You are on page 1of 15

Robotics and Computer–Integrated Manufacturing 86 (2024) 102646

Contents lists available at ScienceDirect

Robotics and Computer-Integrated Manufacturing


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

Robotic compliant grinding of curved parts based on a designed active


force-controlled end-effector with optimized series elastic component
Haiqing Chen , Jixiang Yang *, Han Ding
School of Mechanical Science and Engineering, State Key Laboratory of Intelligent Manufacturing Equipment and Technology, Huazhong University of Science and
Technology, Wuhan 430074, PR China

A R T I C L E I N F O A B S T R A C T

Keywords: High-accuracy, fast-response, and low-overshoot force control are important to guarantee the material removal
Robotic grinding accuracy and surface quality of robotic compliant grinding. In order to achieve the above target, this study
Curved parts develops an active compliant force-controlled end-effector based on a series elastic actuator for the robotic
Active compliance
grinding of curved parts. Firstly, a decoupled robotic grinding system composed of an industrial robot and an
End-effector
Force control
end-effector is developed, and a novel force-controlled end-effector is designed by adding an elastic component
between the servo motor and the load to improve compliance. Secondly, the influences of the elastic component
and grinding tool stiffness on the stability of the force-controlled end-effector system are analyzed by establishing
a contact model of the compliant end-effector and the transfer function of the entire force control system. Then,
the stiffness of the grinding tool and the elastic component are optimized with full consideration of the system
cutoff frequency and the end-effector compliance. To improve the force tracking accuracies of the developed
compliant end-effector, a proportional-integral (PI) controller with first-order differential force feedforward
control is designed. Finally, grinding experiments are conducted for verification. The effects of spring stiffness
and grinding tool stiffness on the force control system are tested, which match well with the theoretical analysis.
Grinding results show that the maximum force control error of the compliant force-controlled end-effector is
decreased by 70% compared to that of the rigid force-controlled end-effector, and the overshoot of the force
control is reduced from 30% to almost 0 under the same response speed. The maximum and average absolute
grinding depth errors using the designed end-effector are reduced by 57.2% and 58.6%, respectively, compared
with those of the traditional rigid end-effector, and the average surface roughness of the finished part is reduced
by 19.2%. Which demonstrates the effectiveness and advantages of the proposed compliant force-controlled end-
effector.

1. Introduction studies. Xiao and Huang [9,10] developed a multi-axis machine tool for
the adaptive belt grinding of blades, and the experimental results
Complex curved parts (e.g., blades and blisks) are widely used as the showed that the grinding efficiency and surface quality after the pol­
core components of the aero-engines, steam turbines, and other energy ishing had obvious improvement. Zhang et al. [11] developed a five-axis
conversion devices. The surface roughness after milling cannot meet the method using an abrasive belt flap wheel to grind the blisks. Compared
requirements because of the tool marks left on the surface, therefore, the with multi-axis machine tools, robots have the advantages of low cost,
milled curved parts must be ground to improve the surface quality good flexibility, and large operating space. Therefore, robotic grinding
[1–4]. At present, manual grinding is still the main finishing method for has attacked attention of both the academic and industrial fields in
complex curved parts. However, the low processing efficiency and ac­ recent years [12–16].
curacy makes manual grinding difficult to satisfy the processing re­ In the process of robotic grinding, because the material removal is
quirements of the complex curved parts [5–8]. directly affected by the contact pressure between the compliant tool and
In order to improve the grinding quality and efficiency, automatic the workpiece, the force control accuracy has a significant influence on
grinding of complex curved parts has attracted attention of a lot of the profile accuracy and the surface quality of the curved parts.

* Corresponding author.
E-mail address: jixiangyang@hust.edu.cn (J. Yang).

https://doi.org/10.1016/j.rcim.2023.102646
Received 7 April 2023; Received in revised form 2 August 2023; Accepted 3 September 2023
Available online 15 September 2023
0736-5845/© 2023 Elsevier Ltd. All rights reserved.
H. Chen et al. Robotics and Computer-Integrated Manufacturing 86 (2024) 102646

Fig. 1. Schematic of the robotic grinding system.

Theoretically, the grinding force could be controlled by adjusting the extensive studies. For example, Calanca et al. [32], Aguirre-Ollinger
relative motion between the robot and the workpiece. In order to control et al. [33], and Zhang [34] used SEA to improve exoskeleton flexi­
the grinding force, Zhao et al. [17] and Lakshminarayanan et al. [18] bility and assist human walking. Kong et al. [35,36] designed a rotary
developed impedance controllers to achieve compliant grinding, in SEA and used it in human assistant robots to improve the performance
which the dynamic relationship between the contact force and robotic during interaction. Yu et al. [37,38] used SEA in rehabilitation robots to
motion was controlled. A force/position hybrid control method based on assist patients with treatment. The series elastic actuators produce a
fuzzy PID control was put forward by Zhang et al. [19,20] for grinding small output impedance and increase the flexibility of force control,
the aviation blades. However, when the impedance control and for­ which effectively solves the phenomenon of force control instability of
ce/position hybrid control are used, because the robot must control the the high stiffness end-effector [39–42]. Most of existing studies on SEA
contact force and the end-effector position at the same time, it is difficult were designed and utilized in the interaction between humans and the
to ensure good force control accuracy [21,22]. environment, which paid more attention to the force control flexibility
In order to improve force control accuracy, the force-controlled end- than the accuracy. There have been very few examples applied to robotic
effectors were developed by researchers [23–26], where the end-effector grinding processing which has both high requirement of force control
itself was applied to realize the contact force control, and the robot only compliance and accuracy. A rigid-flexible coupling actuator was
controlled the position of the end-effector. There are two main force developed based on SEA structure [43], which showed advantages to
control methods for end-effectors, one is passive control and the other is suppress overshoot of the force control, but the modeling and optimi­
active control. In passive control methods, elastic elements are added to zation of the spring and grinding head stiffness based on the control
the end-effector to achieve passive compliance control. For example, performance and mechanical restriction is not systematically studied.
Wang et al. [27] designed a passive compliant grinding device to avoid By taking the advantage of flexibility of the SEA, this study develops
the influence of contact force fluctuation on the grinding quality. To an active compliant force-controlled end-effector used for robotic
improve force control accuracy, active force control methods measure grinding of curved parts, to solve the problem of high overshoot and low
the actual forces by sensors, and actively eliminate force errors by accuracy of the traditional rigid end-effector. In the end-effector, the
adjusting the distance between the grinding head in the end-effector and rigid ball screw system driven by the servo system has the advantages of
the workpiece. Mohammad et al. [28] and Xiao et al. [29] designed fast response and wide bandwidth, and the elastic component improves
force-controlled end-effectors in which voice coil motors were driven to the compliance and reduces the impact of the grinding force. The main
extend and retract the polishing head. Chen et al. [30] designed a contributions of this paper are summarized in the following points. First,
force-controlled end-effector driven by two orthogonal ball-screw sys­ the influences of the elastic component and the grinding tool stiffness on
tems, and Li et al. [31] used the designed end-effector to grind complex the stability of force-controlled end-effector system is analyzed by
surfaces using the planned contact forces based on the material removal establishing a contact model of the compliant end-effector and the
model. These force-controlled end-effectors [28–31] are rigidly con­ transfer function of the entire force control system. Second, an optimi­
nected between the motor and the load, which have the advantage of zation method for the spring stiffness and grinding head stiffness is
fast response. However, during the grinding process of the complex proposed under the simultaneous consideration of the system cutoff
curved parts, the typical high-stiffness force-controlled end-effectors frequency and the end-effector compliance. Last, the developed force-
have the characteristic of high mechanical output impedance, which is controlled end-effector significantly improves the profile accuracy and
easy to induce force-controlled overshoot and vibration of low-rigidity the surface quality of curved parts by reducing the overshoot and
parts, and thereafter affects the surface quality of parts. Therefore, improving the tracking accuracy of the force control in the robotic
high-quality grinding of complex curved parts places high demands on grinding.
the compliant control of the contact force. The rest of this paper is organized as follows. Section 2 proposes a
Low-impedance series elastic actuator (SEA) can be developed by grinding system which is composed of an industrial robot and a
adding elastic components to the force-controlled end-effector, which compliant force-controlled end-effector. The system modeling, identifi­
improves the compliance of the output force. This fact has attracted cation, and stability analysis of the end-effector are conducted in

2
H. Chen et al. Robotics and Computer-Integrated Manufacturing 86 (2024) 102646

Fig. 2. Control system of the robotic compliant grinding.

Section 3. In Section 4, the spring stiffness and grinding tool stiffness in


the end-effector are reasonably optimized, and the force controller is
designed in Section 5. Experiments and comparisons are conducted in
Section 6. Conclusions are provided in Section 7.

2. Design of the decoupled robotic grinding system

The developed robotic grinding system, which is composed of an


industrial robot and a compliant force-controlled end-effector, is sche­
matically shown in Fig. 1. The robot controller dominates the relative
motion of the grinding tool to track the planned grinding trajectory
along the curved surface. The designed end-effector controlled by the Fig. 3. Physical model of the active compliant end-effector.
Beckoff controller adjusts the contact force by combining active and
passive control method during the grinding process. Fig. 1 depicts the 3. Stability analysis of the active compliant force-controlled
structure of the designed force-controlled end-effector. The linear end-effector
movement of the ball-screw system driven by a servo motor adjusts the
contact force between the grinding tool and the workpiece. The electric In the force control of the compliant end-effector, the spring stiffness
spindle controls the grinding speed to remove the surface material from determines the flexibility of the end-effector, and the grinding tool
the parts. Springs are added between the ball screw system and the stiffness determines the contact dynamics between the grinding head
grinding head to increase the compliance of the contact forces. and the workpiece. Therefore, both the spring stiffness and grinding tool
The control system for the grinding process is shown in Fig. 2. In the stiffness have great influences on the dynamic response of the system
system, the Beckhoff controller is used to control the contact force of the and the surface quality of the grinding. Therefore, it is necessary to
end-effector. The actual contact force is measured by the force sensor analyze the relationship between the spring stiffness, the grinding tool
and sent back to the Beckhoff controller to realize closed-loop control. stiffness, and the system stability.
Generally, the measurement result of the force sensor is affected by the
gravity of the end-effector. Therefore, the robot host computer C5G
Open transmits the collected real-time robot pose information to the 3.1. Modeling of the active compliant force-controlled end-effector
Beckhoff controller through TCP/IP communication, and then the atti­
tude angle of the robot end is obtained to compensate for the influence The physical model of the active compliant end-effector is estab­
of the end-effector gravity on the force sensor. In the grinding process, lished as shown in Fig. 3, where Xm is the output displacement of the
the control commands produced by the Beckhoff controller are sent to motor, Cf is the damping of the spring, X0 is the displacement of the load
the drive system of the servo motor through the EtherCAT filed bus, then end, M is the load mass, and F0 is the contact force between the grinding
the servo motor tries to minimize the discrepancy between the measured head and the workpiece, Ks is the stiffness of the spring, and Ke is the
contact force and the expected contact force. Meanwhile, the analog stiffness of the grinding tool. The contact force is measured by the force
output signals generated by the Beckhoff controller are sent to the sensor. Because the contact damping between the tool and the work­
electric spindle drive system to achieve the desired grinding speed. piece is relatively small, the influence of the contact damping can be
neglected. In the experimental platform, the load mass M is 2.9 kg, and
the spring damping Cf is 0.01Ns/mm.
On the basis of the force balance equation for the end-effector in
Fig. 3, it is obtained that

3
H. Chen et al. Robotics and Computer-Integrated Manufacturing 86 (2024) 102646

function and the closed-loop transfer function of the system. Since the
operation mode of the servo motor is chosen as the speed control, the
speed loop transfer function of the servo motor needs to be identified.
The control block diagram of the developed compliant end-effector is
shown in Fig. 5, where Kv and Ti are the speed loop gain and integral
constant, respectively, ki is the transfer function of the current loop, J is
the equivalent moment of inertia of the motor and ball-screw system, B
is the equivalent damping coefficient, and rg is the ball screw lead.
From Fig. 5, the system transfer function model of the speed loop is
Vm (s) Kv ki Ti s + Kv ki
= (5)
Vc (s) JTi s2 + (BTi + Kv ki Ti )s + Kv ki

which is a second-order model. In the identification, the Pseudo Random


Binary Sequence (PRBS) signals serve as the input commands (Fig. 6(a)),
and the actual speed are used as output signals (Fig. 6(b)).
Using the input and output data in Fig. 6, due to the existence of
filters in the actual speed loop, the speed loop transfer function identi­
Fig. 4. Frequency response comparison of the designed active compliant end- fied using the least squares method is
effector and the traditional rigid end-effector.
− 20.15s + 550.5
G(s) = (6)
s2 + 36.7s + 564.5
M Ẍ 0 − Ks (Xm − X0 ) − Cf (Ẋ m − Ẋ 0 ) + F0 = 0 (1)
In order to verify the correctness of the identified system, the iden­
F0 = K e X 0 (2) tified transfer function is used to generate the response using the iden­
tical PBRS signals as the input signals. It can be seen from Fig. 7 that the
Combining Eqs. (1) and (2), the transfer function from the simulated output speed obtained using the identified system transfer
displacement of the motor end Xm to the contact force F0 is expressed in function matches well with the actual speed. Therefore, the identified
the Laplace domain as transfer function can reflect the response of the end-effector system
F0 (s) Ke Cf s + Ke Ks
= (3)
Xm (s) Ms2 + Cf s + Ke + Ks

In contrast, in the traditional rigid end-effector, the motor output is


directly connected to the load. Therefore, the transfer function from the
displacement of the motor end X′m to the contact force F′0 is

F′0 (s)
= Ke (4)
X′m (s)

Fig. 4 demonstrates the characteristics of the active compliant end-


effector compared to the traditional rigid end-effector. In order to
analyze the influence of the added elastic component on the end-
effector, the grinding tool stiffness Ke is set to 30 N/mm, and the
spring stiffness Ks is set to 1.5 N/mm. As can be seen in Fig. 4, the active
compliant end-effector has better compliance than the rigid end-effector
as it provides greater load displacement over a wide frequency band­
width. Because the active compliant end-effector is a second-order sys­
tem, resonance occurs in the experiments at a frequency of
approximately 17 Hz.

3.2. System identification of the compliant force-controlled system


Fig. 6. Input and output signals of the system identification experiment: (a)
Pseudo Random Binary Sequence (PRBS) signals served as input; (b) actual
When analyzing the dynamic performance of the designed active
speed used as output signals.
compliant end-effector, it is necessary to identify the open-loop transfer

Fig. 5. Control block diagram of the developed compliant end-effector.

4
H. Chen et al. Robotics and Computer-Integrated Manufacturing 86 (2024) 102646

Fig. 7. Verification of identification effect of speed loop transfer function.

Fig. 8. Effects of different spring stiffnesses on the frequency response of the open-loop system: (a) Nyquist plot; (b) Bode plot.

effectively.
Table 1
Based on Fig. 5, the open-loop transfer function in Eq. (7) is obtained
Frequency domain characteristics of the open-loop system with different spring
by combining the speed loop transfer function in Eq. (6), and the contact
stiffnesses.
model of the compliant end-effector in Eq. (3)
Spring stiffness Amplitude Phase margin Cutoff frequency
F0 (s) rg Ke Cf s + Ke Ks Ks(N/mm) margin γ(◦ ) Wc(rad/s)
Gol (s) = = G(s)⋅ ⋅ 2 (7) Kg(dB)
Vc (s) s Ms + Cf s + Ke + Ks
0.5 15.9 79.8 2.1
where rg=5 mm is the ball-screw lead. Combined with proportional- 1 11.3 70.0 3.7
integral (PI) controller used to realize closed-loop control of the con­ 1.5 8.9 62.1 5.0
3 5.6 46.2 7.6
tact force, the closed-loop transfer function of the entire force control
4 4.6 39.3 8.8
system is obtained as 6 3.5 30.4 10.2
( )
F0 (s) Kp s + Ki Gol (s)
Gcl (s) = = ( ) (8)
Fd (s) s + Kp s + Ki Gol (s) N/mm, and 6 N/mm, respectively (see Fig. 8). The grinding tool stiffness
Ke is set to 3 N/mm in the tests.
3.3. Stability analysis As shown in the Nyquist plot in Fig. 8(a), the frequency characteristic
curve approaches the point (-1, j0) as the spring stiffness rises, which
Based on the open-loop and closed-loop transfer function models reduces the system stability. From the Bode plot in Fig. 8(b), the
established above, the influences of the spring stiffness and the grinding amplitude margins, phase margins, and cutoff frequencies of the open-
tool stiffness on the stability of the compliant force-controlled end- loop system corresponding to different spring stiffnesses are calculated
effector system is analyzed in this subsection. by using MATLAB as listed in Table 1. It can be seen that as the spring
stiffness increases, the amplitude margin and phase margin decrease,
3.3.1. Analysis of the open-loop system of the compliant force-controlled and the system stability also becomes worse. In addition, the open-loop
end-effector cutoff frequency increases as the spring stiffness is higher, which means
that the response speed of system becomes higher.
(1) Influence of the spring stiffness on the frequency response of the
open-loop system (2) Influence of the grinding tool stiffness on the frequency response
of the open-loop system
In order to analyze the influence of the different spring stiffnesses on
the stability of the open-loop system, the frequency response diagrams In order to analyze the influence of the different grinding tool stiff­
of the system are drawn when the spring stiffness Ks are 0.5 N/mm, 1.5 nesses on the frequency response of the open-loop system, the frequency

5
H. Chen et al. Robotics and Computer-Integrated Manufacturing 86 (2024) 102646

Fig. 9. Effects of different grinding tool stiffnesses on the stability of open-loop systems: (a) Nyquist plot (b) Bode plot.

Table 2 Table 3
Frequency domain characteristics of the open-loop system with different Frequency domain characteristics of the closed-loop system with different spring
grinding tool stiffnesses. stiffnesses.
Grinding tool stiffness Amplitude margin Phase Cutoff frequency Spring stiffness Resonance peak Resonance Cutoff frequency
Ke(N/mm) Kg(dB) margin γ(◦ ) Wc(rad/s) Ks(N/mm) Mr(dB) frequency Wr(rad/s) Wd(rad/s)

0.5 14.9 79.5 1.8 0.5 0.017 0.05 2.6


1 11.7 73.2 3.0 1 0.023 0.07 6.8
3 8.9 62.1 5.0 1.5 0.039 0.12 13.0
5 8.3 58.0 5.8 3 3.5 11.4 19.5
10 7.8 54.2 6.5 4 5.6 12.4 21.1
20 7.5 52.0 6.9 6 8.5 13.3 22.3

response diagrams of the system are drawn when the grinding tool
Table 4
stiffness Ke are 0.5 N/mm, 3 N/mm, and 20 N/mm, respectively (see
Frequency domain characteristics of the closed-loop system with different
Fig. 9). The spring stiffness Ks is set to 3 N/mm in the tests.
grinding tool stiffnesses.
As shown in the Nyquist plot in Fig. 9(a), the frequency characteristic
curve approaches the point (-1, j0) as the grinding tool stiffness rises, Grinding tool Resonance peak Resonance Cutoff frequency
stiffness Ke(N/mm) Mr(dB) frequency Wr(rad/ Wd(rad/s)
which reduces the system stability. From the Bode plot in Fig. 9(b), the s)
amplitude margins, phase margins, and cutoff frequencies of the open-
0.5 0.019 0.01 2.3
loop system corresponding to different grinding tool stiffness are
1 0.028 0.01 4.4
calculated as listed in Table 2. It can be seen that as the grinding tool 3 0.039 0.12 13.0
stiffness increases, the amplitude margin and phase margin decrease, 5 0.3 6.7 15.1
and the system stability also decreases. In addition, the open-loop cutoff 10 1.1 8.9 16.3
frequency rises as the tool stiffness is higher, which means that the 20 1.6 9.6 16.9

system response becomes faster.


From the comparison in Tables 1 and 2, it can be indicated that the Section 3.2, the resonance peaks, resonance frequencies, and cutoff
amplitude margin, phase margin, and cutoff frequency are more sensi­ frequencies of the closed-loop system corresponding to different spring
tive to the spring stiffness than to the grinding tool stiffness. Thus, it can stiffnesses are calculated as listed in Table 3.
be concluded that the spring stiffness has a greater impact on the It can be seen from Table 3 that when the spring stiffness increases,
compliant end-effector system than the grinding tool stiffness. the resonance peak of the closed-loop system rises and the relative sta­
bility of the system falls. In addition, the cutoff frequency increases
3.3.2. Analysis of the closed-loop system of the compliant force-controlled when the spring stiffness becomes higher, so the response capability of
end-effector the compliant end-effector becomes faster. When the spring stiffness Ks
In order to further analyze the influence of the spring stiffness and is between 0.5 and 1.5 N/mm, the system does not produce overshoot,
the grinding tool stiffness on the compliant force-controlled end- and the resonance peak is near 0. When the spring stiffness Ks is between
effector, it is necessary to conduct analysis of the closed-loop system. 3 and 6 N/mm, the system produces obvious overshoot, and the reso­
nance peak is changed from 3.5 to 8.5 dB.
(1) Influence of the spring stiffness on the frequency response of the
closed-loop system (2) Influence of the grinding tool stiffness on the frequency response
of the closed-loop system
In the closed-loop control system of the compliant force-controlled
end-effector, a proportional-integral (PI) controller parameters are set By fixing the spring stiffness at Ks=1.5 N/mm, the resonance peaks,
to Kp=1 and Ki=0.01. The grinding tool stiffness is set to Ke=3 N/mm. resonance frequencies, and cutoff frequencies of the closed-loop system
According to the closed-loop transfer function model identified in corresponding to different grinding tool stiffnesses are calculated as

6
H. Chen et al. Robotics and Computer-Integrated Manufacturing 86 (2024) 102646

Fig. 10. Three grinding heads with different stiffnesses: (a) belt grinding head with sponge wheel; (b) belt grinding head with rubber wheel; (c) belt grinding head
with metal wheel.

listed in Table 4.
According to Table 4, when the grinding tool stiffness increases, the
resonance peak of the closed system of the compliant end-effector in­
creases and the relative stability of the system decreases. The response
speed of the closed-loop system becomes higher as the cutoff frequency
rises. When the grinding tool stiffness Ke is between 0.5 and 3 N/mm, the
system does not produce overshoot and the resonance peak is almost 0.
When the grinding tool stiffness Ke is between 5 and 20 N/mm, the
system produces overshoot, and the calculated resonance peak of the
system is changed from 0.3 to 1.6 dB.
In conclusion, as the spring stiffness and grinding tool stiffness in­
crease, the force-controlled end-effector system becomes less stable and
less compliant, whereas the response capability of the compliant end-
effector becomes faster as the cutoff frequency increases. In addition,
according to the Tables 3 and 4, the spring stiffness has a more pro­
nounced effect on the force-controlled system performance than the
grinding tool stiffness.
Fig. 11. Stiffness calibration results of different belt grinding heads.
4. Stiffness optimization of the compliant end-effector
as linear. The least square method is used to calibrate the stiffnesses, and
From the frequency analysis of the compliant force-controlled end- the stiffnesses of the belt grinding heads with the sponge wheel, rubber
effector in Section 3, a lower spring stiffness and grinding tool stiffness wheel, and metal wheel are calibrated to be 3 N/mm, 20.3 N/mm, and
results in smaller output impedance and increase the force control 45.6 N/mm, respectively. Since the rigidity of the belt grinding head
compliance, whereas the force control cutoff frequency is reduced. with sponge wheel is smaller than that of the other two types of grinding
Therefore, the spring stiffness and the grinding tool stiffness need to be heads, the selection of the belt grinding head with sponge wheel is ex­
reasonably designed to achieve a balanced performance. pected to have a better performance in reducing force control errors and
overshoots in the grinding process, based on the theoretical analyses in
Section 3.3. Experiments using three different types of grinding heads
4.1. Selection of the grinding tool
will be conducted in Section 6 to verify the effectiveness of the theo­
retical analyses.
Belt grinding is a common and efficient machining method for
improving surface quality and has the advantages of good surface
quality and high efficiency [44–47]. Based on the analyses in Section 4.2. Stiffness optimization of the spring
3.3, as the stiffness of the grinding tool increases, the compliance of the
force-controlled end-effector becomes lower, and the force overshoot According to the analyses in Section 3.3, as the spring stiffness in­
becomes larger. Therefore, among the commonly used belt grinding creases, the response capability of the compliant end-effector becomes
heads, a low-rigidity grinding head is preferred to help improve the faster while the force-controlled end-effector system becomes less stable
surface quality of curved parts by reducing the fluctuations in grinding and less compliant. Therefore, on the premise of meeting the mechanical
forces. structure and cutoff frequency requirements, the spring with a lower
The belt grinding heads that are commonly used in the grinding of stiffness is preferred to improve the compliance of the force-controlled
complex parts include: belt grinding head with sponge wheel, belt end-effector.
grinding head with rubber wheel, and belt grinding head with metal
wheel as shown in Fig. 10. (1) The minimum spring stiffness is restricted by the mechanical
In the grinding head stiffness calibration experiments, the rigid end- structure of the compliant end-effector.
effector (the springs in the proposed compliant end-effector are
removed) is used. The contact force F is measured by the force sensor As shown in Fig. 12, if the spring stiffness Ks is too small, the linear
and the deformation of the grinding head X is measured by the servo bearing interferes with the fixed plate under the hybrid influence of
motor encoder. The experimental results are shown in Fig. 11, which gravity G and contact force F0, which results in no flexibility in the force
demonstrates that the stiffness of the grinding head can be approximated control. In order to ensure that there is no mechanical interference, the

7
H. Chen et al. Robotics and Computer-Integrated Manufacturing 86 (2024) 102646

Fig. 12. Schematic diagram of the spring stiffness selection for the compliant force-controlled end-effector: (a) initial compression position of the spring; (b) position
of the spring compression under the load gravity and contact force.

√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
following relationship must be satisfied (− 20.15ω∗ )2 +550.52
Eq. (6), |G(jω∗ )| = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
∗2 2 ∗ 2
. If the required open-loop cutoff
(564.5− ω ) +(36.7ω )
H1 ≥ H2 , where H2 = (G + F0 )/Ks (9)
frequency is ω∗ = 5rad /s, the spring stiffness Ks that satisfies Eq. (12) is
In the experiments, according to the physical model of the compliant solved to be Ks=1.5 N/mm by using the Newton iteration numerical
end-effector, the maximum distance of H1 that the linear bearing can calculation method in MATLAB.
move under the initial spring compression is 30 mm. H2 is the distance Considering the hybrid constraints of the mechanical structure and
that the linear bearing moves under the hybrid influence of the load the open-loop cutoff frequency, the spring stiffness should not be lower
gravity and contact force. The load gravity G is 2.9 kg, and the planned than 1.5 N/mm. Based on the analyses in Section 3.3, a low-stiffness
maximum contact force F0 is about 6 N. According to Eq. (9), it can be spring is preferred to improve the force control compliance. Therefore,
calculated that the spring stiffness should satisfy to avoid mechanical the optimum spring stiffness is chosen to be 1.5 N/mm to guarantee the
interference compliance of the force-controlled end-effector on the premise of
satisfying the constraints of the mechanical structure and the open-loop
Ks ≥ 1.2N/mm (10) cutoff frequency.

(2) The spring stiffness is optimized to satisfy the requirement of the 5. Normal contact force control
open-loop cutoff frequency.
Parts with complex curved surfaces such as blisks and blades have
According to the open-loop transfer function (Eq. (7)), the amplitude characteristics of severe bending/torsion and large variance of the cur­
of the open-loop frequency characteristic is vature. If they are ground with a constant force, over-grinding is likely to
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( )2 ̅ occur at the position of large curvature due to the large contact stress; on
rg Ke Cf ω + (Ke Ks )2 the contrary, under-grinding may occur at the position of small curva­
|Gol (jω)| = |G(jω)|⋅ ⋅√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
)2̅ (11) ture due to small contact stress. The over- or under- grinding severely
ω ( )2 (
Ke + Ks − M ω2 + Cf ω
affect the profile accuracy of finished parts. Therefore, in order to ensure
Based on the classical control theory [48], the open-loop cutoff fre­ the uniform material removal depth at different positions of the complex
quency is the frequency w∗ that satisfies |Gol (jω∗ )| = 1. Therefore, the curved surface, the contact force must be planned based on the material
corresponding spring stiffness Ks that satisfies the requirement of the removal model that considers the effect of curvature variance. Based on
open-loop cutoff frequency w∗ can be obtained by solving the function the Preston equation [49], the contact force plan method to achieve the
√(̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ ̅ required material depth of the curved surface was developed in a pre­
)
∗ 2 2 vious work [31]
r K e Cf ω + (K e Ks )
(12)
g
|G(jω∗ )|⋅ ∗ ⋅√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( )2 ( )2 = 1 ( )2
ω Ke + Ks − M ω∗2 + Cf ω∗ hVt R1 R2
F= (13)
K h Vr R1 + R2
In the experiments, the grinding tool stiffness Ke is 3 N/mm, the load
mass M is 2.9 kg, and the spring damping Cf is 0.01Ns/mm. According to where F is the planned normal contact force, h is the desired material

8
H. Chen et al. Robotics and Computer-Integrated Manufacturing 86 (2024) 102646

Fig. 13. Workpiece used in experiments.

normal contact force at different positions of the tool path is planned as


shown in Fig. 14.
Generally, the measurement result of the force sensor is affected by
the gravity of the tool. Therefore, in order to accurately obtain the
normal contact force during the grinding process, the components of
gravity need to be compensated for in the reading of the force sensor.
The components of gravity in the force sensor coordinate system are
calculated as

[ Gx Gy Gz ] = [ 0 0 G ]⋅T (14)

where G is the gravity of the tool, Gx, Gy, and Gz are gravity components
in the x, y and z directions of the force sensor coordinate system. T =
Rot(Z, α)⋅Rot(Y, β)⋅Rot(Z, γ)is the transfer matrix between the force
sensor coordinate system (x-y-z) and the global coordinate system (X-Y-
Z) as shown in Fig. 15, with
⎡ ⎤ ⎡ ⎤
cα − sα 0 c β 0 sβ
Fig. 14. The planned contact force at different positions. ⎣ ⎦
Rot(Z, α) = sα cα 0 , Rot(Y, β) = ⎣ 0 1 0 ⎦, Rot(Z, γ)
0 0 1 − sβ 0 cβ
⎡ ⎤
cγ − sγ 0

= sγ cγ 0 ⎦ (15)
0 0 1

Here, ca and sa is abbreviation of cos(a) and sin(a), respectively. α, β,


γ are the Euler angles from the tool center point coordinate system to the
base coordinate system, which are transferred to the controller of the
end-effector by the TCP/IP communication protocol between the C5G
Open system and the Beckoff controller (see Fig. 1), and the corre­
sponding sampling frequency of the TCI/IP communication protocol in
experiments is 100 Hz.
Because the tool center point coordinate system direction is the same
as the force sensor coordinate system direction, then the actual normal
contact force after gravity compensation is

Fy = Fy − Gy = Fy − Gsinβsinγ (16)

where Fy is the reading of the force sensor. By using Eq. (16), the in­
fluence of the tool gravity on the measurement result of force senor is
compensated.
In order to improve the responsiveness and stability of the compliant
Fig. 15. Schematic diagram of gravity compensation. force-controlled end-effector control system, a proportion-integral (PI)
controller with first-order differential force feedforward control is used
removal depth, Vt is the tangential feedrate of the tool along the work­ in the compliant force-controlled end-effector. The control block dia­
piece profile, Vr is the electric spindle speed, R1 is the radius of the gram is shown in Fig. 16.
grinding head, R2 is the curvature radius of the workpiece and Kh is the According to the Eqs. (3) and (5), the expression for the error E(s) is
material removal coefficient. By using the multi-factor orthogonal tests
method introduced in [50], the identified material removal coefficient is
Kh = 9.87 × 10− 7 . In the experiment, the material removal depth along
the profile is set to a constant value (20 μm). For the workpiece profile
used in the experiments shown in Fig. 13, the spindle speed Vr is set to
5000 rpm, and the tangential feedrate Vt is set to 0.2 mm/s, so the

9
H. Chen et al. Robotics and Computer-Integrated Manufacturing 86 (2024) 102646

Fig. 16. Control block diagram of the compliant force-controller end-effector.

Fig. 17. Grinding experiments on curved parts.

)
( Kv ki Ti s+Kv ki rg Ke Cf s+Ke Ks system is
1− Kf s 2
⋅ ⋅ 2 Fd (s)
JTi s +(BTi +Kv ki Ti )s+Kv Ki s Ms +Cf s+Ke +Ks lim e(t) = lims⋅E(s) = 0 (18)
E(s)= ( )
Ki Kv ki Ti s+Kv ki rg Ke Cf s+Ke Ks t→∞ s→0
1+ Kp + ⋅ ⋅
s JTi s2 +(BTi +Kv ki Ti )s+Kv ki s Ms2 +Cf s+Ke +Ks Therefore, the designed closed loop controller for the compliant end-
( 6 ( ) ( ) ( ) )
as +bs5 + c− Kf rg Ti Cf e s4 + d− Kf rg f s3 + e+Kv ki Ks − Kf rg Ks e s2 Fd (s) effector achieves zero steady errors in the control results of the step
= 6 5 4
( )3 2
(
as +bs +cs + d+geTi Cf s +(e+h+gef )s + geKs +rg Ki ef s+rg Ki eKs
) signal or unit ramp signal.
Therefore, the input speed command Vc is written as
(17)
( )
Ki ( )
where a=JTiM, b=JTiCf+MBTi+MKvkiTi, c=JTi(Ke+Ks)+ Vc (s) = Kp + Fd − Fy + Fd K f s (19)
s
Cf(BTi+KvkiTi)+MKvki, e=KvkiKe, d=(Ke+Ks)(BTi+KvkiTi)+CfKvki,
f=TiKs+Cf, g=rgKp, h=KvkiKs+rgKiTiCf. where Kp and Ki are the proportional and integral coefficients of the
When the input signal is a step signal or a unit ramp signal, Fd (s) = 1s feedback controller, respectively, and Kf is the feedforward coefficient.
1
or Fd (s) = s2
, and the steady-state error of the designed force-control

10
H. Chen et al. Robotics and Computer-Integrated Manufacturing 86 (2024) 102646

Fig. 18. Force control comparison of springs and grinding heads with different stiffnesses: (a) grinding using step force with different spring stiffnesses; (b) grinding
using step force with different stiffness of grinding tools; (c) grinding using the planned variable force with different spring stiffnesses; (d) grinding using the planned
variable force with different stiffnesses of grinding tools.

6. Experiments In the robotic grinding operation, the trajectory of the robot along
the workpiece profile is generated by the CAM software, and the feed
The experimental platform is shown in Fig. 17. The designed direction of the end-effector is along as the normal direction of the
compliant force-controlled end-effector is clamped at the end of the contact point on the workpiece surface to adjust the normal grinding
robot, and the workpiece is fixed on the worktable. The robot controller force. Therefore, the motion of the grinding head caused by force control
dominates the relative motion of the grinding tool to track the planned does not bring the tangential motion error of the trajectory, so it is not
grinding trajectory along the workpiece. The compliant end-effector necessary to consider the influence of the grinding head motion in the
controls the contact force between the grinding tool and the work­ path planning or trajectory control process.
piece to follow the planned contact force, and the control period is 10 To verify the influences of the spring stiffness and grinding tool
ms. stiffness on the compliant force-controlled system, robotic grinding

11
H. Chen et al. Robotics and Computer-Integrated Manufacturing 86 (2024) 102646

Fig. 19. Comparison of constant force control between the compliant end-effector and the rigid end-effector: (a) the actual and the desired grinding forces; (b)
tracking errors of the grinding forces.

experiments on curved parts are carried out in four different cases: (1) surface. In the experiment, the desired material removal depth is set to
step force grinding with different spring stiffnesses; (2) step force 20 μm.
grinding with different grinding tool stiffnesses; (3) variable force In cases 1 and 3, three springs with different stiffnesses are used,
grinding with different spring stiffnesses; (4) variable force grinding where K1=1.5 N/mm, K2=2.7 N/mm, and K3=3.9 N/mm. In cases 2 and
with different grinding tool stiffnesses. In the variable force control, the 4, grinding tools with different stiffnesses are used (see Fig. 10), which
reference force signal is planned by using the force planning method are belt grinding head with sponge wheel (Ksponge), belt grinding head
proposed in the authors’ previous work [31] (see Eq. (13) in Section 5) with rubber wheel (Krubber), and belt grinding head with metal wheel
to guarantee the required material removal depth along the curved (Kmetal). The relationships between the stiffnesses of the three grinding

Fig. 20. Comparison of the variable force control between the compliant end-effector and the rigid end-effector: (a) the actual and the desired grinding forces; (b)
tracking errors of the grinding forces.

12
H. Chen et al. Robotics and Computer-Integrated Manufacturing 86 (2024) 102646

heads are Ksponge < Krubbber < Kmetal.


In cases 1 and 3, the grinding head is always selected as the belt
grinding head with sponge wheel to analyze the influence of different
spring stiffnesses on the system, and the same controller parameters are
used as Kp=3, Ki=0.01, and Kf=0.1. In the experiments of grinding with
step force commands (Fig. 18(a)), when the spring stiffness are K1=1.5
N/mm, K2=2.7 N/mm, and K3=3.9 N/mm, respectively, the steady-
state force errors are ±0.2 N, ±0.4 N, and ±0.6 N, respectively, and
the overshoots are about 0%, 6%, and 10%, respectively. In addition, in
the experiments of grinding with planned variable force commands
(Fig. 18(c)), when the spring stiffness are K1=1.5 N/mm, K2=2.7 N/mm,
and K3=3.9 N/mm, respectively, the steady-state force errors are about
±0.3 N, ±0.4 N, and ±0.6 N, respectively, and the overshoots are 2%,
6%, and 13%, respectively.
In cases 2 and 4, the spring stiffness is always selected as 1.5 N/mm
to analyze the influence of different grinding tool stiffnesses on the
system, and the same controller parameters are used as Kp=3, Ki=0.01,
and Kf=0.1. In the experiments of grinding with step force response
commands (Fig. 18(b)), when the grinding heads use sponge wheel
(Ksponge), rubber wheel (Krubber), and metal wheel (Kmetal), respectively,
the steady-state force errors are ±0.2 N, ±0.4 N, and ±0.6 N, respec­
tively, and there is almost no overshoot. In addition, in the experiments
of grinding with the planned variable force commands (Fig. 18(d)), the
steady-state force errors for the grinding heads with a sponge wheel
(Ksponge), rubber wheel (Krubber), and metal wheel (Kmetal), are about
±0.3 N, ±0.4 N, and ±0.6 N, respectively, and the overshoots are 0%,
2%, and 5%, respectively.
It can be concluded from the above experimental results that with an
increase of the spring stiffness and grinding tool stiffness, the steady-
state force error and the force overshoot become larger, and the
response speed of the system becomes higher. Therefore, the results Fig. 21. The workpiece profile is measured by the coordinate
match well with the theoretical analysis in the Section 3. measuring machine.
In order to verify the superiority of the developed compliant force-
controlled end-effector compared with the traditional rigid force-
controlled end-effector, force-controlled grinding experiments are con­
ducted on curved surfaces (Fig. 17). In the compliant force-controlled
end-effector, the belt grinding head with sponge wheel (3 N/mm) is
used and the spring stiffness is 1.5 N/mm. The traditional rigid force-
controlled end-effector is the same as the compliant end-effector while
the only difference is that the elastic component (spring) is removed.
In the step force control experiments, the desired grinding force is set
to 5 N, and the controller parameters of the compliant force-controlled
end-effector are set to Kp=3, Ki=0.01, and Kf=0.3. In order to ensure
that the compliant end-effector and the rigid end-effector have the same
response time, the controller parameters of the rigid force-controlled
end-effector are set to Kp=1.5, Ki=0.02, and Kf=0.1. The experimental
results in Fig. 19 show that the overshoot of the rigid force-controlled
end-effector is 1.8 N, whereas the overshoot of the developed
compliant force-controlled end-effector is neglectable under the same
response speed. The stable force control error of the rigid end-effector is
within ±1 N, whereas the stable force control error of the developed
compliant end-effector is within ±0.3 N. In the stable force tracking
condition, the developed compliant end-effector reduces force control
errors by 70% compared to the rigid end-effector. Fig. 22. Comparison of the material removal accuracy between the compliant
In the variable force control experiments, the controller parameters and the rigid end-effectors after grinding.
of compliant force-controlled end-effector are set to Kp=3, Ki=0.01, and
Kf=0.3. In order to ensure same response time, the controller parameters Therefore, it can be seen that the developed compliant end-effector
of rigid force-controlled end-effector are set to Kp=1.5, Ki=0.02, and has better performance of lower overshoot and higher tracking accu­
Kf=0.1. The experimental results in Fig. 20 show that the overshoot of racy for both step and variable force control compared with the tradi­
the rigid force-controlled end-effector is 1.3 N, whereas the overshoot of tional rigid force-controlled end-effector.
the developed compliant force-controlled end-effector is neglectable For further explorations, the material removal accuracy of the curved
under the same response speed. The stable force control error of the rigid surfaces after grinding with the planned variable contact force are
end-effector is within ±1 N, whereas the stable force control error of the measured by a coordinate measurement machine (CMM) (Fig. 21), and
developed compliant end-effector is within ±0.3 N. Under the variable the roughness of the finished workpiece is measured by a surface
force tracking condition, the developed compliant end-effector reduces roughness tester.
the stable force control error by 70% compared to the rigid end-effector.

13
H. Chen et al. Robotics and Computer-Integrated Manufacturing 86 (2024) 102646

Fig. 23. Comparison of the surface roughness between the compliant and the rigid end-effectors after grinding: (a)compliant end-effector; (b)rigid end-effector.

From Fig. 22, it can be seen from the experimental results that the force control is reduced from 1.5 N to almost 0 N under the same
maximum and average absolute grinding depth errors of the designed response speed. The maximum and average absolute grinding depth
compliant end-effector are 5.9 μm and 2.4 μm, respectively, whereas the errors and the surface roughness of the curved surface finished by the
maximum and average absolute grinding depth errors of the traditional designed compliant end-effector are reduced by 57.2%, 58.6%, and
rigid end-effector are 13.8 μm and 5.8 μm, respectively. The maximum 19.2%, respectively, compared with those of the traditional rigid end-
and average absolute grinding depth errors of the designed compliant effector. The experimental results demonstrate the effectiveness and
end-effector are reduced by 57.2% and 58.6%, respectively, compared advantages of the proposed compliant force-controlled end-effector.
with those of the traditional rigid end-effector. As shown in Fig. 23, the
average roughness of the surface finished by the compliant end-effector CRediT authorship contribution statement
is Ra=0.346 μm, whereas the average roughness of the surface finished
by the traditional rigid end-effector is Ra=0.428 μm. The average Haiqing Chen: Writing – original draft, Investigation, Methodology,
roughness of the surface finished by the designed compliant end-effector Formal analysis, Validation. Jixiang Yang: Writing – review & editing,
is reduced by 19.2% compared to that of the traditional rigid end- Methodology, Supervision, Project administration. Han Ding: Supervi­
effector. The higher material removal accuracy by using the developed sion, Project administration.
compliant end-effector is benefited by the higher control accuracy of the
contact force because it directly affects the material removal rate. The
lower roughness of the surface finished by the developed compliant end- Declaration of Competing Interest
effector is benefited by the lower force fluctuation during the grinding
process. Therefore, the developed compliant end-effector has a better The authors declare that they have no conflicts of interest to this
grinding performance in terms of the profile accuracy and surface work.
quality of curved workpieces compared with the rigid force-controlled We declare that we do not have any commercial or associative in­
end-effector. terest that represents a conflict of interest in connection with the work
submitted.
7. Conclusion
Data availability
In this paper, a compliant force-controlled end-effector is developed
based on the structure of serial-elastic-actuator to improve the force Data will be made available on request.
control accuracy and compliance of the robotic grinding. A model of the
compliant force-controlled end-effector is established, and the in­
fluences of the elastic component stiffness and the grinding tool stiffness
Acknowledgments
on the system response are analyzed. The grinding tool stiffness and
spring stiffness are reasonably chosen with full consideration of the
This research is supported by the National Natural Science Founda­
cutoff frequency and end-effector compliance of the system. A
tion of China under Grant Nos. 52122512, 52188102 and 52375496,
proportional-integral (PI) control structure with the first-order differ­
and the Natural Science Foundation of Hubei Province, China under
ential force feedforward control is designed for the closed-loop control
Grant No. 2021CFA075.
of the designed end-effector. Finally, the effect of the stiffness optimi­
zation on the force-control performance of the end-effector is verified by
References
grinding experiments. The experimental results show that the force
control error of the proposed compliant end-effector is decreased by [1] D. Zhu, X. Feng, X. Xu, Z. Yang, W. Li, S. Yan, H. Ding, Robotic grinding of complex
70% compared to that of the rigid end-effector, and the overshoot of the components: a step towards efficient and intelligent machining – challenges,
solutions, and applications, Robot. Comput. Integr. Manuf. 65 (2020), 101908.

14
H. Chen et al. Robotics and Computer-Integrated Manufacturing 86 (2024) 102646

[2] X. Xu, D. Zhu, H. Zhang, S. Yan, H. Ding, TCP-based calibration in robot-assisted [27] Z. Wang, L. Zou, L. Duan, X. Liu, C. Lv, M. Gong, Y. Huang, Study on passive
belt grinding of aero-engine blades using scanner measurements, Int. J. Adv. compliance control in robotic belt grinding of nickel-based superalloy blade,
Manuf. Technol. 90 (2017) 635–647. J. Manuf. Process. 68 (2021) 168–179.
[3] Y. Yang, H. Li, Z. Liao, D. Axinte, W. Zhu, A. Beaucamp, Controlling of compliant [28] A.E.K. Mohammad, J. Hong, D. Wang, Design of a force-controlled end-effector
grinding for low-rigidity components, Int. J. Mach. Tools Manuf. 152 (2020), with low-inertia effect for robotic polishing using macro-mini robot approach,
103543. Robot. Comput. Integr. Manuf. 49 (2018) 54–65.
[4] Y. Wang, J. Yang, D. Li, H. Ding, Tool path generation with global interference [29] C. Xiao, Q. Wang, X. Zhou, Z. Xu, X. Lao, Y. Chen, Hybrid force/position control
avoidance for the robotic polishing of blisks, Int. J. Adv. Manuf. Technol. 117 strategy for electromagnetic based robotic polishing systems, in: Proceedings of the
(2021) 1223–1232. Chinese Control Conference (CCC), Guangzhou, China, IEEE, 2019, pp. 7010–7015.
[5] H. Huang, Z.M. Gong, X.Q. Chen, L. Zhou, Robotic grinding and polishing for [30] F. Chen, H. Zhao, D. Li, L. Chen, C. Tan, H. Ding, Robotic grinding of a blisk with
turbine-vane overhaul, J. Mater. Process. Technol. 127 (2002) 140–145. two degrees of freedom contact force control, Int. J. Adv. Manuf. Technol. 101
[6] S. Tan, J. Yang, H. Ding, A prediction and compensation method of robot tracking (2019) 461–474.
error considering pose-dependent load decomposition, Robot. Comput. Integr. [31] D. Li, J. Yang, H. Zhao, H. Ding, Contact force plan and control of robotic grinding
Manuf. 80 (2023), 102476. towards ensuring contour accuracy of curved surfaces, Int. J. Mech. Sci. 227
[7] M. Rososhansky, F.J. Xi, Coverage based tool-path planning for automated (2022), 107449.
polishing using contact mechanics theory, J. Manuf. Syst. 30 (2011) 144–153. [32] A. Calanca, S. Toxiri, D. Costanzi, E. Sartori, R. Vicario, T. Poliero, C.D. Natali, D.
[8] H. Huang, L. Zhou, X.Q. Chen, Z.M. Gong, SMART robotic system for 3D profile G. Caldwell, P. Fiorini, J. Ortiz, Actuation selection for assistive exoskeletons:
turbine vane airfoil repair, Int. J. Adv. Manuf. Technol. 21 (2003) 275–283. matching capabilities to task requirements, IEEE Trans. Neural Syst. Rehabil. Eng.
[9] G. Xiao, Y. Huang, Constant-load adaptive belt polishing of the weak-rigidity blisk 28 (2020) 2053–2062.
blade, Int. J. Adv. Manuf. Technol. 78 (2015) 1473–1484. [33] G. Aguirre-Ollinger, H. Yu, Lower-limb exoskeleton with variable-structure series
[10] G. Xiao, Y. Huang, Equivalent self-adaptive belt grinding for the real-R edge of an elastic actuators: phase-synchronized force control for gait asymmetry correction,
aero-engine precision-forged blade, Int. J. Adv. Manuf. Technol. 83 (2016) IEEE Trans. Robot. 37 (2021) 763–779.
1697–1706. [34] J. Zhang, The passive series stiffness that optimizes torque tracking for a lower-
[11] J. Zhang, Y. Shi, X. Lin, Z. Li, Parameter optimization of five-axis polishing using limb exoskeleton in human walking, Front. Neurorobot. 11 (2017) 68.
abrasive belt flap wheel for blisk blade, J. Mech. Sci. Technol. 31 (2017) [35] K. Kong, J. Bae, M. Tomizuka, Control of rotary series elastic actuator for ideal
4805–4812. force-mode actuation in human–robot interaction applications, IEEE ASME Trans.
[12] F. Nagata, Y. Kusumoto, Y. Fujimoto, K. Watanabe, Robotic sanding system for new Mechatron. 14 (2009) 105–118.
designed furniture with free-formed surface, Robot. Comput. Integr. Manuf. 23 [36] K. Kong, J. Bae, M. Tomizuka, A compact rotary series elastic actuator for human
(2007) 371–379. assistive systems, IEEE ASME Trans. Mechatron. 17 (2012) 288–297.
[13] F. Tian, C. Lv, Z. Li, G. Liu, Modeling and control of robotic automatic polishing for [37] H. Yu, S. Huang, G. Chen, N. Thakor, Control design of a novel compliant actuator
curved surfaces, CIRP J. Manuf. Sci. Technol. 14 (2016) 55–64. for rehabilitation robots, Mechatronics 23 (2013) 1072–1083.
[14] W. Ji, L. Wang, Industrial robotic machining: a review, Int. J. Adv. Manuf. Technol. [38] H. Yu, S. Huang, G. Chen, Y. Pan, Z. Guo, Human–robot interaction control of
103 (2019) 1239–1255. rehabilitation robots with series elastic actuators, IEEE Trans. Robot. 31 (2015)
[15] Q. Wang, W. Wang, L. Zheng, C. Yun, Force control-based vibration suppression in 1089–1100.
robotic grinding of large thin-wall shells, Robot. Comput. Integr. Manuf. 67 (2021), [39] N. Paine, S. Oh, L. Sentis, Design and control considerations for high-performance
102031. series elastic actuators, IEEE ASME Trans. Mechatron. 19 (2014) 1080–1091.
[16] Y. Sun, D.J. Giblin, K. Kazerounian, Accurate robotic belt grinding of workpieces [40] S. Oh, K. Kong, High-precision robust force control of a series elastic actuator, IEEE
with complex geometries using relative calibration techniques, Robot. Comput. ASME Trans. Mechatron. 22 (2017) 71–80.
Integr. Manuf. 25 (2009) 204–210. [41] Y. Park, N. Paine, S. Oh, Development of force observer in series elastic actuator for
[17] X. Zhao, B. Tao, L. Qian, Y. Yang, H. Ding, Asymmetrical nonlinear impedance dynamic control, IEEE Trans. Ind. Electron. 65 (2018) 2398–2407.
control for dual robotic machining of thin-walled workpieces, Robot. Comput. [42] G. Chen, P. Qi, Z. Guo, H. Yu, Mechanical design and evaluation of a compact
Integr. Manuf. 63 (2020), 101889. portable knee–ankle–foot robot for gait rehabilitation, Mech. Mach. Theory 103
[18] S. Lakshminarayanan, S. Kana, D.M. Mohan, O.M. Manyar, D. Then, D. Campolo, (2016) 51–64.
An adaptive framework for robotic polishing based on impedance control, Int. J. [43] J. Yang, H. Chen, R. Qi, H. Ding, Y. Yin, A novel approach to robotic grinding
Adv. Manuf. Technol. 112 (2021) 401–417. guaranteeing profile accuracy using rigid-flexible coupling force control for free-
[19] H. Zhang, L. Li, J. Zhao, J. Zhao, The hybrid force/position anti-disturbance control formed surfaces, CIRP Ann. 72 (2023) 313–316.
strategy for robot abrasive belt grinding of aviation blade base on fuzzy PID [44] Y.J. Wang, Y. Huang, Y.X. Chen, Z.S. Yang, Model of an abrasive belt grinding
control, Int. J. Adv. Manuf. Technol. 114 (2021) 3645–3656. surface removal contour and its application, Int. J. Adv. Manuf. Technol. 82 (2016)
[20] H. Zhang, L. Li, J. Zhao, J. Zhao, S. Liu, J. Wu, Design and implementation of 2113–2122.
hybrid force/position control for robot automation grinding aviation blade based [45] W. Wang, J. Li, W. Fan, Investigation into static contact behavior in belt rail
on fuzzy PID, Int. J. Adv. Manuf. Technol. 107 (2020) 1741–1754. grinding using a concave contact wheel, Int. J. Adv. Manuf. Technol. 101 (2019)
[21] C. Fan, G.S. Hong, J. Zhao, L. Zhang, J. Zhao, L. Sun, The integral sliding mode 2825–2835.
control of a pneumatic force servo for the polishing process, Precis. Eng. 55 (2019) [46] L. Zou, X. Liu, X. Ren, Y. Huang, Investigation of robotic abrasive belt grinding
154–170. methods used for precision machining of aluminum blades, Int. J. Adv. Manuf.
[22] F. Chen, H. Zhao, D. Li, L. Chen, C. Tan, H. Ding, Contact force control and Technol. 108 (2020) 3267–3278.
vibration suppression in robotic polishing with a smart end effector, Robot. [47] L. Ren, N. Wang, W. Pang, Y. Li, G. Zhang, Modeling and monitoring the material
Comput. Integr. Manuf. 57 (2019) 391–403. removal rate of abrasive belt grinding based on vision measurement and the gene
[23] A. Roswell, F.J. Xi, G. Liu, Modelling and analysis of contact stress for automated expression programming (GEP) algorithm, Int. J. Adv. Manuf. Technol. 120 (2022)
polishing, Int. J. Mach. Tools Manuf. 46 (2006) 424–435. 385–401.
[24] L. Liao, F.(Jeff) Xi, K. Liu, Modeling and control of automated polishing/deburring [48] M.F. Golnaraghi, B.C. Kuo, Automatic Control Systems, 10th ed., McGraw-Hill
process using a dual-purpose compliant toolhead, Int. J. Mach. Tools Manuf. 48 Education, New York, 2017.
(2008) 1454–1463. [49] M.J. Tsai, J.F. Huang, W.L. Kao, Robotic polishing of precision molds with uniform
[25] G. Zhang, G. Yang, Y. Deng, C. Chen, R. Zhu, K. Yang, Modeling and force control material removal control, Int. J. Mach. Tools Manuf. 49 (2009) 885–895.
of a pneumoelectric end-effector for robotic continuous contact operations, Int. J. [50] K. Zhou, H.H. Ding, S.Y. Zhang, J. Guo, Q.Y. Liu, W.J. Wang, Modelling and
Adv. Manuf. Technol. 121 (2022) 1219–1234. simulation of the grinding force in rail grinding that considers the swing angle of
[26] J. Li, Y. Guan, H. Chen, B. Wang, T. Zhang, X. Liu, J. Hong, D. Wang, H. Zhang, the grinding stone, Tribol. Int. 137 (2019) 274–288.
A high-bandwidth end-effector with active force control for robotic polishing, IEEE
Access 8 (2020) 169122–169135.

15

You might also like