Liftoff and tilt inconsistencies often result from issues in
the reconstruction and registration of the virtual geometry for
path planning, such as misalignments between the virtual and
physical surface geometries. Additional factors include limita-
tions in the robot’s hardware and the synchronization between
the ECA data acquisition instrument and the robot. While
post-processing methods can help mitigate these errors and
improve defect detection, some defects may still be missed if
the errors exceed the algorithm’s corrective capacity.
Part Surface Reconstruction Error
Errors in the reconstructed surface of the test specimen can
arise from several sources. One common issue is the alignment
of the model, where misalignments lead to translation errors
along the scan path. If the model is not properly oriented with
its physical counterpart, rotation and liftoff variations will
occur throughout the entire scanning process. Another chal-
lenge is the discrepancy between the virtual model and the
physical surface geometry.
In this work, surface reconstruction of steel samples is
achieved using structured light with a CR-01 scanner offering
0.1 mm accuracy. However, metallic surfaces can reflect light,
causing inaccuracies during scanning, a problem that can be
mitigated using blue-light reconstruction (Zhan et al. 2015).
Additionally, errors may arise if the sample lies outside the
sensor’s effective range, affecting both translation and rotation.
Ray-plane intersection algorithms calculate translation and
rotation based on the mesh faces’ normal directions (Hamilton
et al. 2024).
Rotational errors between scan points can introduce dis-
crepancies, which can be mitigated using detrending algo-
rithms applied to the raster path. These errors arise at various
stages, as illustrated in Figures 4a and 4b from a scan of flat
Sample A (Figure 3a). Figure 4a depicts the delta rotation
acquired from the robot alone, revealing a baseline pathing
error. In raster pathing, certain scan lines shift more than
others due to orientation errors on the mesh. Figure 4b high-
lights spatial variations in the ECA signal caused by positioning
errors from pathing inaccuracies.
Mesh registration also introduces potential body transfor-
mation errors. The process, which relies on point-pair picking
(PPP), aligns the mesh within the robot’s reference frame.
Misalignment in laser calibration or displacement of calibra-
tion marks during sample placement can introduce errors,
as can human inaccuracies in selecting calibration points. To
reduce these errors, the root-mean-square (RMS) error was
minimized to approximately 1 mm. Although PPP provides
an initial approximation, body errors were further corrected
using 2D detrending. Figure 4c illustrates this correction with
four calibration markers (R0–R3), whose reconstructed posi-
tions were compared to physical locations identified via laser
calibration. The overall RMS error across all four points was
0.516 mm.
Robot Mastering Error
This error occurs during the calibration of a robotic arm’s
joints, where each joint requires precise alignment at zero
degrees. While state-of-the-art robots automate this process,
the robotic arm used in the presented system relies on manual
measurements and markings on the robot itself. Consequently,
sub-degree misalignments can occur at each joint. Although
these errors may seem minor, they result in more complex
orientation discrepancies within the robot’s joint system,
1010 1030
x (mm)
1050 1020 1040
—100
—50
R3 R0
R2 R1
0
50
100
—100
—50
0
50
100
0.9
1
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
x (mm)
0.3
0.2
0.1
0
—0.1
—0.2
Figure 4. Effects from mesh pathing errors: (a) changes of rotation
xyz of the probe during the scan of flat Sample A as provided from
the robot controller (b) corresponding voltage trend registered by
the ECA (c) aligned scanning environment from structured-light
reconstruction after selecting point pairs on the flat sample.
ME
|
ROBOTICECA
66
M AT E R I A L S E V A L U AT I O N • A P R I L 2 0 2 5
y
(mm)
y
(mm) Degree
V
(V)

particularly affecting the Denavit–Hartenberg (DH) parameters
rather than the Cartesian coordinate system, as seen with other
error types. These calibration inaccuracies manifest as “wave-
like” patterns, introducing approximately 1 mm of spatial
variance in the ECA scans (see Figure 6a). To mitigate this
issue, a specialized post-processing method was developed.
Synchronization Error Between Eddy Current
Instrument and Robot’s Controller
Errors arising from poor synchronization between the robot’s
real-time orientation data and the real-time acquisition of
the scanning instrument must also be considered. These syn-
chronization issues cause smearing due to lag between the
two data streams. The system managing both data flows—
comprising the robot’s controller and the ECA data acquisi-
tion instrument—operates as two separate systems, leading to
inevitable delays. Although timestamps are recorded alongside
each data stream, a nonconstant delay persists between the
two. On average, this delay was found to be approximately
50 ms. To minimize the effects of this delay, slower scan speeds
of 25 mm/s were employed, reducing smearing to under 1 mm,
which was adequate to detect most corrosion flaws.
Signal Fluctuation Due to Coil Heating and
Environmental Effects
When performing NDE using ECT array probes in absolute
mode, voltage fluctuations are occasionally observed that
are unrelated to probe orientation. These fluctuations may
occur even when the probe is stationary. Such behavior is
typically attributed to high excitation currents in the coils or
ambient temperature changes (García-Martín et al. 2011). For
instance, Figure 5 illustrates a pronounced case of voltage drift
in an ECA probe while it remains stationary 1 mm above the
sample. This drift occurs during the coil warm-up phase
after powering on the ECA data acquisition system, prior to
reaching its steady-state operating temperature. In this case,
the coils were driven at twice the maximum recommended
voltage, and the signals were amplified by 60 dB. Additionally,
the robot, ECA probe, and steel test sample with corrosion
damage were situated near a high-power air duct emitting hot
air into the adjacent lab space. Once the coils reached their
steady-state temperature, the ECA probe was nulled on an
undamaged region of the test sample. After calibration, voltage
variations were reduced to a range of –0.015 V to 0.015 V.
To mitigate environmental effects on the robotic ECA
measurements, the system was programmed to null the array
probe in the damage-free region at fixed intervals. Additionally,
a second-order detrending process was applied to each coil
signal in the time domain.
Image Processing Algorithms for Corrosion Detection
To address liftoff variations and environmental effects, the
post-processing workflow incorporates a specialized array sub-
traction algorithm and detrending. This approach leverages
the shared liftoff errors across the coils to isolate and correct
these inconsistencies effectively. A simplified representation
of the post-processing procedure is provided in Figure 6. A
detailed description can be found in the authors’ previous work
(Hamilton 2024).
Array Subtraction Algorithm
The array subtraction algorithm removes common liftoff effects
across all probe channels by subtracting the mean value of the
line scans, effectively eliminating identical noise and voltage
variations between coils. It is applied to ECA data in the time
domain to prevent spatial misalignment caused by delays
0 1
—40
—35
—300
—25
—20
—15
—10
—5
0
5
2 3 4 5 6 7
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
Time
Coils
0
—
—
0
5
0
—
((min)n
Figure 5. Voltage fluctuations
in the coils (absolute mode)
caused by array probe heating
during warm-up and hot air
circulation around the probe.
A P R I L 2 0 2 5 • M AT E R I A L S E V A L U AT I O N 67
V
(V)
o
geoltage