IEEE Robotics & Automation Magazine - September 2023 - 61

40PH CPP free-field array microphone with 12AL CPP preamplifiers
(GRAS Sound & Vibration; frequency range 0-
20 kHz) measures the sound spectrum. The microphone is
connected to an EMP373 portable computer (Acme) equipped
with a PCI-6229 input-output board (16 b, 250 kS/s) and a
BNC-2110 Bayonet Neill-Concelman breakout board
(National Instruments). Data points are acquired at 50 kHz.
The machining sound spectrum is then postprocessed to
identify the spindle rotation frequency in real time and estimate
the interaction force through (3).
HAPTIC INTERFACE
The haptic interface provides direct interaction with the
machine operator. A bilateral haptic coupling is used between
the haptic device and the teleoperated robot. When the operator
moves the haptic controller, the device continuously sends
commands to the robot while receiving force feedback from
the sound processing block. The haptic device used in the
reported setup, by 3D Systems, has a refresh rate of 1,000 Hz.
ROBOT TELEOPERATION
The robot in this example has been designed for the in situ repair
of an airfoil [17], [24]. As demonstrated in Figure 4(c), the robot
has 4 DoF: rotation around and translation along its base axis and
two bending DoF provided by nitinol compliant joints [joint 1
and joint 2 in Figure 4(c)]. In this work, only 3 DoF are used:
bending angles 1i and 2i and linear motion L [see Figure 4(c)].
Thus, the 3 DoF of the haptic device [in Figure 4(a)] can be
mapped with a biunivocal relation onto the three motion variables
of the robot. The z-axis of the haptic device is linked to the linear
feed of the robot (L), while motion along the x- and y-axes controls
the tip orientation, coupling the bending DoF (, ).12
ii This
coupling can be expressed with a simplified kinematic transformation,
defined as
i
>i =>>H
1
2
L
where t ,x
tx
ty
tz
HH
X
Y
Z
space of the haptic device and robot tip.
MACHINING SOUND ACQUISITION AND PROCESSING
This section introduces a method for machining force estimation
from audible features, with its application to our example.
The robot in Figure 4 is equipped with an air-driven
machining tool for either milling or grinding. Such air-driven
high-speed spindles are characterized by a near-linear
torque-speed relation [23], and any force applied to the robot
end effector, including contact reaction forces, decreases the
spindle rotation velocity and sound frequency. Thus, the radial
force applied to an air-driven spindle can be estimated
from its rotation frequency during grinding and milling,
which are the most common operations in airplane engine
repair. The relationship between those variables can be
experimentally obtained for a known workpiece and machining
tool [18].
Z
(5)
t ,y and tz are scaling parameters between the workLinear
Actuator
(a)
L
Load Cell
Joint 1
Milling
Tool
Slider
(b)
Workpiece
Joint 2
(c)
θ2
θ1
X
Y
Robot Tip
To obtain an accurate radial machining force from sound
features, we propose a method that can be implemented into
any existing machining system in two steps: an online acquisition
of the sound signal generated during the machining process,
with real-time detection of the spindle frequency, and an
offline mapping of radial milling/grinding force and spindle
frequency, which requires prior calibration.
ONLINE SPINDLE FREQUENCY ACQUISITION
The sound generated by a milling/grinding process is composed
of four different main sources: the collision between
the machining tool and workpiece, the rotation of the spindle
turbine when driven by compressed air, background noise,
and the exhaust air blown out of the air-driven spindle. Only
the sound that comes from the first source is needed, as this
frequency is associated with the spindle's, whereas the other
three sound sources can be seen as external disturbances.
In the proposed example, the sound signal is sampled at
50 kHz by data acquisition in LabVIEW, and the capture time
is 0.002 s at every loop (100 data points per capture window).
For each acquisition, the power spectrum density (PSD) is
calculated to detect the spindle frequency within the noise.
An example of a signal PSD plot from a milling operation
Robot
Control Box
Haptic Device
Microphone
FIGURE 4. The system hardware. (a) The main components of
the system, with the haptic device, robot actuation, and sound
acquisition system. (b) The end effector and the workpiece setup
with a load cell to measure the radial milling force for a calibration
experiment. (c) The end effector of the robot in a bent position,
with the DoF of the system highlighted.
SEPTEMBER 2023 IEEE ROBOTICS & AUTOMATION MAGAZINE
61
IEEE Robotics & Automation Magazine - September 2023

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - September 2023
