You are on page 1of 14

Int J Adv Manuf Technol (2014) 75:933–946

DOI 10.1007/s00170-014-6196-6

ORIGINAL ARTICLE

Real-time thermal error compensation method for robotic


visual inspection system
Shibin Yin & Yin Guo & Yongjie Ren & Jigui Zhu &
Shourui Yang & Shenghua Ye

Received: 13 February 2014 / Accepted: 21 July 2014 / Published online: 8 August 2014
# Springer-Verlag London 2014

Abstract The thermally induced error is a critical element in 1 Introduction


the total errors for the robotic visual inspection system during
its long-term operation. This paper investigates the thermal With the rapid development of robot manufacture industry
behavior of the robotic visual inspection system and proposes and the continuous decrease of its production cost in recent
a real-time thermal error compensation method. Based on the years, the industrial robot has become an economical and
fixed-point constraint, an error calibration model is derived by flexible orienting device for industrial inspection. Compact
associating the thermally induced parameter errors with the and portable visual sensors have been increasingly integrated
deviations in the measured coordinates of the fixed reference with the industrial robot to construct the robotic visual inspec-
point during the robot warm up and cool down. A joint-by- tion system [1–3]. The robotic visual inspection system is not
joint test is performed to investigate the link parameters that only economically promising, but also combines the industrial
change significantly with respect to temperature variations. robot’s high-flexibility and visual sensor’s high-throughput
Standard spheres are adopted as the calibration targets, and and high-accuracy. It has been applied extensively for geom-
optimum projected angles for sphere center measurement by etry reverse engineering (GRE) and flexible automated online
visual sensor are obtained based on theoretical analysis and inspection of parts and components in sectors such as auto-
experimental data. Contrary to most other works, the method motive assembly, shipbuilding, and aircraft manufacturing [4,
needs no exact knowledge of the temperature sensors and is 5]. Nevertheless, industrial robots are normally designed for
well suited for online dynamic thermal error compensation for repeated work such as pick and place, spot welding, and so on.
the robotic visual inspection system. Verification experiment It has high repeatability but low accuracy. To some extent, the
is carried out on a car-body assembly line, and results show accuracy of the robotic measurement system has been limited
that the max/mean residual error for the tested points has been by the robot positioning accuracy. With the normal model
reduced from 0.441/0.195 to 0.136/0.078 mm with the signif- specified by the manufacturer, the positioning accuracy of
icant parameters calibrated. the industrial robot is generally worse in comparison with
the required performance, especially when high accuracy
(such as 0.1 mm) is desired. Therefore, there’s an urgent
Keywords Industrial robot . Thermal error . Standard sphere . demand for improving and maintaining the robot’s accuracy
Real-time compensation in the varying manufacturing environments [6].
Robot calibration is a cost-effective way to improve robot
accuracy, which is a procedure performed on each individual
robot. To a large extent, the discrepancy between the robot’s
S. Yin : Y. Ren (*) : J. Zhu : S. Yang : S. Ye
ideal and real behavior is due to the joint angle offsets,
State Key Laboratory of Precision Measuring Technology and geometric link length inaccuracy, and non-geometric factors
Instruments, Tianjin University, Tianjin 300072, China like flexibility, backlash, thermal expansion, and so on. Roth
e-mail: yongjieren@163.com has defined three levels of robot calibration to classify most of
the current approaches for robot calibration [7]. The first level
Y. Guo
Department of Precision Instrument, Tsinghua University, calibration is defined as “joint level”, which is to determine
Beijing 100084, China the relationship between the readout of joint transducer and
934 Int J Adv Manuf Technol (2014) 75:933–946

the actual joint displacement. The second level calibration is calibration and later compensation methods. Manufacturers
defined as entire kinematic calibration, which is to determine generally try to minimize the thermal effect by using low
the basic kinematic geometry of the robot and the correct thermal-expansion materials in the robot design or by
joint-angle relationships [8–10]. The third level calibration is implementing an empirical error correction model based on
defined as “non-geometric” calibration, which is to determine the outputs of several temperature sensors placing inside the
the positioning error due to effects such as link compliance, robot arm. And unlike robot geometrical and compliance
friction, and clearance [11, 12]. Several researches have fo- errors compensation, the error due to thermal deformation
cused on the issue of kinematic calibration, different kinemat- has not been extensively investigated in literatures [21, 22].
ic models and various algorithms for identification and com- Heisel [23] reported the sequential investigation of the thermal
pensation have been developed as summarized by Hollerbach behavior of different types of industrial robots with a stan-
[13] and Mooring [14]. Various measurement techniques such dardized experimental setup and suggested the possibility for
as coordinate measuring machine (CMM), laser tracker, and improving the robot design and the strategy for thermal error
economical ball-bar have been employed for calibration tasks, compensation. Poonyapak [24] proposed a predictive model
which have made the calibration process more convenient for temperature-induced deformation of an industrial robot by
[15–17]. Non-geometrical errors that vary predictably with using thermal imaging camera and incorporated a simple
position such as gravity, joint compliance, and gear transmis- compensation algorithm in the robot controller to reduce the
sion errors (third level) have also received great attention [18]. warm-up time. Gong [25] derived a comprehensive error
Nubiola [19, 20] proposed an absolute calibration method for model which combines geometric errors, position-dependent
an ABB IRB1600 robot using a laser tracker, which takes into compliance errors, and time-variant thermal errors. The ther-
account all possible geometric errors and compliance error. mally induced parameter errors are calibrated at different
Generally, for serial manipulators, the mere geometrical error temperatures, and empirical thermal error models are
is the major font of total errors and the geometrical parameters established to correlate robot thermal errors with the corre-
calibration plus the compliance compensation have supplied sponding temperature fields. However, thermistor sensors are
very good result, improving the accuracy almost up to the attached on the robot arm to construct the temperature field,
robot repeatability. and complex modeling methods are adopted to establish the
When the robotic measurement system has been adopted empirical error correction model. The installation and cabling
for online inspection and run continuously at a high speed, of the sensors are tedious, and the system maintenance is
such as on the car assembly and sub-assembly line, system arduous, so this method is difficult to be implemented on the
long-term stability is required. However, when the robot has robots working on the manufacturing floor continuously.
run at a high speed over a very long period, the internal heat In this paper, a real-time dynamic thermal error compensa-
sources such as motors, bearings, and ambient heat sources tion method was developed for the robotic visual inspection
will cause the robot arm temperature change, which will cause system. The method is designed to be applied on production
not only link expansion but also structure distortions. And the line and correct the thermal error during the robotic system
industrial robots are generally constructed by thin, beam-like operation. Based on the fixed-point constraint, an error cali-
links, the time constants for the heating and cooling are bration model is derived for associating the thermally induced
therefore small, it is not realistic to achieve thermally stable parameter errors with the differences between the measured
operation in the industrial field. Moreover, the aforementioned coordinates of a fixed reference point during the industrial
robot calibration procedures are generally carried out in a robot warm-up/cool-down stage. A joint-by-joint test clearly
temperature-controlled environment, so the correction models displays the thermal behavior of the robot and justifies our
are valid only for those temperature conditions. Therefore, choice of the link parameters that change significantly with
robot thermal errors must be estimated and compensated respect to temperature variations. The significant parameters
online during its operation in order to maintain the robot’s are identified by singular value decomposition (SVD) without
accuracy without affecting the production efficiency. initial estimates. The best implementation of this method is
The thermal errors of the industrial robot originate from the proposed by using standard spheres as the calibration targets,
nonlinear and time-varying thermal deformations, which are and the optimum projected angle for sphere center measure-
caused by the non-uniform temperature variations in the robot ment with line-structured laser sensor has been obtained from
structure and the random changes in the ambient temperatures. experimental data. Contrary to nearly all other works, no exact
The temperature variations are related to the heat source knowledge of the temperature sensors is necessary for the
location, heat source intensity, thermal expansion coefficient, thermal error compensation method presented here and no
and the robotic system configuration. Difficulty in mathemat- complicated empirical error model is established to estimate
ically characterizing the influence of temperature variations on the robot thermal error during it works. Therefore, it meets the
the robot accuracy in a general way for any robot has tradi- demand of real-time thermal error compensation and is well
tionally prevented the development of thermal error suited for application on the robotic visual inspection system.
Int J Adv Manuf Technol (2014) 75:933–946 935

The rest of the paper is organized as follows. Section 2 where esT denotes the transform relationship between the
introduces the principle of the robotic visual inspection system laser sensor frame and the robot end-effector frame, be T is the
and thermal behavior of the system. In section 3, the robot transformation between robot end-effector frame and the robot
kinematic model and error model will be briefly reviewed first base frame, which could be obtained from robot forward
and then the dynamic thermal error calibration algorithm will kinematic model, and wb T is the transformation between the
be described. Section 4 gives a detail for the implementation robot base frame and the workpiece frame. es T and wb T are
of the thermal error compensation method. Section 5 analyzes constants after the robotic system has been constructed, and
the digitizing error of reference sphere measurement. Test and for a same featured point on each product flowing on the
result of the compensation method are given in section 6. The production line, be T is almost invariable as the robot follows
paper will finish in section 7 with a short conclusion and a the programmed trajectory repeatedly, so changes in the mea-
summary. sured results will accurately reflect the deviation in the prod-
ucts’ dimension.
The robotic visual inspection system researched in this
2 Robotic visual inspection system and its thermal work is constructed by an ABB IRB2400 industrial robot
behavior and an off-the-shelf visual sensor nxSensor-I 3D made by
NextWare. Inc. The visual sensor is a line-structured laser
The robotic flexible and online inspection system is based on sensor that works on the principle of triangulation. With high
the multi-joints movement and plays a significant role in integration of state-of-art laser technology, electronic imaging,
quality monitoring and control of products on the assembly and a digital signal processor (DSP), the nxSensor-I 3D laser
line. As shown in Fig. 1, the system mainly consists of an sensor has an accuracy of 25μm and a resolution up to 5μm. It
industrial robot, serving as the orienting device, and a non- could be connected to the computer via universal serial bus
contact visual sensor. In the practical production, a robot (USB). As for the IRB2400 industrial robot, its unidirectional
trajectory is firstly planned according to the distribution of repeatability is specified to be 0.060 mm, which we have
the features on the inspected product, and then the industrial evaluated to be 0.063 mm using the laser tracker.
robot drives the visual sensor to inspect every product on the Although robotic online inspection is a kind of repeated
assembly line following the planned trajectory. The coordinate work and the robot repeatability is the primary element for
systems of the robotic visual inspection system consist of system accuracy, absolute calibration of the robot was per-
robot base frame (BF), end-effector frame (EF), laser sensor formed first in order to eliminate the influence of robot posi-
frame (SF), and workpiece frame (WF), as shown in Fig. 1. tioning error due to geometric errors and compliance errors.
The measured results are usually unified and assessed in the The robot calibration procedure is based on the method pro-
workpiece frame. Comparing with dimension specifications, posed by Albert Nubiola [19] using a laser tracker as the
quality warning information will be generated when there is a measurement device. A 29-parameters calibration model is
product dimensionally out of tolerance. For a visual point P on established to take all possible kinematic errors and compli-
the workpiece, the relationship between the coordinate Pw in ance in joints 2, 3, 4, and 5 into account. When the robotic
the workpiece frame and Ps in the laser sensor frame is: inspection system works in the field, position and orientation
w e
of the robot end-effector (be T ) are calculated with the calibrated
b
Pw ¼ b T ⋅ e T ⋅ s T ⋅ Ps ð1Þ kinematic parameters.
Based on the above analysis, the variations in the system
measuring result are mainly caused by the dimensional chang-
es in the measured objects. However, when the robotic inspec-
tion system has run at a high speed for a long time, it is
observed that there is an obvious drift tendency in the mea-
sured results of a same featured point. And the drift still
exhibits when the system is adopted to inspect a same product
repeatedly over a long period. Figure 2 shows the nature and
magnitude of the errors in the measured results, comparing
with the standard specification, of a featured hole over a
period of about 3 h, during which about 100 measurement
cycles are performed. An upward drift with a magnitude of
0.2 mm was detected in the y-axis and z-axis, and a downward
drift with a magnitude of 0.4 mm was detected in the x-axis.
Since the robot has followed the same trajectory and the
Fig. 1 Schematic of the robotic visual inspection system inspected product has been clamped in position, only the
936 Int J Adv Manuf Technol (2014) 75:933–946

Fig. 2 Thermally induced error


of a inspected featured hole

temperature on the robot arm varies, the detected drifts are 3.1 Error model for industrial robot
considered to be attributed by the robot thermal errors.
As pointed out in [23], one of the error sources that cause Industrial robot is mostly designed in an open kinematic chain
a significant performance variation is the effect of thermal type construction with a series layout of the axes elements
changes on the structure of the robot. The thermally in- (joints) and the axes connecting elements (links). In this paper,
duced error has been described as the most significant the mathematic representation chosen to calculate the trans-
element of the total error in the robots’ long duration be- formation of coordinates between two consecutive links is
havior. Internal and external heat sources will cause chang- defined in [26], also known as the Denavit–Hatenberg (DH)
es in the links and joints dimensions, as well as unpredict- convention. For two consecutive link frames, i-1th and ith link
able structure variations, all of these will changes the robot frame, four link parameters, namely joint angle θi, link offset
kinematic behavior and cause the robot end-effector devi- di, link twist αi, and link length ai are used to represent the
ates from the pre-planned positions. For the robotic visual relative translational and rotational transformation as follows:
inspection system researched in this paper, links expansion
2 3
and structure deformation during the robot warm up and Cθi −Sθi Cαi Sθi Sαi ai Cθi
cool down will cause the position and orientation of the 6 −Cθi Sαi ai Sθi 7
i 1 6 Sθi
iT ¼ 4
Cθi Cαi 7 ð2Þ
robot end-flange be T deviate from its original value at cool 0 Sαi Cαi di 5
status. As the laser sensor is fixed on the robot end-flange 0 0 0 1
severing as a tool, the relative position of the featured point
Ps in the laser sensor frame will change as well. If the
original value for ebT is still used when transforming the where Sθi, Cθi, Sαi, and Cαi represent sinθi, cosθi, sinαi,
measured result of laser sensor to the workpiece frame, the and cosαi, respectively. As the joints are rotational, only θi is
result in the workpiece frame will inevitably deviate from the joint variable and di, αi, and ai are constants.
the value at cool status. For the serial robotic manipulator with N degrees of free-
dom, the global transformation between end-effector and the
robot base frame is calculated by pre-multiplying each of the
3 Formulation of the thermal error compensation transformation matrices between a frame and the previous
methodology one:

b b
According to the aforementioned considerations, tempera- eT ¼ 1T 12 T 23 T ⋯ N N1T ð3Þ
ture variation has a profound effect on the performance of
robotic visual inspection system, modeling its influence is The error model of the robot is used to investigate the
an important and desirable issue in order to improve the relationship between end-effector positional and orientational
system measuring accuracy. As the robotic visual inspec- errors with respect to the kinematic parameter errors.
tion system generally runs on the industrial field without Supposing that there are errors in the kinematic parameters
intermission, the thermally induced error should be com- of every link, the actual transformation between robot end-
pensated online during its operation. In this paper, we effector and the robot base frame can be given by:
developed a real-time dynamic thermal error compensation
method for the robotic visual inspection system, which is a N n
b b
parametric calibration method and is developed upon the eT þ d eT ¼ ∏ 1
n T þd TÞ
n 1
n ð4Þ
n¼1
robot kinematic model and error model. So, the robot error
model is briefly reviewed first, followed with a detailed
analysis and description on the thermal error compensation If the parameter errors are small and the higher order terms
approach. are negligible, the differential transformation in Eq. (4) can be
Int J Adv Manuf Technol (2014) 75:933–946 937

represented by a linear function of the individual kinematic parameter errors. J is the identification Jacobian matrix
parameter deviation as follows: defined in Eq. (10), it outlines how each kinematic
parameter error influences the robot positional and ori-
∂ n n1T ∂ n n1T ∂ n n1T ∂ n n1T entational accuracy.
d n n1T ¼ Δθi þ Δd i þ Δai þ Δαi
∂θi ∂d i ∂ai ∂αi
 
ð5Þ Mθ Md Ma Mα
J¼ ð10Þ
Rθ 0 0 Rα
where Δθi, Δαi, Δai, and Δdi are small errors for robot
kinematic parameters.
Expanding Eq. (4) and neglecting the high order terms, the 3.2 Thermal error compensation algorithm based
differential transformation of the robot end-effector with re- on fixed-point constraint
spect to the robot base frame can be approximated as in
Eq. (6): The novel thermal error compensation approach proposed in
! this paper is based on fixed-point constraint. Suppose there is
b
X
N b
∂eT
b
∂eT ∂eT
b b
∂eT a reference point fixed in the robot workspace, and the robotic
deT ¼ Δθi þ Δd i þ Δai þ Δαi
i¼1
∂θi ∂d i ∂ai ∂αi visual inspection system has measured the reference point at
cool status and warm status from a same robot pose. Before
ð6Þ
thermal error compensation, the measured results in laser
b i 1
∂ eT b ∂ T sensor frame will be transformed to the robot base frame by
where ∂q ¼ 1T ⋅ 12 T ⋅ ⋯ ⋅ ∂q ⋅ iþ1T ⋅ ⋯ ⋅ N N1T , qi denotes
i i
i i
adopting the original end-effector orientation and position at
one of the kinematic parameters θi, di, αi and ai. Following
cool status, which are represented as follows:
Paul [27], we define the error matrix as:
b e b e
b b
d e T ¼ δ e T⋅e T
b
ð7Þ Pcb ¼ e T ⋅ s T ⋅ Pcs ; Pwb ¼ e T ⋅ s T ⋅ Pws ð11Þ

2 3
0 −δez δey Δex where P cb and Pwb are the measured results in the robot base
6 δez 0 −δex Δey 7
where δ e T ¼ 6 7. Define ΔE ¼
b
4 −δey δex frame at cool status and warm status, respectively. P cb will
0 Δez 5
serve as the reference to be compared at other temperatures.
0 0 0 0
2 3 2 3 As analyzed in section 2, the position and orientation of the
Δex δex
4 Δey 5, ΔΘ ¼ 4 δey 5 , and we can rewrite the robot error robot end-flange at warm status will deviate from their values
at cool status and the relative position of the reference point in
Δez δez
the laser sensor frame will change, that is Pcs ≠Pws . So, the
model in a matrix equation as follows:
measured result of the reference point in Eq. (11) will not be
          consistent. Taking the deviation of the robot position and
ΔE Mθ Md Ma Mα
¼ Δθ þ Δd þ Δa þ Δα orientation into consideration, the actual position of the refer-
ΔΘ Rθ 0 0 Rα ence point in the robot base frame at the warm status is given
ð8Þ as follows and it will be consistent with the measured result at
cool status.
where Δθ=[Δθ1Δθ2 ⋯ΔθN]T, Δd=[Δd1Δd2 ⋯ΔdN]T,  
e e
Δa = [Δa1Δa2 ⋯ ΔaN]T, Δα = [Δα1Δα2 ⋯ Δα N]T, Δθi, b ¼
Pwa
b
eT
b b
þ d e T ⋅ s T ⋅ Pws ¼ Pcb ¼ e T ⋅ s T ⋅ Pcs ð12Þ
Δdi, Δai, and Δαi are small deviations of robot kinematic
parameters. Mθ, Md, Ma, Mα, Rθ, Rα are 3×N position-
dependent constant matrices of partial derivatives of the end- where dbe T represents the thermally induce error for the
effector with respect to the kinematic errors, and they are position and orientation of the robot end-effector relative to
independent of temperature changes. In an alternative way, the robot base frame, Pwa
b is the actual position of the reference
Eq. (8) can be written in a compact form as: point at warm status. According to Eqs. (11) and (12), we can
calculate the deviation between the measured results of the
ΔX ¼ J ⋅ΔP ð9Þ reference point at the cool status and warm status as follows:

b 
where ΔX=[ΔEΔΘ]T is the robot end-effector’s po- Pcb −Pwb ¼ Pwa
b −P b ¼
w
eT þ d be T ⋅es T ⋅Pws − be T ⋅ es T ⋅ Pws
sitional and orientational errors in the robot base frame,
ΔP = [ΔθΔdΔaΔα] T represents the robot kinematic ¼ d be T ⋅ es T ⋅ Pws ð13Þ
938 Int J Adv Manuf Technol (2014) 75:933–946

Basing on the robot differential kinematic in Eq. (7), we laser sensor to measure the position of sphere center at differ-
have: ent poses. If the internal heat sources or the ambient heat
sources cause robot link expansion or structure distortions,
Pcb −Pwb ¼ δbe T ⋅ be T ⋅ es T ⋅ Pws ¼ δbe T ⋅ Pwb the robot kinematic parameters will deviate from the original
2 3 2 wx 3 values. Based on the thermal error calibration method present-
0 −δez δey Δex Pb
6 δez 0 −δex Δey 7 6 Pwy 7 ed in section 3.2, the thermally induced parameter errors will
¼64 −δey
7⋅6 b 7 ð14Þ
δex 0 Δez 5 4 Pwzb
5 be identified dynamically. Since the line-structured laser sen-
0 0 0 0 1 sor usually has a limited field of view, the standard sphere is
not very large in dimension. And considering the limited
In an alternative way, the first three elements in Eq. (14) motion volume when the robot measures a standard sphere,
can be rewritten as follows: more than one standard spheres are adopted and fixed in the
robot workspace at different positions, so that the robot will
2 3
Δex get adequate movement. The final implementation of the
2 3 6 Δey 7 thermal compensation methodology is shown schematically
1 0 0 0 Pwz −Pwy 6 7
b b 6
wx 5 6 Δez 7
7 in Fig. 3. Three identical standard spheres are mounted at the
Pcb −Pwb ¼ 4 0 1 0 −Pwz 0 P ⋅ 6 7
6 δex 7
b b both sides of industrial robot with different heights and differ-
0 0 1 Pwy b −Pwx
b 0 4 δey 5 ent distances relative to the robot base. In order to maintain the
δez robot’s accuracy during its long-term operation, thermal errors
must be estimated and compensated in real time. The robot
¼ Q⋅ΔX ¼ Q⋅J ⋅ΔP ð15Þ
drives the laser sensor to measure the spheres at regular
It can be seen from Eq. (15) that the deviation between intervals, and the thermally induced parameter errors are
measured results of the reference point at warm status and cool identified and compensated dynamically by comparing the
status has been associated to the thermally induced parameter measured sphere centers with the sphere centers measured at
errors by a linear position-dependent function. Let Ai =Qi ⋅J i cool status. It can be seen from the figure that the joints 1, 2, 3
of the robot have to move in order to measure all the three
b −Pb ), and suppose we have measured the refer-
and Yi =(P ci wi

ence point in m (m>4×N/3, N; robot degrees of freedom) spheres and the other three joints have to rotate in order to
different robot poses, we can obtain the following matrix measure the spheres at different poses, so the robot will get
equation: adequate movement and the measured positions of the sphere
centers will fully reflect the thermal errors of the robot.
Y ¼ A ⋅ ΔP ð16Þ Tungsten carbide with 6 % cobalt added as a binder was
selected as a promising material for the calibration spheres. It
In Eq. (16), the deviations in the kinematic parameters are is extremely hard (9.5 on the Mohs hardness scale) with low
the parameters to be identified. A singular value decomposi- coefficient of thermal expansion. It has a high resistance to
tion (SVD) method, which is a rapid and computationally thermal shock and is able to withstand the temperature fluc-
efficient algorithm, can be used to solve the matrix equation tuations. Moreover, it is commercially available in the form of
as follows: precision ball bearings, and the off-the-shelf standard ball
 
ΔP ¼ U ⋅ S þ ⋅ V T Y ð17Þ

where U, V are the right and left singular matrix of A, and


Laser sensor
S + is the singular value matrix.

Industrial robot
4 Implementation of the thermal error compensation
methodology
Standard sphere
As stated previously, the real-time thermal error compensation
method presented in this paper is based on the fixed-point
constraint. In the practical implementation, standard sphere is
adopted as the calibration target and the sphere center is
referred to as the fixed reference point. For the robotic visual
inspection system, the standard sphere is mounted in the robot
workspace and the industrial robot drives the line-structured Fig. 3 Implementation of the thermal error compensation method
Int J Adv Manuf Technol (2014) 75:933–946 939

bearings are available in convenient sizes for the purpose of


thermal calibration. The selected sphere diameter is 38.1 mm
Laser tracker Industrial robot
and the spheres have first been inspected on a CMM by
sampling 20 points on each sphere surface, the sphericity
(the maximum deviation of the points from the least-squares
fitted sphere) of the spheres is found to be less than 10μm. In
order to maintain the nominal positions of standard spheres
and to assume negligible thermal effects over the standard
sphere, all the support units are produced in carbon fiber tube,
with a nominal coefficient of thermal expansion (CTE) of
0.7×10−6K−1.
Theoretically, all link parameters will deviate when the
robot warms up and cools down, so all the link parameter
errors need to be identified in Eq. (17). Each sphere center
measurement provides three parametric equations, so at least Fig. 4 Experimental setup for investigating the robot thermal behavior
eight sphere center measurements (24 parametric equations)
are needed for an entire identification of the parameter errors.
In order to improve the identification accuracy, over- center of the SMR is referred to as tool center point (TCP) in
determined equation system whose parametric equations are the thermal error investigation.
3~5 times of parameters to be identified is usually construct- When the robot rotates joint-by-joint, link-expansion or
ed. It means 30~40 sphere center measurements are needed structure distortions will only exhibit at the rotated joint, and
for each thermal error compensation procedure. However, the measured thermal errors are compensated by just calibrat-
more sphere center measurements mean longer cycle time ing the link parameters of the rotated joint. In order to find out
for inspecting one product and lower production efficiency the significant parameter terms, the thermal errors are com-
for the assembly line, which is not expected by the manufac- pensated by selecting different parameters and the most effec-
turers. In practice, only some of the link parameters will have tive terms are referred to as the significant parameter terms.
significant changes with respect to the temperature variations. Starting from the cool status, each joint is rotated separately
Therefore, it is of great significance to find out the significant (with the others being blocked). The nature and magnitude of
parameter terms and delete the insignificant parameters in the TCP drift during the robot warm up are determined by the
order to obtain a cost-effective thermal error model and en- laser tracker. Multiple positions are measured by the laser
hance the compensation efficiency. tracker according to the angle extent of each single-joint
In order to find the link parameters that change significant- rotational movement. When the robot has stayed at the home
ly with respect to thermal variations, thermal behavior of configuration for more than 2 h with all the motors power off,
industrial robot must be investigated first. Moreover, for serial the TCP displacement is measured over multiple positions.
robots, the axis coordinates must be transformed into the base Then the tested joint moves at 70 % of the maximum speed for
by means of coordinate transformation. This means that the 3 h to warm up, and the TCP positions are measured again by
deformation of elements of the kinematic chain should be the laser tracker. We define the thermal error at a certain TCP
analyzed and measured individually. So we performed a position as the difference between the TCP displacements
joint-by-joint test to learn the thermal behavior of each joint measured at warm status and cool status, which is given as
with the help of Leica AT901 laser tracker. Figure 4 shows the follows:
experimental setup for investigating the thermal behavior of
the IRB2400 industrial robot. According to the specification, qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
the typical measurement uncertainty (deviation between the Δr ¼ ð x − x0 Þ 2 þ ð y − y0 Þ 2 þ ð z − z 0 Þ 2 ð18Þ
measured and the nominal coordinate of a tested point) of the
laser tracker is ±15 μm+6 μm/m. All measurements were where (x, y, z) and (x’, y’, z’) represent the TCP positions at
performed at ambient temperature varying within a range of cool status and warm status measured by the laser tracker,
±0.5 °C indicated by the environmental thermometer. During respectively. Since the compliance errors and geometric errors
all the measurements, the laser tracker was never displaced are assumed to be constant, the TCP position drift measured
from its location shown in Fig. 4, in order to avoid the during the robot warm-up stage is considered to be attributed
accumulation of errors. For the determination of temperature to thermal variations.
deformation of the industrial robot, an end-effector has been The thermal errors are firstly compensated by calibrating
designed and mounted at the robot end-flange to hold a 1.5 in. all the link parameters of the tested joints and then compen-
spherically mounted reflector (SMR), as shown in Fig. 4. The sated by selecting different parameter terms. After thermally
940 Int J Adv Manuf Technol (2014) 75:933–946

induced link parameter errors calibrated, the residual errors are parameters is almost the same remarkable as the result com-
calculated according to Eq. (18), where (x, y, z) is the TCP pensated with all link parameters. All of these have justified
position at warm status measured by the laser tracker and that the thermal error could be compensated effectively by just
(x’, y’, z’) is the TCP position calculated with the calibrated calibrating the significant parameters. It should be pointed out
link parameters. Figure 5 shows the comparison of thermal that the thermal errors in multiple-joints movements will trade
errors and the residual errors after error calibration, only the off against each other, hence the thermal deflection measured
compensation results by all the link parameters and the most here is the worst case.
significant terms are given. It can be seen from Fig. 5 that the According to the identification result, link parameters d1,
thermal errors depend not only on the parameter errors, but a2, d2, a3, d4, a5, and d6 have significant changes with
also on the robot positions. The maximum and mean values respect to temperature changes, and these seven parameters
for the thermal errors over the measured positions are given in are selected as the significant terms. d1, a2, a3, d4, a5, and d6
Table 1. Without compensation, the maximum changes of represent the length of links, thermal error Δd1 represents the
TCP displacement during the robot warm-up stage is about expansion of the base frame, and Δa2, Δa3, Δd4, and Δd6
0.2 mm, and it is most significant on joint 2, which exhibits a represent the links expansion while the robot warm up, Δd2
thermal error as large as 0.413 mm. It was also observed from and Δa5, represents the joint expansion or distortions. And
a thermocouple, attached to the “hot spot” (the hottest location the selection of the significant parameters can also explain the
on the outside of structure associated with each of the joint significant change direction in Table 1 well. According to the
motors), that the record temperature on joint 2 has increased DH kinematic model, joint angle θi and link twist αi are angles
by about 17 °C. After compensation, the maximum residual between the adjacent x-axes and y-axes, respectively, which
error for TCP displacement has been decreased to less than are structural parameters and will not change significantly
0.1 mm and the mean value has been decreased to about with respect to the temperature change. As the angle encoders
0.06 mm. It is also noted that the maximum/mean error for at the joints are heated uniformly while the motors heating, the
the TCP displacement compensated with the significant joint angles will have no significant changes while the robot

Fig. 5 Compensation of the


thermal error at each robot joint. a
Joint 1, b joint 2, c joint 3, d joint
4, e joint 5, f joint 6

(a) Joint 1 (b) Joint 2

(c) Joint 3 (d) Joint 4

(e) Joint 5 (f) Joint 6


Int J Adv Manuf Technol (2014) 75:933–946 941

Table 1 Measurement informa-


tion and compensation result for Joints Interval Significant Significant Maximum/mean error (mm)
the joint-by-joint test directions parameters
Not compensated All compensated Partial compensated

1 −40°,40° z d1 0.143/0.102 0.074/0.052 0.083/0.058


2 −20°,50° y a2, d2 0.413/0.298 0.096/0.066 0.093/0.075
3 −20°,50° y a3 0.237/0.156 0.070/0.057 0.084/0.060
4 −150°,150° x, z d4 0.209/0.190 0.062/0.042 0.081/0.049
5 −90°,90° x, z a5 0.203/0.177 0.095/0.052 0.076/0.050
6 −180°,180° x, z d6 0.150/0.141 0.051/0.044 0.056/0.045

warms up. Link offset di and link length ai are the distance As the laser sheet projected from a semiconductor laser
between the adjacent x-axes and y-axes, respectively, which diode has a certain thickness, the intersected arc-shaped con-
are physical parameters of the links and will change signifi- tour will have a certain width. The width effect of the laser
cantly with respect to the temperature change. Various selec- sensor is there will be a deviation between the identified center
tions of link parameters were tried for each joint, and the of the laser contour and the actual geometric center of the laser
compensation results showed that none of the other selections stripe on the sphere surface. Taking the effect of the projected
is significantly better than these seven terms. angle into account, the laser contour will be wider when the
projected angle is larger, and more deviation will exist in the
identified stripe center, as analyzed in [29]. So, when the line-
5 Analysis of digitizing error in reference sphere structured laser sensor is adopted to inspect a standard sphere,
measurement a salient deviation will be yielded when the intersecting circle
is far away from the sphere center, where the projected angle is
The standard sphere is adopted as the calibration target in the large.
robot thermal error compensation, and its center acts as the The normal vector of the light plane (a, b, c), the circle
physical constraint of fixed-point in the robot workspace. center Oc (xz, yz, zz), and the sphere center Op (xc, yc, zc) under
Therefore, the accurate position measurement of the sphere the laser sensor frame satisfy the following relationship:
center with the line-structured laser sensor is essential to 8
ensure the effectiveness of the thermal error compensation < xc ¼ xz  at
approach. Based on the principle of triangulation, the line- y ¼ yz  bt ð20Þ
: c
structured laser sensor is able to obtain the 3D coordinates zc ¼ zz  ct
from the laser line projected on the surface to be measured. It
mainly consists of a camera (CCD and lens) and a laser diode
with a cylindrical lens capable of projecting a stripe. A contour
line of laser light is created when the laser shots on the surface
to be inspected, then the laser contour is imaged in the image
row-column of the CCD camera array. Based on the calibrated
measurement model, the coordinates of each spot on the laser
contour can be reconstructed by intersecting the correspond-
ing camera projection ray with the laser plane.
Method of measuring the sphere center is illustrated in
Fig. 6, when the laser is projected on the sphere surface, an
image of the arc-shaped laser contour is captured by the
camera and the point cloud of the arc are extracted based on
the method of Gaussian curve approximation [28], then circle
center Z(xz, yz, zz) and radius r can be determined by fitting a
circle with the point cloud. Defining the angle between laser
plane and the surface normal direction at the measured point
as the projected angle β (angle ∠OPAOc in Fig. 6), then we
have:

cosβ ¼ r=R ð19Þ Fig. 6 Laser sensor measuring a standard sphere


942 Int J Adv Manuf Technol (2014) 75:933–946

Laser sensor (a) Compensation with 18 calibration points and 24 parameters

Linear rail
Standard sphere

(b) Compensation with 18 calibration points and 7 parameters


Fig. 9 Compensation with all parameters and significant parameters. a
Fig. 7 Experimental setup for sphere center measurement Compensation with 18 calibration points and 24 parameters. b Compen-
sation with 18 calibration points and 7 parameters
pffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
where t ¼ R2 −r2 = a2 þ b2 þ c2 represents the dis-
tance between the fitted circle center and the sphere center,
plus or minus is determined by the position of the sphere
center relative to the circle center. Structural parameters of
the light plane (a, b, c) are estimated in the process of sensor
calibration. So the radius of the standard sphere must be
known in order to determine the sphere center at one robot
position by one laser shot, and the relationship between the
sphere radius tolerance and the error for t is given as follows:

ΔR (a) Compensation in x axis


Δt ¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi ð21Þ
  2
a þ b þ c ⋅ 1−ðr=RÞ
2 2 2

According to Eq. (21), when the laser plane is too close to


the sphere center (r/R≈1), a small error in the sphere radius
(ΔR) will result in a considerable error for t, so the measured
sphere center is very sensitive to the sphere diameter tolerance
when the projected angle is small. (b) Compensation in y axis

(c) Compensation in z axis


Fig. 10 Compensation with 18 sphere center measurements and 7
parameters. a Compensation in x-axis. b Compensation in y-axis. c
Fig. 8 Distance error with respect to the measuring position Compensation in z-axis
Int J Adv Manuf Technol (2014) 75:933–946 943

Table 2 Maximum/mean value for errors in the three coordinated axes and the experiment has been repeated several times with
(unit: mm)
different rail moving directions. Results similar to Fig. 8 are
Axis Maximum/mean value at Maximum/mean value at exhibited. With the experiments and statistical analysis, we
sphere centers tested points can conclude that the projected angle for most accurate sphere
center measurement is within the area of 20°~40°, and the
Measured Compensated Measured Compensated
radius of intersecting circle for the best measurement is within
x 0.345/0.073 0.054/0.030 0.251/0.122 0.083/0.045 the area of R×cos20o ~R×cos40°.
y 0.190/0.096 0.055/0.023 0.134/0.049 0.074/0/043
z 0.263/0.131 0.073/0.029 0.241/0.122 0.068/0/039
6 Test and results

According to the above discussion, measurement uncer- In order to verify the effectiveness of the thermal error com-
tainty of the sphere center will become great when the pensation method proposed in this paper, experiment is per-
projected angle is either too large or too small, so there will formed on the robotic visual inspection system installed on an
be an optimum measurement area where the line-structured automotive assembly line. The robotic visual inspection sys-
laser sensor can project the laser stripe in order to obtain the tem is constructed as described in section 2, and it is devoted
best measurement accuracy of the sphere center. Due to the to monitor the quality of car body assembly and sub-assembly.
difficulty of analytically describing the relationship between Three standard spheres were installed according to the imple-
the measurement accuracy and projected angle, the optimum mentation of the thermal error compensation method present-
projected angle for sphere center measurement is going to be ed in section 4.
obtained base on experimental data. As shown in Fig. 7, an In order to eliminate the influence of dimensional devia-
experimental setup is installed to find the best projected angle tions among different car bodies, a car body was clamped in
for sphere center measurement by the line-structured laser position. Ten featured holes on the car body are selected as the
sensor. The laser sensor is mounted on the robot end-effector inspection targets, and a robot trajectory has been planned to
and a standard sphere with a diameter of 38.1 mm is fixed on a measure the featured holes. Starting from the cool status, the
one-dimensional linear rail. Adjust the image plane of the robot drives the laser sensor to measure the featured holes
sensor to be nearly parallel with the rail plane and move the repeatedly at 70 % of the maximum speed over a period of 3 h.
reference sphere into the sensor’s field of view at the sensor Without thermal error compensation, the measured positions
working offset. Keep the laser sensor quiescent and move the of the featured holes will drift similarly as Fig. 1. In order to
reference sphere with a step of 1 mm by controlling the linear compensate the thermal error, 18 sphere center measurements
rail, the moving step can be calculated with the measured (6 points at each sphere) are performed based on the measure-
sphere centers in laser sensor frame and it is also measured ment strategy described in section 5. Before compensation,
by a high-resolution grating ruler. Figure 8 shows the step the sphere centers and hole-centers in the robot base frame are
error between the measurement of the grating ruler and the calculated with the original link parameters and the measured
laser sensor, along with the fitted radii of the intersecting results serve as the reference to be compared at the warm
circles. status. The thermal errors are calculated according to
From Fig. 8, we can see that the reference sphere has Eq. (18) by comparing the measured results at warm
moved 31 mm along the linear rail and the range of the status with the measured results at cool status. Based on
projected angle is about 0°~60°. The distance error curve the thermal error compensation algorithm presented in
shows that great errors will exhibit when the intersecting circle section 3.2, the thermally induced link parameter errors are
is close to the sphere center (projected angle is small) and it identified. After thermal error calibration, the sphere centers
will decline sharply while the intersecting circle deviate from and hole-centers at the warm status are recalculated with the
the sphere center, but when it moves too far away from the identified link parameters and the residual errors are calculat-
sphere center (projected angle is large), the step error goes up ed according to Eq. (18) by comparing the compensated
clearly. The moving direction of the linear rail in the laser sphere centers and hole-centers with the measured values at
sensor frame can be changed by rotating the robot’s last joint cool status.

Table 3 Nominal value and cali-


brated deviations for the signifi- Parameters d1 a2 d2 a3 d4 a5 d6
cant parameters (unit: mm)
Nominal 615 705 0 135 755 0 85
Deviation 0.1272 0.0616 −0.0742 0.0574 0.2210 0.0265 0.0124
944 Int J Adv Manuf Technol (2014) 75:933–946

In order to investigate the effect of the thermal error com-


pensation method more profoundly, thermal errors and resid-
ual errors are calculated in each of the three coordinate axes by
subtracting the warm values by the cool values, and the errors
in the x-, y-, and z-axes are illustrated in Fig. 10. The thermal
errors are compensated by just calibrating the significant
parameter terms. And the maximum/mean values for the
thermal errors in the three axes are given in Table 2. As the
(a) Compensation with 12 calibration points and 7 parameters thermal errors at the coordinated axes have possibility for plus
or negative, the maximum/mean values are calculated with the
absolute values. It can be seen from the Fig. 10 that there are
thermal errors in all the three axes, and it is most remarkable in
x-axis where the maximum error is up to 0.345 mm. After
compensation with the significant parameters, the mean errors
of the measured sphere centers have been reduced to about
0.03 mm in all the three axes. And the compensation effect at
the tested points is also slightly worse than the sphere centers
(b) Compensation with 9 calibration points and 7 parameters because the tested points have not been used to identify the
Fig. 11 Result of compensation with different sphere center measure- parameter errors. The calibrated deviation for the significant
ments. a Compensation with 12 calibration points and 7 parameters. b parameters is given in Table 3. It can be seen from the table
Compensation with 9 calibration points and 7 parameters that d1 and d4 have the most significant changes with respect
to the temperature change which represent the expansion of
The thermal errors and the compensation results are shown the robot base and the link 4. With all the link parameters
in Fig. 9. The dashed line separates the sphere centers and the calibrated, the identified results of the seven significant pa-
tested points. Before compensation, the maximum/mean value rameters are similar to the result in Table 3, and the identified
for the thermal errors at the sphere centers and the tested results of the other non-significant parameters are on the order
points is 0.441/0.195 mm. With all the parameters or only of 0.001 mm.
the significant parameters compensated, residual error at the In the practical application, it’s desirable to decrease the
18 calibrated postures and the 10 tested postures are calculated sphere center measurements as much as possible in order to
separately. After compensation by calibrating all the parame- increase the production efficiency. But the number of sphere
ters, the maximum/mean values for the residual errors at the center measurements plays a significant role in the parameter
sphere centers and tested hole-centers are 0.079/0.041 identification. In order to investigate the relationship between
and 0.096/0.064 mm, respectively. And the residual the number of calibration points and the compensation effect,
errors obtained after temperature correction with signif- we have repeated the above experiment with different num-
icant parameters also show a favorable improvement bers of sphere center measurements. Since many of the exper-
with respect to the non-corrected values, the maximum/mean iments have similar results, only two of them, with calibration
values for the residual errors are 0.092/0.051 and 0.136/ and testing, are presented here for brevity, that are compensa-
0.078 mm, respectively. The residual errors at the tested tion with 12 sphere center measurements (4 points at each
hole-centers are slightly larger than the errors at the sphere sphere) and compensation with nine sphere center measure-
centers, which can be explained by the fact that the sphere ments (3 points at each sphere). The compensation results are
centers are used to identify the link parameter errors while the shown in Fig. 11, from which we can see that the residual
tested points not. errors after compensation have increased with less sphere

Fig. 12 Dynamical A B C
compensation test result
Int J Adv Manuf Technol (2014) 75:933–946 945

center measurements. After compensation with 12 sphere industrial robot warm-up/cool-down stage. Seven significant
center measurements, the maximum/mean values for the re- link parameters have been selected and justified based on a
sidual errors at the sphere centers and the hole-centers are joint-by-joint test. Based on this, the best implementation of
0.107/0.452 and 0.135/0.823 mm, respectively. And after the method is proposed by using standard spheres as the
compensation with nine sphere center measurements, the calibration targets and the optimum projected angle for sphere
maximum/mean values for the residual errors at the sphere center measurement by line-structured laser sensor is investi-
centers and the hole-centers are 0.203/0.127 and 0.192/ gated. Experimental results show that by applying this meth-
0.087 mm, respectively. Although the results show a obvious odology the max/mean residual error of the measured result
improvement with respect to the non-corrected values, it’s still has been reduced from 0.441/0.195 to 0.136/0.078 mm with
recommended to guarantee more than 12 sphere center mea- only the significant parameters calibrated. It has proved that
surements are performed (4 points at each sphere) to obtain an the method is feasible and valid in maintaining the accuracy of
approving thermal error compensation effect. the robot visual inspection system during its long-term oper-
Since the method presented in this paper is designed to ation. The method has been successfully applied to the car
compensate the robot thermal error dynamically during the body-in-white visual inspection stations installed on the car-
robotic visual system operation, an online compensation ex- body assembly lines. Since it is easy and convenient to im-
periment is also carried out. A robot trajectory is planned to plement, we expect the proposed thermal error compensation
measure the sphere centers and the featured holes on the car method will have wide application on the manufacturing floor
body. Eighteen sphere center measurements (6 points at each for robot online accuracy enhancement and maintenance.
sphere) are performed to calibrate the thermally induced pa- Future effort will also be devoted to extend application of this
rameter errors. The robot repeats the planned trajectory at method to other robotic systems such as articulated arm coor-
70 % of the maximum speed, and the laser sensor measures dinate measuring machines and so on.
the sphere centers and the featured holes about every 2 min.
Then, thermally induced parameter errors are calibrated with
the measured sphere centers and the thermal errors at the Acknowledgments The work was supported by National High Tech-
nology Research and Development Program of China (863 Program,
tested points are compensated in real time. The experiment No.2012AA041205); National Natural Science Funds for Distinguished
lasts about 3 h and the featured holes are measured about 100 Young Scholar (51225505); National Natural Science Foundation of
times. The compensation result of one of the featured holes is China (51305297). The authors would like to express their sincere ap-
illustrated in Fig. 12, which has been divided into three stages. preciation to them, and comments from the reviewers and the editors are
very much appreciated.
Starting from the cool status, it can be seen from stage A that
there is a clear drift tendency while the robot warms up. As
can be seen from the figure, once applying the thermal error
compensation in stage B, the measured error for the tested References
point has been closely reduced by an order of magnitude
compared with thermal errors before compensation. With the
1. Brosed F, Santolaria J, Aguilar J, Guillom AD (2012) Laser triangu-
thermal error compensation procedure cancelled in stage C, lation sensor and six axes anthropomorphic robot manipulator model-
the measured error for the tested point rebounds to the drift ling for the measurement of complex geometry products Robot.
curve in stage A immediately. The deviations between the Comput Integr Manuf 28:660–671
mean value of the measure errors in stage B and the error at 2. Larsson S, Kjellander J (2006) Motion control and data capturing for
laser scanning with an industrial robot. Robot Auton Syst 54:453–
the cool status are in the range of ±0.1 mm for all the three 460
axes. 3. Xie SQ, Cheng D, Wong S, Haemmerle E (2008) Three-dimensional
object recognition system for enhancing the intelligence of a KUKA
robot. Int J Adv Manuf Technol 38(7–8):822–839
4. Paoli A, Razionale AV (2012) Large yacht hull measurement by
7 Conclusions integrating optical scanning with mechanical tracking-based method-
ologies Robot. Comput Integr Manuf 28:592–601
A real-time dynamic thermal error compensation method for 5. Liu CJ, Yang XY, Zhu JG, Ye SH (2006) Flexible coordinate mea-
robotic visual inspection system has been presented in this surement system based on industrial robot for car body-in-white. J
Optoelectron Laser 17:207–210
paper. This method is designed to be applied on the production 6. Conrad KL, Shiakolas PS, Yih T (2000) Robotic calibration issues:
line and is capable of correcting the errors due to temperature Accuracy, repeatability and calibration Proc. 8th Medit. Conf. on
changes during the system operation. It is a parametric cali- Control and Automation (MED2000). pp 17-19
bration method developed without using temperature sensors. 7. Roth Z, Mooring B, Ravani B (1987) An overview of robot calibra-
tion. IEEE J Robot Autom 3:377–385
An error calibration model is derived for associating the 8. Ha IC (2008) Kinematic parameter calibration method for industrial
thermally induced parameter errors with the differences be- robot manipulator using the relative position. J Mech Sci Technol 22:
tween the measured results of a reference point during the 1084–1090
946 Int J Adv Manuf Technol (2014) 75:933–946

9. Bai Y (2007) On the comparison of model-based and modeless 19. Nubiola A, Bonev IA (2012) Absolute calibration of an ABB IRB
robotic calibration based on a fuzzy interpolation method. Int J Adv 1600 robot using a laser tracker Robot. Comput Integr Manuf 29:
Manuf Technol 31(11–12):1243–1250 236–245
10. Gatti G, Danieli G (2008) A practical approach to compensate for 20. Slamani M, Nubiola A, Bonev IA (2012) Modeling and as-
geometric errors in measuring arms: Application to a six-degree-of- sessment of the backlash error of an industrial robot. Robotica
freedom kinematic structure Measu. Sci Technol 19:015107 30:1167–1175
11. Drouet P, Dubowsky S, Zeghloul S, Mavroidis C (2002) 21. Eastwood S, Webb P (2009) Compensation of thermal deformation
Compensation of geometric and elastic errors in large manipulators of a hybrid parallel kinematic machine. Robot Comput Integr Manuf
with an application to a high accuracy medical system. Robotica 20: 25:81–90
341–352 22. Santolaria J, Yag EJA, Jimnez R, Aguilar JJ (2009) Calibration-based
12. Aoyagi S, Kohama A, Nakata Y, Hayano Y, Suzuki M (2010) thermal error model for articulated arm coordinate measuring ma-
Improvement of robot accuracy by calibrating kinematic model using chines. Precis Eng 33:476–485
a laser tracking system-compensation of non-geometric errors using 23. Heisel U, Richter F, Wurst KH (1997) Thermal behaviour of indus-
neural networks and selection of optimal measuring points using trial robots and possibilities for error compensation. CIRP Ann
genetic algorithm IEEE/RSJ Int. Conf. on Intelligent Robots and Manuf Technol 46:283–286
Systems (IROS). pp 5660-5665 24. Poonyapak P, Hayes MJD (2006) Towards a Predictive Model for
13. Hollerbach JM (1989) A survey of kinematic calibration. In: Khatib Temperature-Induced Deformation of an Industrial Robot Proc of
O, Craig JJ, Lozano T (eds) The robotics review, 1st edn. MIT Press, EuCoMeS the first European Conf. on Mechanism Science,
Cambridge, pp 207–242 Obergurgl (Austria)
14. Mooring BW, Roth ZS, Driels MR (1991) Fundamentals of manip- 25. Gong C, Yuan J, Ni J (2000) Nongeometric error identification and
ulator calibration. Wiley, New York compensation for robotic system by inverse calibration Int. J Mach
15. Driels MR, Swayze LW, Potter LS (1993) Full-pose calibration of a Tools Manuf 40:2119–2137
robot manipulator using a coordinate-measuring machine. Int J Adv 26. Denavit J, Hartenberg RS (1955) A kinematic notation for
Manuf Technol 8(1):34–41 lower-pair mechanisms based on matrices. ASME J Appl Mech
16. Santolaria J, Brau A, Velzquez J, Aguilar J (2010) A self-centering 22:215–221
active probing technique for kinematic parameter identification and 27. Paul RP (1981) Robot manipulators: mathematics, programming, and
verification of articulated arm coordinate measuring machines. Meas control: the computer control of robot manipulators
Sci Technol 21:055101–055111 28. Anchini R, Beraldin J, Liguori C (2007) Subpixel location of discrete
17. Yin S, Ren Y, Zhu J, Yang S, Ye S (2013) A vision-based self- target images in close-range camera calibration: a novel approach
calibration method for robotic visual inspection systems. Sensors Electronic Imaging 2007, International Society for Optics and
13:16565–16582 Photonics. pp 649110-649118
18. Jang JH, Kim SH, Kwak YK (2001) Calibration of geometric and 29. Feng HY, Liu Y, Xi F (2001) Analysis of digitizing errors of a laser
non-geometric errors of an industrial robot. Robotica 19:311–321 scanning system. Precis Eng 25:185–191

You might also like