.
Zhang D., Wang L., Lang S.Y.T. (2005). Parallel Kinematic Machines:
Design, Analysis and Simulation in an Integrated Virtual
Environment. Journal of Mechanical Design; 127(4): p. 580-588.
Zheng G., Haynes L.S. (1993). Dynamic control of a parallel link manipulator
using a CMAC neural network. Computers and Electrical Engineering;
19(4): p. 265-76.
Zheng X.Z., Bin H.Z., Luo Y.G. (2004). Kinematic analysis of a hybrid serial-
parallel manipulator. International Journal of Advanced
Manufacturing Technology; 23(11-12): p. 925-30.
175
Zhuang H., Roth Z.S. (1993). Method for kinematic calibration of Stewart
platforms. Journal of Robotic Systems; 10(3): p. 391-405.
Zhuang H., Yan J., and Masory O. (1998). Calibration of Stewart Platform
and other parallel manipulators by minimizing inverse kinematic
residuals. Journal of Robotic Systems; 15(7): p. 396-406.
Zlatanov D., Bonev I.A., Gosselin C.M., (2002). Constraints singularities of
parallel mechanisms. In: Proceedings of IEEE Conference on
Robotics and Automation. p. 496-502.
Zuo A., Wu Q.M.J., Gruver W.A. (2002) Stereo vision guided control of a
Stewart platform. In: Proceedings of IEEE International Symposium
on Intelligent Control. Vancouver, BC, Canada; 30 October 2002. p.
125-130.
176
PUBLICATIONS FROM THIS RESEARCH
1. Saputra. V.S., Ong S.K., Nee A.Y.C., Development and Trajectory
Planning of a Hybrid Serial-Parallel Manipulator, Asian International
Journal of Science and Technology in Production and Manufacturing
Engineering, 2009, 2(1): p. 99-115.
2. Saputra. V.S., Ong S.K., Nee A.Y.C., A PSO algorithm for mapping
the workspace boundary of parallel manipulators, In Proceedings of
International Conference on Robotics and Automation, Anchorage,
Alaska, US; 3-7 May 2010, p. 4691-4696.
3. Saputra. V.S., Ong S.K., Nee A.Y.C., A Swarm Optimization
Approach for Solving Workspace Determination of Parallel
Manipulators, Robotica, July 2012, under review.
4. Saputra. V.S., Ong S.K., Nee A.Y.C., Optimum Calibration of a
Parallel Kinematic Manipulator (PKM) using Digital Indicators,
Journal of Intelligent and Robotic Systems, October 2012, under
review.
177
You might also like
- TS5040 - LabVIEW Robotics Starter Kit (DaNI) Lab ExercisesfNo ratings yetTS5040 - LabVIEW Robotics Starter Kit (DaNI) Lab Exercisesf25 pages
- Robotic Engineering 2: ROS-Turtlebot Motion Control and NavigationNo ratings yetRobotic Engineering 2: ROS-Turtlebot Motion Control and Navigation34 pages
- Robot Technology: Ch.7 Trajectory Planning of RobotsNo ratings yetRobot Technology: Ch.7 Trajectory Planning of Robots53 pages
- Path Planning Techniques For Unmanned Aerial Vehicles A Review and SolutionsNo ratings yetPath Planning Techniques For Unmanned Aerial Vehicles A Review and Solutions30 pages
- FUNDAMENTALS OF ROBOTICS SYLLABUS 01082023aNo ratings yetFUNDAMENTALS OF ROBOTICS SYLLABUS 01082023a7 pages
- @vision-Based Control of Supernumerary Robotic Limbs For Grasping TasksNo ratings yet@vision-Based Control of Supernumerary Robotic Limbs For Grasping Tasks6 pages
- A Survey of FPGA-Based Robotic ComputingNo ratings yetA Survey of FPGA-Based Robotic Computing27 pages
- Automated Offline Programming For Robotic Welding SystemNo ratings yetAutomated Offline Programming For Robotic Welding System9 pages
- Development of A Smart Metrology Machine-1 - 240611 - 173401No ratings yetDevelopment of A Smart Metrology Machine-1 - 240611 - 17340162 pages
- Path Planning and Control of Mobile RoverNo ratings yetPath Planning and Control of Mobile Rover21 pages
- The Off-Grid Blueprint: DIY Survival Projects for Modern Self-Reliance: DIY Survival Projects for Modern Self-Reliance: DIY Survival Projects for Modern Self-RelianceFrom EverandThe Off-Grid Blueprint: DIY Survival Projects for Modern Self-Reliance: DIY Survival Projects for Modern Self-Reliance: DIY Survival Projects for Modern Self-RelianceNo ratings yet
- Approaching Socially Aware Robot Navigation Framework Where and How To ApproachNo ratings yetApproaching Socially Aware Robot Navigation Framework Where and How To Approach15 pages
- Assignment 3: Path Planning and Object RecognitionNo ratings yetAssignment 3: Path Planning and Object Recognition3 pages
- A Survey of Technical Trends of ADAS and Autonomous DrivingNo ratings yetA Survey of Technical Trends of ADAS and Autonomous Driving4 pages
- Integrated Task Assignment and Path Planning For Capacitated Multi-Agent Pickup and DeliveryNo ratings yetIntegrated Task Assignment and Path Planning For Capacitated Multi-Agent Pickup and Delivery8 pages
- Aircraft Trajectory Planning With Collision Avoidance Using Mixed Integer Linear Programming, ACC.2002.1023918No ratings yetAircraft Trajectory Planning With Collision Avoidance Using Mixed Integer Linear Programming, ACC.2002.10239186 pages
- No More Heavy Lifting: Robotic Solutions To The Container Unloading ProblemNo ratings yetNo More Heavy Lifting: Robotic Solutions To The Container Unloading Problem11 pages
- The Multiple Traveling Salesman Problem Tolga Bektas PDFNo ratings yetThe Multiple Traveling Salesman Problem Tolga Bektas PDF11 pages
- (Xiaolong2017) Robotwarehouse Navigation MethodNo ratings yet(Xiaolong2017) Robotwarehouse Navigation Method4 pages
- Design and Implementation of Stewart Platform Robot For 1No ratings yetDesign and Implementation of Stewart Platform Robot For 1161 pages
- Hybrid Orientation and Position Collaborative Motion Generation Scheme For A Multiple Mobile Redundant Manipulator System Synthesized by A Recurrent Neural NetworkNo ratings yetHybrid Orientation and Position Collaborative Motion Generation Scheme For A Multiple Mobile Redundant Manipulator System Synthesized by A Recurrent Neural Network13 pages
- Reconfigurable Kinematics of General Stewart Platform and SimulatNo ratings yetReconfigurable Kinematics of General Stewart Platform and Simulat115 pages
- Machine Tool Design N K MEHTA 3rd EditioNo ratings yetMachine Tool Design N K MEHTA 3rd Editio478 pages
- The Future of Learning: Revolutionizing Education Through Generative AI: AI Books, #11From EverandThe Future of Learning: Revolutionizing Education Through Generative AI: AI Books, #11No ratings yet
- A Systematic Development Method For Cyber-Physical Machine ToolsNo ratings yetA Systematic Development Method For Cyber-Physical Machine Tools13 pages
- Development of Machining-Based Technique To Generate Geometrical Features On A SurfaceNo ratings yetDevelopment of Machining-Based Technique To Generate Geometrical Features On A Surface29 pages
- Design and Fabrication of Stewart PlatformNo ratings yetDesign and Fabrication of Stewart Platform27 pages
- Process Optimization at Demand Machine Tools and Gauges, FaridabadNo ratings yetProcess Optimization at Demand Machine Tools and Gauges, Faridabad33 pages
- Design and Fabrication of A CNC Precision Router: Senior Design Project ReportNo ratings yetDesign and Fabrication of A CNC Precision Router: Senior Design Project Report57 pages
- Time-dependent Behaviour and Design of Composite Steel-concrete StructuresFrom EverandTime-dependent Behaviour and Design of Composite Steel-concrete StructuresNo ratings yet
- Structural Design of 3-Axis CNC Machine Tool For Wood CarvingNo ratings yetStructural Design of 3-Axis CNC Machine Tool For Wood Carving100 pages
- Aquaponics Construct and Operate: Instructions and Everything You Need to KnowFrom EverandAquaponics Construct and Operate: Instructions and Everything You Need to KnowNo ratings yet
- Securing ChatGPT: Best Practices for Protecting Sensitive Data in AI Language ModelsFrom EverandSecuring ChatGPT: Best Practices for Protecting Sensitive Data in AI Language ModelsNo ratings yet
- Aquaponics Design Plans, Construction, Operation, and Income: Organic FoodFrom EverandAquaponics Design Plans, Construction, Operation, and Income: Organic FoodNo ratings yet
- Blog Smarter, Not Harder: SEO, Blogging, and AI Strategies to Skyrocket Your TrafficFrom EverandBlog Smarter, Not Harder: SEO, Blogging, and AI Strategies to Skyrocket Your TrafficNo ratings yet
- Aquaponics How to do Everything from Backyard to Profitable Business: from BACKYARD to PROFITABLE BUSINESSFrom EverandAquaponics How to do Everything from Backyard to Profitable Business: from BACKYARD to PROFITABLE BUSINESSNo ratings yet
- Aquaponic Design Plans Everything You Needs to Know: Everything You Need to Know from Backyard to Profitable BusinessFrom EverandAquaponic Design Plans Everything You Needs to Know: Everything You Need to Know from Backyard to Profitable BusinessNo ratings yet
